Computational Risk Management

Risks exist in every aspect of our lives and risk management has always been a vital
topic. Most computational techniques and tools have been used for optimizing risk
management and the risk management tools benefit from computational
approaches. Computational intelligence models such as neural networks and
support vector machine have been widely used for early warning of company
bankruptcy and credit risk rating, operational research approaches such as VaR
(value at risk) optimization has been standardized in managing markets and credit
risk, agent-based theories are employed in supply chain risk management and
various simulation techniques are employed by researchers working on problems of
environmental risk management and disaster risk management. Investigation of
computational tools in risk management is beneficial to both practitioners and
researchers. The Computational Risk Management series is a high-quality research
book series with an emphasis on computational aspects of risk management and
analysis. In this series research monographs, conference proceedings are published.

More information about this series at http://www.springer.com/series/8827

Sifeng Liu Yingjie Yang
•

Jeffrey Forrest

Grey Data Analysis

Methods, Models and Applications

123
Sifeng Liu Yingjie Yang
Institute for Grey Systems Studies Centre for Computational Intelligence
Nanjing University of Aeronautics De Montfort University
and Astronautics Leicester, Leicestershire
Nanjing, Jiangsu UK
China
Jeffrey Forrest
and Department of Mathematics
Slippery Rock University of Pennsylvania
Centre for Computational Intelligence Slippery Rock, PA
De Montfort University USA
Leicester, Leicestershire
UK

This work supported by

A Project of Marie Curie International Incoming Fellowship under the

7th Framework Programme of the European Union

A joint project of both the NSFC and the RS of UK

Projects of the National Natural Science Foundation of China and

A project of the Leverhulme Trust International Network

Springer-Verlag
2016

ISSN 2191-1436 ISSN 2191-1444 (electronic)

Computational Risk Management
ISBN 978-981-10-1840-4 ISBN 978-981-10-1841-1 (eBook)
DOI 10.1007/978-981-10-1841-1

Library of Congress Control Number: 2016945870

© Springer Science+Business Media Singapore 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer Science+Business Media Singapore Pte Ltd.
Foreword I

It gives me great pleasure to introduce this 8th edition of Grey System Theory and
Its Applications by Prof. Sifeng Liu. The theory of grey systems was first intro-
duced in 1982 by J.L. Deng (1933–2013) at Huazhong University of Science and
Technology; it established a relatively new approach for addressing poorly defined
problems with a high level of greyness or uncertainty. The theory enables one to
model, analyse, monitor, and control such partially defined systems by generating,
excavating, and extracting useful information from what is available. It built on the
work of Dr. Lotfi A. Zadeh, who introduced the concept of fuzzy sets in the 1960s
that in turn led to breakthroughs in neural networks and soft computing.
Grey System Theory actually combines two critical and overarching areas. The
first concerns systems which attempt to synthesize the various components or
subsystems into an overall functioning system or system of systems. Systems
theory attempts to make transparent the deep connections and interactions among
objects and events, all leading to the enrichment and progress of science and
technology. Many of the historically difficult, hard-to-solve problems in the dif-
ferent scientific fields have been successfully resolved through the application of
systems theory and its allied methodologies, including information theory, cyber-
netics, combinatorics, and genetics. The second concerns the greyness or uncer-
tainty level that is implicit in all natural or man-made systems. Indeed, most
modelling techniques assume the existence of uncertainty or stochasticity, as
defined by either empirical evidence or assumed distributions, including fuzzy sets.
Grey System Theory, then, provides a realistic approach to modelling, analysing,
monitoring, and controlling systems. Professor Sifeng Liu has greatly extended, if
not expanded, Prof. Deng’s earlier efforts. In the 1980s, he put forward a series of
new models and concepts, including sequence operator, absolute degree of grey
incidence, grey cluster evaluation model with fixed weight, and positioned coeffi-
cient of grey matrix. In the 1990s, he proposed a buffer operator and its axiom,
generalized degree of grey incidence, grey number and measurement of its infor-
mation content, drifting and positioning solution, the grey econometrics model GM
(1,1), the grey Cobb–Douglas model, etc. More recently, he proposed the concept

v
vi Foreword I

of general grey numbers, the grey algebraic system based on a kernel and degree of
greyness, and different variations of the model GM (1,1).
The widespread recognition and application of grey system theory reflect its
growing acceptance. A number of universities from around the world have adopted
Prof. Sifeng Liu’s monographs, both in Chinese and English, as their textbooks. In
2002, he won the World Organization of Systems and Cybernetics (WOSC) Prize.
In 2008, as a pre-eminent Chinese scholar, he was elected an Honorary WOSC
Fellow. In 2013, after a strict review by the European Commission, he was selected
to be a Marie Curie International Fellow, thus honouring him as the first such
Fellow with grey systems expertise.
As a systems scientist and engineer, I am honoured to write this preface for the
8th edition of Grey System Theory and Its Applications. I look forward to its
widespread dissemination and its promulgation of grey system applications in
science and engineering.

James M. Tien, Ph.D., D.Eng. (h.c.), NAE

Distinguished Professor and Dean Emeritus, College
of Engineering, University of Miami, Coral Gables, FL, USA

Note Prof. James M. Tien prepared this note for 8th edition of Grey System Theory and Its
Applications (in Chinese) by the same authors, published in 2016. With his permission, it is printed
here as a foreword for this current book.
Foreword II

Grey Systems: Theory and Applications

Written by Sifeng Liu and Yi Lin
Springer-Verlag: Berlin, Heidelberg
2010, 379 pages, ISBN 978-3-642-16158-2 (cloth)
DOI 10.1007/978-3-642-16158-2

Professors Sifeng Liu and Yi Lin have written another pioneering book on the
important topic of grey systems. In 2006, the same authors wrote the well-received
book entitled “Grey Information: Theory and Practical Applications” which was
also published by Springer-Verlag. I am pleased to say that their second book on
Grey Systems constitutes a significant expansion and improvement in their previous
fine book. Accordingly, if you already possess a copy of the 2006 book, you can
make a worthwhile academic investment by obtaining a copy of their recent book in
order to be cognizant of the latest ideas and advancements in the crucial field of
grey systems.
The question that naturally arises is why grey systems are of such great import at
this point in history. The answer is quite straightforward: many challenging prob-
lems facing society consist of interconnected complex systems of systems
exhibiting high uncertainty and having few measurements. For example, in order to
effectively combat climate change, one must understand as much as possible the
complex interactions among natural systems such as atmospheric, oceanic, geo-
logical, and hydrological systems, with societal systems including energy produc-
tion, industrial, agricultural, and city systems. The deep uncertainty involved with
these interconnected systems of systems and their potential emergent behaviour,
coupled with a dearth of observations, mean that formal tools for handling this
uncertainty are in high demand. Fortunately, an arsenal of mathematically based
methodologies and techniques have been developed over the years: a rich variety of
probabilistic-based tools, fuzzy sets founded by Lotfi Zadeh, rough sets started by
Z. Pawlak, information-gap modelling perfected by Yakov Ben-Haim, uncertainty
theory developed by Baoding Liu, and grey systems established by Julong Deng in
1982. The foregoing and other approaches to describing uncertainty are based upon

vii
viii Foreword II

different axioms and are thereby highly complementary for tackling a wide variety
of uncertain situations.
Grey systems are purposefully designed for modelling uncertain systems, or
systems of systems, problems having small samples, and low-quality information.
Grey systems are capable of dealing with partially known information through
generating, excavating, and extracting useful information from what is available.
How this is accomplished is explained in depth in the timely grey systems book of
Professors Liu and Lin.
In their contemporary textbook, Liu and Lin systematically present the theory
and practice of grey systems. In fact, the excellent ideas and applications contained
in their book are based upon the authors’ many years of developing theoretical
concepts, applying their methods to real-world applications, testing and refining
their new techniques with actual data, carrying out stimulating research with their
students and colleagues, teaching their students about their exciting work, and
delivering research papers at international conferences around the globe. Their
comprehensive book contains the latest theoretical and applied advances created by
the authors and other scholars around the world in order to place the readers at the
forefront of international research in grey systems.
The main body of their book contains ten well-explained and interconnected
chapters: Introduction to Grey Systems Theory, Basic Building Blocks, Grey
Incidence and Evaluation, Grey Systems Modeling, Discrete Grey Prediction
Models, Combined Grey Models, Grey Models for Decision Making, Grey Game
Models, Grey Control Systems, and Introduction to Grey Systems Modeling
Software. Moreover, this book includes a computer software package developed for
grey systems modelling to permit both researchers and practitioners to use the new
methodologies. Their book concludes with three appendices. The first appendix
compares grey systems theory and interval analysis while revealing the fact that
interval analysis is a part of grey mathematics. The second presents an array of
different approaches to studying uncertainties. Finally, the last appendix shows how
uncertainties occur using a general systems approach.
This book contains a wealth of mathematical results, techniques, and algorithms
which are presented by the authors for the first time. These contributions include an
axiomatic system of buffer operators and a series of weakening and strengthening
operators; axioms for measuring the greyness of grey numbers; general grey inci-
dences (grey absolute incidence, grey relative incidence, grey comprehensive
incidence, grey analogy incidence, and grey nearness incidence); discrete grey
models; fixed weight grey cluster evaluation; and grey evaluation methods based on
triangular whitenization weight functions, multi-attribute intelligent grey target
decision models, applicable range of the G(1,1), grey econometrics (G-E), grey
Cobb-Douglass (G-C-D), grey input-output (G-I-O), and grey game models (G-G).
In their well-written book, Drs. Liu and Lin do a thorough job in their presen-
tation of many difficult technical concepts. The authors are able to convince the
readers of their book regarding the power and usefulness of their new theory by
presenting many interesting examples of practical applications to real-life problems.
The challenging practical problems addressed in their book include urban economic
Foreword II ix

planning, downtown traffic design, natural disaster prediction, relative strength

evaluation of a state, investment projection of a company, and employee perfor-
mance evaluation.
The depth and scope of the advancements in grey systems covered in this book,
in conjunction with clarity of explanation, make this seminal book attractive to
researchers, students, teachers, and practitioners working in many different fields.
These areas of endeavour include image processing, video processing, multimedia
security, computer vision, machinery, control, agriculture, water resources, medi-
cine, astronomy, earth science, economics, and management. I personally found
grey systems useful for accurately forecasting wastewater time series for which
there is a scarcity of data. I intend to keep a copy of this valuable book easily
accessible in my university office and purchase more copies of the book for use by
my students.

Keith William Hipel

Ph.D., P.Eng., FIEEE, FINCOSE, FCAE, FEIC, FRSC, FAWRA,
University Professor of Systems Design Engineering, University
of Waterloo, Waterloo, ON, Canada, Senior Fellow, Centre
for International Governance Innovation, Waterloo, ON, Canada

Note Prof. Keith William Hipel prepared this note for one of the earlier book by the same authors,
published in 2010. It is published in Grey Systems: Theory and Application, 2011, Vol. 1, No. 3.
With his permission, it is printed here as a foreword for this current book.
Foreword III

With human knowledge maturing and scientific exploration deepening and largely
expanding in the course of time, mankind finally realizes the fundamental fact that
due to both internal and external disturbances and limitations of human and tech-
nical sensing organs, all information received or collected contains some kind of
uncertainty. Accompanying the progress of science and technology and the
aforementioned realization, our understanding about various kinds of uncertainties
has gradually been deepened. Attesting to this end, in the second half of the
twentieth century, the continual appearance of several influential and different types
of theories and methods on unascertained systems and information has become a
major aspect of the modern world of learning. Each of these new theories was
initiated and followed up by some of the best minds of our modern time.
In their recent book, entitled “Grey Information: Theory and Practical
Applications”, published in its traditionally excellent way by Springer,
Profs. Sifeng Liu and Yi Lin presented in a systematic fashion the theory of grey
system, which was first proposed by J.L. Deng in the early 1980s and enthusias-
tically supported by hundreds of scientists and practitioners in the following years.
Based on the hard work of these scholars in the past (nearly) thirty years, scholars
from many countries currently are studying and working on the theory and various
applications of this fruitful scientific endeavour. With this book published by such a
prestigious leading publisher of the world, it can be expected that more scientific
workers from different parts of the world will soon join hands and together make
grey system and information a powerful theory capable of bringing forward prac-
tically beneficial impacts to the advancement of the human society.
This book focuses on the study of such unascertained systems that are known
with small samples or “poor information”. Different of all other relevant theories on
uncertainties, this work introduces a system of many methods on how to deal with
grey information. Starting off with a brief historical introduction, this book carries
the reader through all the basics of the theory. And, each important method studied
is accompanied with a real-life project the authors were involved in during their
professional careers.

xi
xii Foreword III

Many of the methods and techniques the reader will learn in this book were
originally introduced by the authors. They show how from our knowledge based on
partially and poorly known information can be obtained to accurate descriptions
and effective controls of the systems of interest. Because this book shows how the
theory of grey system and information was established and how each method could
be practically applied, this book can easily be used as a reference by scholars who
are interested in either theoretical exploration or practical applications or both.
I recommend this book highly to anyone who has either a desire or a need to learn.

July 2007 Prof. Dr. Dr. h.c. mult. Hermann Haken

Founder of Synergetics
Center for Synergetics, University of Stuttgart
Stuttgart, Baden-Württemberg, Germany

Note Prof. Hermann Haken prepared this note for one of the earlier book by the same authors,
published in 2006. It is published in Grey Systems: Theory and Application, 2011, Vol. 1, No. 1.
With his permission, it is printed here as a foreword for this current book.
Foreword IV

I am much interested and impressed by Dr. Sifeng Liu and Dr. Yi Lin’s recently
published monograph on grey information, dealing with the theory and practical
applications.
This book encompasses many aspects of mathematics under the aegis of
uncertain information. I am greatly in favour of this attitude, concerning the
uncertainty of information, which has been mine since a long time ago. Also, this
book focuses on practice and aims at explorations of new knowledge. It is a
comprehensive, all-in-one exposition, detailing not only with the theoretical
foundation but also real-life applications. Because of this characteristic of quality
and usefulness, Liu and Lin’s book possesses the value of the widest possible range
of reference by the workers and practitioners from all corners of natural and social
sciences and technology.
In this book, Liu and Lin present the theory of grey information and systems
starting on such background information as the relevant history, an attempt to
establish a unified information theory, the basics of grey elements, and reaching all
the most advanced topics of the theory. Complemented by many first-hand and
practical project successes, the authors developed an organic theory and method-
ology of grey information and grey system, dealing with errors. In fact, there is
much more to tell about error than about truth. Error (inexactitude) can be met
everywhere and truth (exactitude) nowhere. But inexactitude contains a part of the
truth. Greyness is the field we live in. Extremes, as whiteness and blackness, are
inaccessible, but very useful, ideal concepts.
With the publication of such a book that contains not only a theory, aspects of
magnificent real-life implications and explorations of new research, but also the

xiii
xiv Foreword IV

history, the theorization of various difficult concepts, and directions for future
works, there is no doubt that Drs. Liu and Lin have made a remarkable contribution
to the development and applications of systems science.

June 2007 Prof. Robert Vallée

Member of the French-speaking Academy
of Engineers, President of the World Organisation
of Systems and Cybernetics, Université
Paris-Nord, Paris, France

Note This note is a book review written by Prof. Robert Vallée for one of the earlier book by the
same authors, published in 2006. It is published in Kybernetes: The International Journal of
Cybernetics, Systems and Management Science, 2008, Vol. 37, No. 1. With his permission, it is
printed here as a foreword for this current book.
Preface

In this book, we answer the calls of the readers of our previous publications and
systematically present the main advances in grey system theory and applications.
By following our readers’ feedback and suggestions, this volume introduces the
most recent research results and updates on what is presented in our earlier books.
In particular, the following content, which represents the authors’ recent research, is
highlighted in the book: general grey numbers and their operations, grey incidence
models based on similarity and closeness, three-dimensional degree of grey inci-
dence models, grey evaluation models based on centre-point mixed possibility
functions, grey evaluation models based on endpoint mixed possibility functions,
original difference grey model (ODGM), even difference grey model (EDGM),
multi-attribute intelligent grey target decision models, weight vector group of kernel
clustering, and the weighted coefficient vector of kernel clustering for
decision-making. We also attach software designed for grey system modelling,
which was developed by Bo Zeng using Visual C#, the widely employed C/S
software tool. This user-friendly software allows users to conveniently input and/or
upload data and clearly distinguish module functions. Also, the software has the
ability to present users with operational details, as well as periodic and partial
results. Additionally, users can adjust the levels of computational accuracy based on
their practical needs.
During the writing of this book, we prioritized theoretical simplicity and clarity
to make it easy for the reader to follow the main arguments made. With a good
number of practical applications, we intended to illustrate the methodology of grey
system theory and modelling techniques so that we could emphasize the practical
applicability of grey system thinking. We drew on the most recent research
developments from various research groups around the world, and tried to present
the most complete picture of this new area of scientific endeavour in a concise
manner.
The overall planning and organization of topics contained in this book were
carried out by Sifeng Liu, who also authored Chaps. 1, 4, 6, and 10. Yingjie Yang
produced Chaps. 2, 3, and 11, Jeffrey Forrest composed Chaps. 7 and 8, Naiming

xv
xvi Preface

Xie wrote Chap. 9, and Chap. 12 and the attached computer software were
developed by Zeng Bo. Zhigeng Fang, Yaoguo Dang, Lirong Jian, and Chunhua Su
and colleagues also worked with the authors to refine some of the book's content.
Sifeng Liu was responsible for unifying the terms used throughout the book and for
finalizing the manuscript.
Finally, we would like to encourage you to communicate with us and send us
any comments you might have about this book. It is only by working together, as a
team, that we can grow and mature as researchers. Sifeng Liu can be
reached at [email protected] and [email protected]. Yingjie Yang can be
reached at [email protected] and Jeffrey Forrest at [email protected] or
[email protected].

Nanjing, China Sifeng Liu

April 2016
Acknowledgements

The work presented in this book is supported by a Marie Curie International

Incoming Fellowship under the 7th Framework Programme of the European Union,
entitled “Grey Systems and Its Application to Data Mining and Decision Support”
(Grant No. FP7-PIIF-GA-2013-629051). It is also supported by a joint project of
both the NSFC and the RS in the UK entitled “Grey System Theory and
Computational Intelligence” (71111130211), a project of the Leverhulme Trust
International Network entitled “Grey Systems and Its Applications” (IN-2014-020),
and projects linked to the National Natural Science Foundation of China (91324003,
90924022, 70473037, 70971064). Also, the authors would like to acknowledge the
support provided by the Fundamental Research Funds for Central Universities
(NP2015208), the Research Funds of the Think Tanks for Science and Technology
Development (BR2014100), the Foundation for Key Research Base of Philosophy
and Social Science in Universities of Jiangsu Province (2014JDXM015), the
Foundation for National Outstanding Teaching Group of China (No. 10td128),
the Foundation for Doctoral Programs (grant No. 20020287001, 200802870020,
20093218120032), and the Key Project of Natural Science Foundation of Jiangsu
Province (grant No. BK2003211). The Science Foundation for the Excellent and
Creative group in Science and Technology (on Grey System) of Nanjing University
of Aeronautics and Astronautics and Jiangsu Province (No. Y0553-091), and the
Foundation for the Excellent Young Scientists in Henan Province (No. 964040600)
have also provided their support.
Over the years, our research has been highly commended by many first class
scholars, such as Julong Deng, the founder of grey system theory, L.A. Zadeh, the
founder of fuzzy mathematics, Hermann Haken, the founder of synergetics, Robert
Vallee, the president of the World Organization of Cybernetics and Systems, James
Tien, former vice-president of IEEE and member of the American Academy of
Engineering, K.W. Hipel, the president of The Academy of Science of the Royal
Society of Canada, Jifa Gu, former president of the International Federation for
Systems Research, and academicians of the Chinese Academy of Sciences and the
Chinese Academy of Engineering, such as Qun Lin, Da Chen, Chunsheng Zhao,

xvii
xviii Acknowledgements

Haiyan Hu, Suzi Yang, Youlun Xiong, Zhongtuo Wang, Shanlin Yang, Guozhi Xu,
and others. Xuesen Qian, a world-renowned scientist, has also congratulated us for
our achievements in this area of research. Indeed, such positive feedback on our
work has given us the impetus to develop this book.
Finally, many colleagues and administrators have supported us in the process of
writing this book, as the authors have consulted many scholars and a wide range of
relevant published literature throughout the development of this project. In par-
ticular, they are deeply in debt to Dr. Caroline Moraes, who proofread all chapters
of the book; Prof. Dash Wu, the editor of a book series who has encouraged us to
publish the book; and Wei Zhao and Hannah Qiu, who provided great editorial
guidance through to the final publication of this book. The authors would like to
express their appreciation to them all.
Contents

1 Introduction to Grey Systems Research . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Appearance and Growth of Grey Systems Research . . . . . . . . . . 1
1.2 Development History and Current State . . . . . . . . . . . . . . . . . . . 2
1.3 Characteristics of Uncertain System . . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 Incomplete Information . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.2 Inaccuracies in Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.3 The Scientific Principle of Simplicity . . . . . . . . . . . . . . 7
1.3.4 Precise Models Suffer from Inaccuracies . . . . . . . . . . . . 8
1.4 Comparison of Several Studies of Uncertain Systems . . . . . . . . . 10
1.5 Most Actively Studied Uncertain Systems Theories . . . . . . . . . . 11
1.6 Elementary Concepts of Grey System . . . . . . . . . . . . . . . . . . . . . 13
1.7 Fundamental Principles of Grey Systems . . . . . . . . . . . . . . . . . . 14
2 The Grey Systems Theory Framework . . . . . . . . . . ............. 17
2.1 Grey Models and Framework . . . . . . . . . . . . . . ............. 17
2.2 The Thinking, Models and Framework of Grey Systems
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Grey Numbers and Its Operations . . . . . . . . . . . . . . . . . 18
2.2.2 The Grey Sequence Operator . . . . . . . . . . . . . . . . . . . . . 19
2.2.3 The Grey Prediction Models . . . . . . . . . . . . . . . . . . . . . 20
2.2.4 Grey Incidence Analysis Models . . . . . . . . . . . . . . . . . . 21
2.2.5 Grey Clustering Evaluation Models . . . . . . . . . . . . . . . . 22
2.2.6 Grey Decision-Making Models . . . . . . . . . . . . . . . . . . . 24
2.2.7 Combined Grey Models . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.8 Grey Control Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 The New Framework and Main Components
of Grey Systems Theory . . . . . . . . . . . . . . . . . . ............. 26
3 Grey Numbers and Their Operations . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1 Grey Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 The Whitenization of a Grey Number and Degree
of Greyness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

xix
xx Contents

3.3 Degree of Greyness Defined by Axioms . . . . . . . . . . . . . . . . . . . 33

3.4 The Operations of Interval Grey Numbers . . . . . . . . . . . . . . . . . 35
3.5 General Grey Numbers and Their Operations . . . . . . . . . . . . . . . 37
3.5.1 Reduced Form of Interval Grey Numbers . . . . . . . . . . . 37
3.5.2 General Grey Numbers and Their Reduced Form . . . . . 38
3.5.3 Synthesis of Degree of Greyness and Operations
of General Grey Numbers . . . . . . . . . . . . . . . . . . . .... 40
4 Sequence Operators and Grey Data Mining . . . . . . . . . . . . . . . .... 45
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 45
4.2 Systems Under Shocking Disturbances and Buffer
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 47
4.2.1 The Trap for Shocking Disturbed System
Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.2 Axioms that Define Buffer Operators . . . . . . . . . . . . . . 48
4.2.3 Properties of Buffer Operators . . . . . . . . . . . . . . . . . . . . 49
4.3 Construction of Practically Useful Buffer Operators . . . . . . . . . . 50
4.3.1 Weakening Buffer Operators . . . . . . . . . . . . . . . . . . . . . 50
4.3.2 Strengthening Buffer Operators . . . . . . . . . . . . . . . . . . . 54
4.3.3 The General Form of Buffer Operator . . . . . . . . . . . . . . 55
4.4 Average Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.5 The Quasi-Smooth Sequence and Stepwise Ratio Operator . . . . 58
4.6 Accumulating and Inverse Accumulating Operators . . . . . . . . . . 60
4.7 Exponentiality of Accumulating Generation . . . . . . . . . . . . . . . . 62
5 Grey Incidence Analysis Models. . . . . . . . . . . . . . . . . . . . . . . . . .... 67
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 67
5.2 Grey Incidence Factors and Set of Grey Incidence
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 69
5.3 Degrees of Grey Incidences Model . . . . . . . . . . . . . . . . . . . .... 72
5.4 Absolute Degree of Grey Incidence Model . . . . . . . . . . . . . .... 76
5.4.1 Relative and Synthetic Degree of Grey
Incidence Models . . . . . . . . . . . . . . . . . . . . . . . . . .... 81
5.4.2 Similarity, Closeness and Three-Dimensional
Degree of Grey Incidence Models . . . . . . . . . . . . . .... 85
5.5 Superiority Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 92
5.5.1 Practical Application . . . . . . . . . . . . . . . . . . . . . . . .... 99
6 Grey Clustering Evaluation Models . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.2 Grey Incidence Clustering Model . . . . . . . . . . . . . . . . . . . . . . . . 107
6.3 Variable Weight Grey Clustering Model . . . . . . . . . . . . . . . . . . . 110
6.4 Fixed Weight Grey Clustering Model . . . . . . . . . . . . . . . . . . . . . 117
Contents xxi

6.5 Grey Clustering Evaluation Models Based on Mixed

Possibility Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 121
6.5.1 Grey Clustering Evaluation Model
Based on End-Point Mixed Possibility Functions . .... 121
6.5.2 Grey Clustering Evaluation Model Based
on Center-Point Mixed Possibility Functions . . . . . .... 127
6.6 Practical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 131
7 Series of GM Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
7.2 The Four Basic Forms of GM (1, 1). . . . . . . . . . . . . . . . . . . . . . 142
7.2.1 The Basic Forms of Model GM (1, 1). . . . . . . . . . . . . . 142
7.2.2 Properties and Characteristics of the Basic Model . . . . . 145
7.3 Suitable Ranges of Different GM (1, 1) . . . . . . . . . . . . . . . . . . . 149
7.3.1 Suitable Sequences of Different GM (1, 1) . . . . . . . . . . 149
7.3.2 Applicable Ranges of EGM . . . . . . . . . . . . . . . . . . . . . . 157
7.4 Remnant GM (1,1) Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
7.5 Group of GM (1, 1) Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.6 The Models of GM (r, h) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.6.1 The Model of GM (0, N) . . . . . . . . . . . . . . . . . . . . . . . 169
7.6.2 The Model of GM (1, N) . . . . . . . . . . . . . . . . . . . . . . . 171
7.6.3 The Grey Verhulst Model . . . . . . . . . . . . . . . . . . . . . . . 173
7.6.4 The Models of GM (r, h) . . . . . . . . . . . . . . . . . . . . . . . 176
7.7 Practical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
8 Combined Grey Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 183
8.1 Grey Econometrics Models . . . . . . . . . . . . . . . . . . . . . . . . . .... 183
8.1.1 Determination of Variables Using the Principles
of Grey Incidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.1.2 Grey Econometrics Models . . . . . . . . . . . . . . . . . . . . . . 184
8.2 Combined Grey Linear Regression Models. . . . . . . . . . . . . . . . . 187
8.3 Grey Cobb-Douglas Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
8.4 Grey Artificial Neural Network Models . . . . . . . . . . . . . . . . . . . 191
8.4.1 BP Artificial Neural Model and Computational
Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
8.4.2 Steps in Grey BP Neural Network Modeling. . . . . . . . . 192
8.5 Grey Markov Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
8.5.1 Grey Moving Probability Markov Model . . . . . . . . . . . 194
8.5.2 Grey State Markov Model . . . . . . . . . . . . . . . . . . . . . . . 195
8.6 Combined Grey-Rough Model . . . . . . . . . . . . . . . . . . . . . . . . . . 197
8.6.1 Rough Membership, Grey Membership and Grey
Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
8.6.2 Grey Rough Approximation . . . . . . . . . . . . . . . . . . . . . 199
8.6.3 Combined Grey Clustering and Rough Set Model . . . . . 202
8.7 Practical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
xxii Contents

9 Techniques for Grey Systems Forecasting . . . . . . . . . . . . . . . . . . . . . 209

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
9.2 Interval Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
9.3 Grey Disaster Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.4 Wave Form Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
9.5 System Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.5.1 The Five-Step Modeling Process . . . . . . . . . . . . . . . . . . 221
9.5.2 System Models for Prediction . . . . . . . . . . . . . . . . . . . . 222
9.6 Practical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
10 Grey Models for Decision-Making . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
10.2 Grey Target Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
10.3 Other Approaches to Grey Decision . . . . . . . . . . . . . . . . . . . . . . 234
10.3.1 Grey Incidence Decision . . . . . . . . . . . . . . . . . . . . . . . . 234
10.3.2 Grey Development Decision . . . . . . . . . . . . . . . . . . . . . 239
10.3.3 Grey Clustering Decision . . . . . . . . . . . . . . . . . . . . . . . 241
10.4 Multi-attribute Intelligent Grey Target Decision Model . . . . . . . 241
10.4.1 The Uniform Effect Measure . . . . . . . . . . . . . . . . . . . . . 242
10.4.2 The Weighted Synthetic Effect Measure . . . . . . . . . . . . 244
10.5 The Paradox of Rule of Maximum Value and Its Solution . . . . . 248
10.5.1 The Weight Vector Group of Kernel Clustering . . . . . . 249
10.5.2 The Weighted Coefficient Vector of Kernel
Clustering for Decision-Making . . . . . . . . . . . . . . . .... 250
10.5.3 Several Functional Weight Vector Groups
of Kernel Clustering . . . . . . . . . . . . . . . . . . . . . . . .... 252
10.6 Practical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 253
11 Grey Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
11.2 Controllability and Observability of Grey System. . . . . . . . . . . . 258
11.3 Transfer Functions of Grey System . . . . . . . . . . . . . . . . . . . . . . 260
11.3.1 Grey Transfer Function . . . . . . . . . . . . . . . . . . . . . . . . . 260
11.3.2 Transfer Functions of Typical Links . . . . . . . . . . . . . . . 261
11.3.3 Matrices of Grey Transfer Functions . . . . . . . . . . . . . . . 265
11.4 Robust Stability of Grey System. . . . . . . . . . . . . . . . . . . . . . . . . 266
11.4.1 Robust Stability of Grey Linear Systems . . . . . . . . . . . . 266
11.4.2 Robust Stability of Grey Linear Time-Delay
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 269
11.4.3 Robust Stability of Grey Stochastic Linear
Time-Delay System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
11.5 Typical Grey Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
11.5.1 Control with Abandonment . . . . . . . . . . . . . . . . . . . . . . 276
11.5.2 Control of Grey Incidence . . . . . . . . . . . . . . . . . . . . . . . 277
11.5.3 Control of Grey Prediction . . . . . . . . . . . . . . . . . . . . . . 278
Contents xxiii

12 Introduction to Grey Systems Modeling Software . . . . . . . . . . . . . . 285

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
12.2 Software Features and Functions . . . . . . . . . . . . . . . . . . . . . . . . 286
12.3 Main Components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
12.4 Operation Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
12.4.1 The Confirmation System . . . . . . . . . . . . . . . . . . . . . . . 290
12.4.2 Using the Software Package . . . . . . . . . . . . . . . . . . . . . 291
Farewell to Our Tutor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
About the Authors

Prof. Sifeng Liu Ph.D. SM IEEE, Honorary Fellow of WOSC, and Senior Fellow
of Marie Curie International Incoming Fellowship under the 7th Framework
Programme of the European Union.
Sifeng Liu received Ph.D. in systems engineering from Huazhong University of
Science and Technology, China, in 1998. Dr. Liu is currently a distinguished
professor of Nanjing University of Aeronautics and Astronautics and a research
professor of De Montfort University. He is serving as the founding director of
Institute for Grey Systems Studies, the founding president of International
Association of Grey Systems and Uncertainty Analysis, the founding chair of TC of
IEEE SMC on Grey Systems, and the founding president of Grey System Society of
China. Professor Liu is also serving as the founding editor of Grey Systems Theory
and Application (Emerald) and the editor in chief of The Journal of Grey Systems
(Research Information). He had worked at Slippery Rock University in Slippery
Rock and New York Institute of Technology in New York, USA; Sydney
University in Sydney, Australia; and De Montfort University in Leicester, UK, as a
visiting professor. And he led the College of Economics and Management, NUAA,
from 2001 until 2012.
Professor Liu’s main research activities are in grey system theory and applica-
tions. He has directed more than 50 research projects from China, UK, UN, and EU.
He has published over 600 research papers and 26 books by Science Press,
Springer-Verlag, Taylor & Francis Group, and John Wiley & Sons, Inc. He is
currently serving as the editor of the book series of “Grey Systems” published by
Science Press. His works cited 32 thousand times by others and had translated to
Korean, German, and Romanian. Over the years, he has been awarded 18 provincial
and national prizes for his outstanding achievements in scientific research and
applications. In 2002, Dr. Liu was recognized by the World Organization of
Systems and Cybernetics. Sifeng Liu’s H-index is 55.
Dr. Liu has won several accolades such as the “National Excellent worker of
science and technology”, “National Excellent Teacher”, “National Advanced
Individual for Returnee”, “Expert Enjoying Government’s Special Allowance”, and
“National Expert with Prominent Contribution”.

xxv
xxvi About the Authors

Prof. Yingjie Yang is currently a full professor in Computational Intelligence at

the Centre for Computational Intelligence of De Montfort University, Leicester,
UK. He received the B.Sc. (Hons.), M.Sc., and Ph.D. degrees in engineering from
Northeastern University, Shenyang, China, in 1987, 1990, and 1994, respectively.
He was awarded his Ph.D. degree in computer science at Loughborough University,
Loughborough, UK, in 2008. Before joining De Montfort University in 2002,
Dr. Yang had worked at Beijing Jiaotong University, Nottingham Trent University,
and Loughborough University for various research projects. He has published over
100 papers on grey systems, fuzzy sets, rough sets, neural networks, and their
applications to civil engineering, transportation, and environmental engineering.
His research interests include the representation and modelling of various uncer-
tainties and the application of computational intelligence to real-world problems.
Dr. Yang is the executive president of International Association of Grey Systems
and Uncertainty Analysis, a senior member of IEEE Systems, Man and Cybernetics
Society, a co-chair of IEEE SMC Technical Committee on Grey Systems and the
vice-chair of the Task Force on Competitions for Fuzzy Systems Technical
Committee of IEEE Computational Intelligence Society. He is also a member of the
Analytics and Risk Technical Committee of IEEE Systems Council. He has been
involved in over 15 research projects with total fund value exceeding £1 million
from various funding sources including EU FP6, EU FP7, European Space Agency,
Royal Society, EPSRC, Royal Academy of Engineering, Leverhulme Trust, and De
Montfort University. He is currently focusing on extended fuzzy sets, grey systems,
rough sets, and their applications to decision support and risk analysis. Dr. Yang
has co-chaired several international conferences/special sessions,and been involved
as programme committee member in over 80 international conferences. Dr. Yang is
serving as associate editors in 5 international academic journals, including IEEE
Transaction on Cybernetics, Journal of Intelligent and Fuzzy Systems, World
Scientific Journal, Journal of Grey Systems, and Journal of Grey System: Theory
and Applications. His pioneer work on factor analysis using neural networks has
been adopted by many researchers since its publication more than 10 years ago. His
work on the integration of neural networks with GIS is an early attempt in this field.
His work on extended fuzzy sets and grey systems has been well received in
research community.
Prof. Jeffrey Forrest earned all his educational degrees in pure mathematics. His
Ph.D. degree was granted in 1988 by Auburn University, Alabama; and he did
one-year postdoctoral research in statistics at Carnegie Mellon University,
Pittsburgh, during 1990–1991. Dr. Forrest is currently a specially appointed pro-
fessor in economics, finance, and systems science of Nanjing University of
Aeronautics and Astronautics; a specially appointed professor in mathematics and
systems science of National University of Defense Technology, China; and a
tenured professor of mathematics at Slippery Rock University of Pennsylvania.
Dr. Forrest is a founder and the present president of the International Institute for
General Systems Studies (IIGSS), a non-profit organization registered in PA in
the mid-1990s. Since 1984, Forrest has had over 300 research papers and over
About the Authors xxvii

30 monographs and edited volumes published by a large array of prestigious

publishers, such as Springer, World Scientific, Kluwer Academic (currently part of
Springer), Academic Press (currently part of Elsevier), Wiley, Taylor and Francis,
and Meteorological Press. He serves or served on the editorial boards of 11 pro-
fessional journals, including Kybernetes: The International Journal of Systems,
Cybernetics and Management Science, Journal of Systems Science and
Complexity, and International Journal of General Systems. He is currently a
co-editor of the book series “Systems Evaluation, Prediction and Decision-
Making”, published by CRC Press, an imprint of Taylor and Francis.
Some of Dr. Forrest’s research was funded by the United Nations, the State of
Pennsylvania, the National Science Foundation of China, and the German National
Research Centre for Information Architecture and Software Technology. Over the
years, Dr. Forrest's scientific achievements have been recognized by various pro-
fessional organizations and academic publishers. In 2001, he was inducted into the
honorary fellowship of the World Organisation of Systems and Cybernetics. In
2016, he was elected to fellowship of the International Association of Grey Systems
and Uncertainty Analysis. His research interests are wide ranging, covering areas
such as mathematical and general systems theory and applications, foundations of
mathematics, data analysis, predictions, economics and finance, management sci-
ence, and philosophy of science.
Abstract

This book systematically presents the fundamental methods, models, and practical
application techniques of grey system theory. It is a result of the authors’ combined
expertise in theoretical exploration, real-life application, and teaching in this
research area, which the authors have developed over the past thirty years. This
book covers up-to-date theoretical and applied advances in grey systems from
across the world and vividly presents the reader with the overall picture of this new
theory and its frontier research.
This book contains 12 chapters, including Introduction to Grey System Theory,
The Novel Framework of Grey System Theory, Grey Numbers and their Operations,
Sequence Operators and Grey Data Mining, Grey Incidence Analysis Models, Grey
Cluster Evaluation Models, Series of GM Models, Combined Grey Models,
Techniques for Grey System Forecasting, Grey Models for Decision-Making,
Techniques for Grey Control, and Introduction to the Software of Grey System
Modeling. This is the first book of its kind to address general grey numbers and their
operations, the axiomatic system of buffer operators and a series of weakening and
strengthening operators, absolute degree of grey incidence model, relative degree of
grey incidence model, synthetic degree of grey incidence model, grey incidence
models based on similarity and closeness, three-dimension degree of grey incidence
models, grey fixed weight clustering model, grey evaluation models based on mixed
possibility functions, original difference grey model (ODGM), even difference grey
model (EDGM), discrete grey model (DGM), multi-attribute intelligent grey target
decision models, weight vector group of kernel clustering, and the weighted coef-
ficient vector of kernel clustering for decision-making.
This book will be appropriate as a reference and/or textbook for graduate stu-
dents or high-level undergraduate students, majoring in areas of science, technol-
ogy, agriculture, medicine, astronomy, earth science, economics, and management.
It can also be utilized as a reference book by researchers and technicians in research
institutions, business entities, and government agencies.

xxix
Chapter 1
Introduction to Grey Systems Research

1.1 Appearance and Growth of Grey Systems Research

On the basis of dividing the spectrum of scientific and technological endeavors into
fine sections, the overall development of modern science has shown the tendency of
synthesis at a high level. This higher level synthesis has led to the appearance of
various studies of systems science with their specific methodological and episte-
mological significance. Systems science reveals deep and intrinsic interconnections
between objects and events, and has greatly enriched the overall progress of science
and technology. Many of the historically difficult problems in different scientific
fields have been resolved successfully along with the appearance of systems science
and its specific branches. Furthermore, because of the emergence of various new
areas in systems science, our understanding of nature and the laws that govern
objective evolutions has been gradually deepened. At the end of the 1940s, there
appeared systems theory, information theory, and cybernetics. Toward the end of
1960s and the start of 1970s, there appeared the theory of dissipative structures,
synergetics, catastrophe, and bifurcations. Then, in the mid to late 1970s, new
transfield and interfiled theories of systems science such as the ultracircular theory
and dynamic systems theory emerged.
Due to both the existence of internal and external disturbances and the limita-
tions of our understanding, when investigating systems the available information
tends to contain various kinds of uncertainty and noises. Along with the develop-
ment of science, technology and the progress of the mankind, our understanding of
systems’ uncertainties has been gradually deepened and the research of uncertain
systems has reached a new height. During the second half of the 20th century, the
seemingly non-stoppable emergence of various theories and methodologies of
uncertain systems has been particularly significant in the areas of systems science
and systems engineering. For instance, L.A. Zadeh established fuzzy mathematics
in the 1960s (Zadeh 1965), J.L. Deng developed grey systems theory (Deng 1982)
and Z. Pawlak advanced rough set theory in the 1980s (Pawlak 1982). These works

© Springer Science+Business Media Singapore 2017 1

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_1
2 1 Introduction to Grey Systems Research

represent some of the most important efforts in uncertain systems research of this
time period and provide the theories and methodologies for describing and dealing
with uncertain information from different angles.
Grey systems theory, established by Julong Deng in 1982, is a relatively new
methodology that focuses on the study of problems involving small samples and
poor information. It deals with uncertain systems that contain partially known
information through generating, excavating, and extracting useful information from
what is available. Through this process, systems’ operational behaviors and their
laws of evolution can be correctly described and effectively monitored. In the
natural world, uncertain systems with small samples and poor information exist
commonly. This fact determines the wide applicability of grey systems theory.

1.2 Development History and Current State

In 1982, Professor Julong Deng’s paper titled “The Control Problems of Grey
Systems” was the first paper on grey systems to be published in the Systems and
Control Letters journal (Deng 1982a). In that same year, Professor Deng also
published “Grey Control System” in Chinese and the paper was published by the
Journal of Huazhong University of Science and Technology (Deng 1982b). The
publication of these two seminal articles indicated that a new and cross-sectional
discipline named grey system theory came into the world.
In 1989, The Journal of Grey System was launched by Research Information Ltd
in the UK. Currently, this publication is indexed by Mathematical Review of the
United States, Science Citation Index, and other important indexing agencies from
around the world. In 1997, a Chinese publication named Journal of Grey System,
was launched in Taiwan, China, and it was only in 2004 that this publication began
to be published in English. Additionally, in 2011 Emerald launched a new journal
named Grey Systems: Theory and Application, edited by the faculty of the Institute
for Grey System Studies at Nanjing University of Aeronautics and Astronautics.
There are currently over one thousand different professional journals in the world
that have published papers in grey systems theory, many of which are top journals
in a variety of fields. As of this writing, many journals and publishers such as the
journal of the Association for Computing Machinery (USA), Communications in
Fuzzy Mathematics (Taiwan, China), Kybernetes: The International Journal of
Systems and Cybernetics, Transaction of Nanjing University of Aeronautics and
Astronautics, China Ocean Press, Chinese Agricultural Science Press, Henan
University Press, Huazhong University of Science and Technology Press Co. Ltd,
IEEE Press, Springer-Verlag have respectively published special issues or pro-
ceedings on grey system theory.
Numerous universities around the world have set up grey system theory cur-
riculums. For example, in Nanjing University of Aeronautics and Astronautics
(NUAA), the curriculums of the grey system theory are found not only in PhD and
Master’s programs, but also in undergraduate programs of many disciplines across
1.2 Development History and Current State 3

the university, as an elective module. In 2008, the grey system theory course of
NUAA was selected one of the national level model courses in China. In 2013, the
same course was selected as the national excellent resource sharing course, which
became a free open learning resource for all grey system hobbyists.
Many universities are recruiting and funding doctoral and postdoctoral researchers
in grey system theory and its application. Examples include Huazhong University of
Science and Technology, Nanjing University of Aeronautics and Astronautics,
Southeast University, Wuhan University of Technology, Fuzhou University, Shantou
University, America Central Florida University, Nebraska-Lincoln University,
Canada Warterloo University, Toronto University, De Montfort University, Spain
Pablo de Olavide University, Turkey Bogazici University, Cape Town University in
South Africa, Romania Bucharest Economics University, Japan Kanagawa
University and many universities in Taiwan. There are also tens of thousands of
graduate students and PhDs currently engaged in scientific research applying grey
system thinking and methods across the world.
Numerous publishing agencies such as Science Press, Defense Industries Press,
Huazhong University of Science and Technology Press Co. Ltd, Jiangsu Science
and Technology Press, Shandong People’s Press, Science and Technology
Literature Press of China, China Science and Technology Book Press of Taiwan,
Gaoli Books Limited Company of Taiwan, ASE Press of Romania, Japan
Polytechnic Press, IIGSS Academic Press, CRC of Taylor & Francis Group,
Springer-Verlag, Springer-Verlag London Ltd, and John Wiley & Sons, Inc. have
published hundreds of academic works on grey systems, in many different lan-
guages including Chinese, Traditional Chinese, English, Japanese, Korean,
Romanian, and German.
Additionally, over the years a group of cutting-edge disciplines such as grey
hydrology, grey geological geology, grey breeding, and grey medical science has
emerged. Following this, many national and local science funding agencies are
actively supporting grey system research. There are hundreds of research projects
on grey systems and their applications currently receiving support from the
National Natural Science Foundation of China, The European Commission, The
Royal Society, Leverhulme Trust, as well as Canada, Spain, and Romania national
funds.
Since 2000, 18 regional domestic conferences on grey system theory and its
applications have been held. Such conferences have been supported by The
Leverhulme Trust, Institute for Grey System Studies, Nanjing University of
Aeronautics and Astronautics, De Montfort University, Wuhan University of
Technology, Educational Society of Pudong, Shanghai, and China Center of
Advanced Science and Technology; the latter directed by Mr. Tsung-Dao Lee, a
Nobel Prize winner, and two of the former presidents of the Chinese Academy of
Sciences, Mr. Zhou Guangzhao and Mr. Lu Yongxiang. As a result, grey system
theory has attracted a large number of young scholars to such events.
Many special sessions and tracks on grey system theory have been organized at
significant international conferences such as International Conference on Uncertain
System Modeling, International Conference on System Forecast and Control,
4 1 Introduction to Grey Systems Research

International Conference on General System Studies, International Congress of

World Organization of Systems and Cybernetics, IEEE International Conference on
Systems, Man and Cybernetics. The topicality of grey systems theory and its
popularity in such high-profile international conferences have certainly played an
active role in furthering understanding of, and promoting, this theory among peers
in the world of systems science.
In 2007, 2009, 2011, 2013 and 2015, the first, second, third, fourth and fifth
IEEE International Conference on Grey Systems and Intelligent Services were held
in Nanjing,Macao and Leicester, respectively. Each conference received a signifi-
cant number of submissions from many countries and regions including China,
America, England, Germany, France, Spain, Switzerland, Hungary, Poland, Japan,
South Africa, Russia, Turkey, Romania, Holland, Malaysia, Iran, Ukraine,
Kazakhstan, Pakistan, Iran, Taiwan, Macao, and Hong-Kong. More than one
thousand articles featured in the five conferences were indexed by EI database,
among which more than three hundreds excellent papers were published by
Kybernetes, Grey Systems: Theory and Application, The Journal of Grey System,
Transaction of Nanjing University of Aeronautics and Astronautics (English ver-
sion) and Springer-Verlag.
Many prominent scholars have commended grey system research. Such scholars
include Professor Lotfi A. Zadeh (US), the founder of fuzzy mathematics, Professor
Herman Haken (Germany), the founder of synergetics, Professor James M. Tien
(US), former vice-president of IEEE and member of the National Academy of
Engineering, Professor Robert Valee (France), president of World Organization of
Systems and Cybernetics, Professor Alex Andrew (UK), the Secretary General of
the World Organization of Systems and Cybernetics and President of the Canadian
Royal Academy of Sciences, as well as many Academicians of the Chinese
Academy of Sciences and the Chinese Academy of Engineering, including
Professor Qian Xuesen, Professor Yang Shuzi, Professor Xiong Youlun, Professor
Lin Qun, Professor Chen Da, Professor Zhao Chunsheng, Professor Hu Haiyan,
Professor Xu Guozhi, Professor Wang Zhongtuo and Professor Yang Shanlin.
In 2005, the Grey System Society of China,namely CSOOPEM, was approved
by China Association for Science and Technology, and the Ministry of Civil Affairs
of China. At the beginning of 2008, the Technical Committee of IEEE SMC on
Grey Systems was established. In 2012, the first Workshop of European grey
system research collaboration network was held by De Montfort University, and
delegates from twelve member states of the European Union attended the event. In
2013, Professor Sifeng Liu was selected for a Marie Curie International Incoming
Fellowship (FP7-PEOPLE- IIF-GA-2013-629051) of the 7th Research Framework
Program of the European Commission. Furthermore, in 2014 an international
network project entitled “Grey Systems and Its Applications” (IN-2014-020) was
funded by The Leverhulme Trust. Supported by this project, a series of grey system
theory cooperative research and academic exchange activities have been held in
Europe, North America and China. In 2015, the International Association of Grey
Systems and Uncertainty Analysis (GSUA) was established. As an emerging dis-
cipline, grey system theory is standing strong in the scientific community.
1.3 Characteristics of Uncertain System 5

1.3 Characteristics of Uncertain System

The fundamental characteristic of uncertain systems is the incompleteness and

inadequacy of their information. Due to the dynamics of system evolution, the
biological limitations of the human sensory system, as well as the constraints of
relevant economic conditions and technological availabilities, uncertain systems
exist commonly.

1.3.1 Incomplete Information

Incompleteness in information is one of the fundamental characteristics of uncertain

systems. The most common situations involving incomplete system information
include cases where:
(1) Information about system elements (parameters) is incomplete;
(2) Information on the structure of the system is incomplete;
(3) Information about the boundaries of the system is incomplete; and
(4) Information on the system’s behaviors is incomplete.
Incomplete information is a common phenomenon in our social, economic, and
scientific research activities. For instance, in agricultural production, even if we
have exact information regarding plantation, seeds, fertilizers, and irrigation,
uncertainties in areas such as labor quality, natural environment characteristics,
weather conditions, and the commodity markets make it extremely difficult to
precisely predict the production output and consequent economic value of agri-
cultural fields. For biological prevention systems, even if we know the relationship
between insects and their natural enemies, it is still really difficult to achieve the
expected prevention effects due to uncertainty regarding the relationships between
insects and their baits, insects’ natural enemies and their baits, and a specific kind of
natural enemy with another kind of natural enemy. As for the adjustment and
reform of pricing systems, it is often difficult for policy makers to take actions
because of the lack of information regarding price elasticity of demand and how
price changes on a certain commodity would affect the prices of other commodities.
In security markets, even the brightest market analysts cannot be assured of winning
constantly due to their inability to correctly predict economic policy and interest
rate changes, management changes at various companies, the direction of political
changes, investors’ behavioral changes in international markets, and the effects of
price changes in one block of commodities on another. As for the general economic
system, because there are no clear relationships between the “inside” and the
“outside” of the system, and between the system itself and its environment, and
because the boundaries between the inside and the outside of the system are difficult
to define, it is also difficult to analyze the effects of economic input on economic
output.
6 1 Introduction to Grey Systems Research

Incompleteness in available information is absolute, while completeness in

information is relative. Humans employ their limited cognitive ability to observe
the infinite universe in order to try and obtain complete information. However, it is
impossible for us to do so. In fact, the concept of large samples in statistics rep-
resents the degree of tolerance man has to incompleteness. In theory, when a sample
contains at least 30 objects, it is considered “large.” However, in some situations,
even when a sample contains thousands or several tens of thousands of objects, the
true statistical laws of a given system still cannot be successfully uncovered.

1.3.2 Inaccuracies in Data

Another fundamental characteristic of uncertain systems is naturally occurring

inaccuracy in available data. In grey systems theory, the meanings of uncertain and
inaccurate are roughly the same. Both terms stand for errors or deviations from
actual data values. Based on the essence of how uncertainties are caused, inaccu-
racies can be categorized into three types: the conceptual, level, and prediction type
inaccuracies.
(1) The Conceptual Type
Inaccuracies of the conceptual type emanate from the expression of a certain
event, object, concept, or wish For instance, all such frequently used concepts
as “large,” “small,” “many,” “few,” “high,” “low,” “fat,” “thin,” “good,”
“bad,” “young,” and “beautiful” are inaccurate due to lack of clear definition.
It is very difficult to use exact quantities to express these concepts. As a second
example, suppose that a job seeker with an MBA degree wishes to get an
annual salary offer of no less than $150,000, or that a manufacturing firm plans
to control its rate of defective products to be less than 0.1 %. These are all
cases of conceptual type inaccuracies.
(2) The Level Type
This kind of data inaccuracy is caused by a change at the level of research or
observation. This means that the available data might be accurate when seen at
the level of the system of concern, that is, the macroscopic level, or at the level
of the whole, that is, the cognitive conceptual level. However, when data are
seen at a lower level, that is, a microscopic level, or at a partial localized level
of the system, they generally become inaccurate. For example, the height of a
person can be measured accurately to the unit of centimeters or millimeters.
However, if the measurement has to be accurate to the level of one
ten-thousandth micrometers, the former accurate reading will become extre-
mely inaccurate.
(3) The Prediction (or Estimation) Type
Because it is difficult to have complete understanding of the laws of evolution,
any prediction of the future tends to be inaccurate. For instance, it is estimated
that two years from now, the GDP of a certain country will surpass $10 billion
1.3 Characteristics of Uncertain System 7

dollars; it is estimated that a certain bank will attract savings from individual
residents of between $70 thousand and $90 thousand for the year 2017; it is
predicted that in the coming years the temperature in Leicester, UK, during the
month of June will not go beyond 30 °C, and so on. All these examples
provide uncertain numbers of the prediction type. In statistics, it is often the
case that samples are collected to estimate the whole. Therefore, much sta-
tistical data are inaccurate. As a matter of fact, no matter what method is used,
it is very difficult for anyone to obtain any absolutely accurate (estimated)
value. When we draw out plans for the future and make decisions about what
course of action to take, we in general have to rely on inaccurate predictions
and estimates.

1.3.3 The Scientific Principle of Simplicity

In the history of science, the achievement of simplicity has been a common goal
among most scientists. As early as the sixth century BC, natural philosophers had a
common wish to understand the material laws of nature: to build knowledge of the
material world on the basis of a few common, simple elements. The ancient
Pythagoras of Greece introduced the theory of four elements (earth, water, fire, and
gas) at around 500 BC. The Greeks believed that all material matters in the universe
were composed of these four simple elements. Around the same time, ancient
Chinese philosophers also developed a theory of five elements including water, fire,
wood, gold, and earth. These are the most primitive and elementary thoughts about
simplicity.
The scientific principle of simplicity originates from the simplicity of thinking
employed in the process of understanding nature. As the natural sciences matured
over time, simplicity became the foundation and guiding principle of scientific
research. For example, Newtonian laws of motion unify the macroscopic phe-
nomena of objective movements in their form of extreme simplicity. In his
Mathematical Principles of Natural Philosophy, Newton pointed out that nature
does not do useless work; because nature is fond of simplicity, it does not like to
employ extra reasons to flaunt itself. During the ear of relativity, Albert Einstein
introduced two criteria for testing a theory: external confirmation and internal
completeness, that is, logical simplicity. Einstein believed that a true scientific
theory must comply with the principle of simplicity in order to reflect the harmony
and orderliness of nature. In the 1870s, Ampere, Weber, Maxwell, and others
established theories to explain the phenomenon of electromagnetism based on their
different assumptions. Because Maxwell’s theory is the one that best complies with
the principle of simplicity, it became well accepted. Another example is the
well-known Kepler’s third law of planetary motion: T2 = D3; in form it looks very
simple.
8 1 Introduction to Grey Systems Research

According to the slaving principle of synergetics (H. Haken, 1978), one can
transform an original high-dimensional equation into a low-dimensional evolution
equation of order-parameters by eliminating the fast-relaxing variables in the
high-dimensional nonlinear equation that describes the evolution process of a
system. Because the order-parameters dominate the dynamic characteristics of the
system near its boundary points, through solving the evolution equation of
order-parameters one can obtain the system’s time structure, space structure or
time-space structure, so that one can materialize efficient control over the system’s
behavior.
The simplicity of scientific models is actualized by employing simple expres-
sions and by ignoring unimportant factors of the system of concern. In economics,
the methods of using Gini coefficient to describe differences among consumers’
incomes (Gini 1921) and of employing Cobb-Dauglas production function to
measure the contribution of advancing technology in economic growth are all
introduced on the basis of simplifying realistic systems (Cobb and Douglas 1928).
Modigliani and Brumbergh (1954) use the following model to describe the average
propensity to consume:

Ct y0
¼ a þ b ; a [ 0; b [ 0
yt yt

The curve Phillips (1958) employs to describe the relationship between the rate
of inflation Dp
p and the unemployment rate x is:

Dp 1
¼ aþb
p x

Additionally, the well-known capital asset pricing model (CAPM, Sharpe 1964)
can be seen below:

E½ri  ¼ rf þ bi (E[rm   rf )

Essentially, all of these equations can be reduced to the simplest linear regres-
sion model with a few straightforward transformations.

1.3.4 Precise Models Suffer from Inaccuracies

When available information is incomplete and the collected data inaccurate, any
pursuit of precise models in general becomes meaningless. This fact was well
described by Lao Tzu more than two thousand years ago. The principle of
incompatibility proposed by L.A. Zadeh, the founder of fuzzy mathematics, also
addresses this matter: when the complexity of a system increases, our ability to
precisely and meaningfully describe the characteristics of the system decreases
1.3 Characteristics of Uncertain System 9

accordingly until such a threshold that, as soon as it is surpassed, the preciseness

and meaningfulness become two mutually excluding characteristics (Zadeh 1994).
This mutually antagonistic principle reveals that the pursuit of preciseness can
reduce the operationality and meaningfulness of a cognitive outcome. Therefore,
precise models are not necessarily an effective means to address complex matters.
In 1994, Jiangping Qiu and Xisheng Hua respectively established a theoretically
delicate statistical regression model and relatively coarse grey model based on the
deformation data and leakage data of a certain large scale hydraulic dam. Xisheng
Hua’s work shows that his grey model provided a better fit than Jiangping Qiu’s
statistical regression model. When comparing the errors between the predictions of
the two models with actual observations, it is found that the prediction accuracy of
the grey model is generally better than that of the regression model; see Table 1.1
for details.
In 2001, Dr. Haiqing Guo as well as Zhongru Wu and colleagues respectively
established a statistical regression model and a grey time series combined model
using the observational data of displacement in the vertical direction of a certain
large clay-rock filled dam of inclined walls. They compared the data fit and pre-
dictions of the two models against actual observations and found that the data fit of
the grey combined model was significantly superior to that of the statistical
regression model.
On the other hand, Xiaobing Li, Haiyan Sun and colleagues employed fuzzy
prediction functions (a type of uncertainty prediction) to dynamically trace and
precisely control the fuel oil feeding temperature for anode baking. The control
effect was clearly better than that obtained by utilizing the traditional PID control
method (Li and Sun 2009).
Finally, Caixing Sun and his research group made use of grey incidence analysis,
grey clustering, and various new types of grey prediction models to diagnose and
predict insulation-related accidents related to electric transformers. Their substantial
results indicate that these relatively coarse methods and models are more opera-
tional and provide more efficient results than traditional models (Sun et al. 2002,
2003; Li et al. 2002).

Table 1.1 Comparison between the prediction errors of a statistical model and a grey model
Order No. Type Average error
Statistical model Grey model
1 Horizontal displacement 0.862 0.809
2 Horizontal displacement 0.446 0.232
3 Vertical displacement 1.024 1.029
4 Vertical displacement 0.465 0.449
5 Water level of pressure measurement hole 6.297 3.842
6 Water level of pressure measurement hole 0.204 0.023
10 1 Introduction to Grey Systems Research

1.4 Comparison of Several Studies of Uncertain Systems

Probability and statistics, fuzzy mathematics, grey system theory and rough set theory
are four of the most widely used research methods in the investigation of uncertain
systems. Their research objects contain specific kinds of uncertainty, which represent
their commonality. It is precisely the differences among the uncertainties in the
research objects that make these four theories of uncertainty distinct from each other.
Probability and statistics study the phenomena of stochastic uncertainty with
emphasis placed on revealing historical statistical laws. They investigate the chance
of each possible outcome of the stochastic uncertain phenomenon to occur. Their
starting point is the availability of large samples, which are required to satisfy a
typical form of distribution.
Fuzzy mathematics emphasizes the investigation of problems with cognitive
uncertainty, where research objects possess the characteristic of clear intension and
unclear extension. For instance, “young man” is a fuzzy concept, because each
person knows the intension of “young man.” However, if we determine the exact
age range within which everybody is young and outside which each person is not
young, then we will have great difficulty. That is because the concept of young man
does not have a clear extension. In fuzzy mathematics, this kind of cognitive
uncertainty problem with clear intension and unclear extension is addressed by
making use of experience and the so-called membership function.
Additionally, rough set theory tries to study uncertain systems by using the
accuracy mathematical method. The main thought of rough set theory is to describe
and address the inaccuracy or uncertain knowledge using the known knowledge
library. Professor Z. Pawlak included all the units which cannot be acknowledged to
have boundaries. He defined boundary as the difference set between upper
approximate set and lower approximate set. The boundary is then described through
the upper approximate set approaching the lower approximate set.
The focus of grey system theory, on the other hand, is on the uncertainty
problems of small data sets and poor information, which are different to the
problems addressed by probability, fuzzy mathematics or rough set theory. It
explores and uncovers the realistic laws of evolution, motion of events and mate-
rials through information coverage by possibility function, and through the works
of sequence operators. One of its characteristics is construct models with small
amounts of data. What is clearly different about grey systems theory compared to
fuzzy mathematics is that grey system theory emphasizes the investigation of
objects that process clear extension and unclear intension. For example, by the year
of 2050, China will control its total population within the range of 1.5–1.6 billion
people. This range from 1.5 to 1.6 billion is a grey concept. Its extension is definite
and clear. However, if one inquires further regarding exactly which specific number
within the said range it will be, then he will not be able to obtain any meaningful
and definite answer. The possibility function is used to describe the possibility that
a value is a grey number. We summarize the differences among these four main
uncertainty research methods in Table 1.2.
1.5 Most Actively Studied Uncertain Systems Theories 11

1.5 Most Actively Studied Uncertain Systems Theories

Fuzzy mathematics, grey system theory, and rough set theory are currently the most
actively studied uncertainty theories. From the ISI and EI Compendex databases it
is found that the number of research papers with one of the keywords “fuzzy set,”
“grey system,” and “rough set” has increased rapidly (see Table 1.3).
A search using the Chinese Database of Scholarly Periodicals (CNKI) shows
that, from 1990 to 2015, the number of scholarly publications with one of the
keywords “fuzzy mathematics,” “grey system,” and “rough set,” presents an
up-trending development (see Tables 1.4, 1.5 and 1.6).
Research on uncertain (fuzzy, grey, and rough) systems can be categorized into
the following three aspects:
(1) The mathematical foundation of uncertain systems theories;
(2) The modeling of uncertain systems and computational schemes, including
various uncertain systems modeling, modeling combined with other relevant
methods, as well as related computational methods; and
(3) The wide-range of applications of uncertain systems theories in natural and
social sciences.
Uncertain (fuzzy, grey, rough) systems theories have been widely applied in
many areas of natural science, social science, and engineering, including aviation,
spaceflight, civil aviation, information, metallurgy, machinery, petroleum, chemical
industry, electrical power, electronics, light industries, energy resources, trans-
portation, medicine, health, agriculture, forestry, geography, hydrology, seismol-
ogy, meteorology, environment protection, architecture, behavioral science,

Table 1.2 Comparison among the four methods of uncertainty research

Uncertainty Grey system Prob. statistics Fuzzy math Rough set
research
Research objects Poor Stochastics Cognitive Boundary
information
Basic set Grey number Cantor set Fuzzy set Approximate set
set
Describe method Possibility Density func. Membership func. Upper, lower
func. appr.
Procedure Sequence Frequency Cut set Dividing
operator
Data Any Known Known Equivalent rel.
requirement distribution distribution membership
Emphasis Intension Intension Extension Intension
Objective Law of reality Historical law Cognitive Approx.
expression approaching
Characteristics Small data Large sample Depend on Information form
experience
12 1 Introduction to Grey Systems Research

Table 1.3 Search outcomes of ISI and EI Compendex databases(2010–2015)

Keyword Fuzzy set Grey system Rough set
# in the ISI database 17,086 4398 7231
# in the EI Compendex database 9670 5087 6588

Table 1.4 Search outcomes of “fuzzy mathematics”

Time 1990 1991 1992 1993 1994 1995 1996 1997 1998
# of papers 345 373 401 346 543 575 574 551 514
Time 1999 2000 2001 2002 2003 2004 2005 2006 2007
# of papers 530 605 583 598 714 720 799 908 1006
Time 2008 2009 2010 2011 2012 2013 2014 2015 Total
# of papers 933 911 880 850 800 775 752 699 17,285

Table 1.5 Search outcomes of “grey system”

Time 1990 1991 1992 1993 1994 1995 1996 1997 1998
# of papers 149 181 195 203 517 477 481 483 448
Time 1999 2000 2001 2002 2003 2004 2005 2006 2007
# of papers 456 418 435 512 556 550 576 652 730
Time 2008 2009 2010 2011 2012 2013 2014 2015 Total
# of papers 762 733 765 693 695 602 517 502 13,288

Table 1.6 Search outcomes of “rough set”

Time 1990 1991 1992 1993 1994 1995 1996 1997 1998
# of papers 0 0 0 0 0 0 1 1 9
Time 1999 2000 2001 2002 2003 2004 2005 2006 2007
# of papers 19 50 102 142 267 412 553 710 779
Time 2008 2009 2010 2011 2012 2013 2014 2015 Total
# of papers 919 1018 954 830 809 698 675 651 9599

management science, law, education and military science. These practical appli-
cations of uncertain systems theories have brought significant social and economic
benefits.
Both theoretical and applied research on uncertain systems has been extremely
active. However, such research has shown an emphasis on applications without
much attention to theory development and methodological innovation. In particular,
not enough attention has been given to investigating the differences and com-
monalities among the various uncertain systems theories, and scant work has been
done to try and combine traditional theories and methods with emerging uncertain
systems theories and methods This fact more or less affected the development of
uncertain (fuzzy, grey, rough) systems theories.
1.5 Most Actively Studied Uncertain Systems Theories 13

In fact, traditional and emerging uncertain systems theory and methods are
complementary and cannot be clearly separated. Each uncertain systems theory and
method has its own strengths and can be used for different kinds of uncertainty
problems: such theories complement and supplement each other and are not the-
oretically incommensurable. Many complex, dynamic uncertainty problems are
way beyond the scope and capacity of any single uncertain systems theory and
method. They require that the researcher combine various kinds of traditional
theories with new uncertain systems theories and methods. Further research seeking
to develop this kind of interaction, exchange, and combination of relevant theories
and methods is essential for the future development of science.

1.6 Elementary Concepts of Grey System

Many social, economic, agricultural, industrial, ecological and biological systems

are named by considering the features of classes of the research objects, while grey
systems are labeled using the color of the systems of concern.
In the theory of control, scholars often make use of colors to describe the degree
of clearness of available information. For instance, Ashby refers to objects with
unknown internal information as black boxes. This terminology has been widely
accepted in the scientific community. As another example, as a society moves
toward democracy, citizens gradually demand more information regarding policies
and the meanings of such policies. That is, citizens want to have an increased
degree of information transparency (i.e. white information). Thus, we use “black” to
indicate unknown information, “white” to indicate completely known information
and “grey” to convey partially known and partially unknown information.
Accordingly, systems with completely known information are regarded as white,
while systems with completely unknown information are considered black, and
systems with partially known information and partially unknown information are
seen as grey.
In this context, incompleteness in information is the fundamental meaning of
“grey.” However, the meaning of “grey” can be expanded or stretched from dif-
ferent angles and in varied situations (see Table 1.7).

Table 1.7 Extensions of the concept of “grey”

Situation/concept Black Grey White
Information Unknown Incomplete Completely known
Appearance Dark Blurred Clear
Processes New Changing Old
Properties Chaotic Multivariate Order
Methods Negation Change for the better Confirmation
Attitude Letting go Tolerant Rigorous
Outcomes No solution Multi-solutions Unique solution
14 1 Introduction to Grey Systems Research

At this point, the difference between “system” and “box” Must be highlighted.
On the one hand, the term “box” is used when one does not pay much attention, or
does not attempt, to utilize information regarding the interior characteristics of an
object, while focusing mainly on the external characteristics of such an object. In
this case, the researcher generally investigates the properties and characteristics of
the object through analyzing the input-output relation. On the other hand, the term
“system” is employed to indicate the study of the object’s structure and functions
through the analysis of existing organic connections between the object, relevant
factors, its environment, and related laws of change.
The research objects of grey systems theory consist of uncertain systems that are
known only partially with small samples and poor information. The theory focuses
on the generation and excavation of partially known information through grey
sequence operators of possibility functions to enable an accurate description and
understanding of the material world.

1.7 Fundamental Principles of Grey Systems

In the process of developing grey systems theory, Julong Deng established six
fundamental principles containing intrinsic philosophical intensions, as discussed
below (Deng 1985).
Axiom 1.7.1 The Principle of Informational Differences “Difference” implies
the existence of information. Each piece of information must carry some kind of
“difference”.
When we say that object A is different from object B, we mean that there is some
special information about object A that is not true for object B. All “differences”
between natural objects and events have provided us with elementary information
in order for us to understand their nature.
If information “I” has changed our understanding or impression of a complicated
matter, then the piece of information “I” is definitely different from what we ini-
tially understood the complicated matter to be. Great breakthroughs in science and
technology have provided us with necessary information, which we generally call
knowledge and tools, to understand and change the world around us. Such
advanced information is surely different from pre-scientific information. The more
content a piece of information “I” contains, the more the differences from an earlier
version of such information will become apparent.
Axiom 1.7.2 The Principle of Non-Uniqueness The solution to any problem with
incomplete and indeterminate information is not unique.
Because of the principle of non-uniqueness, which is a basic law of the appli-
cation of grey systems theory, one is set free to look at problems with flexibility.
With flexibility, one becomes more effective in reaching their goals.
1.7 Fundamental Principles of Grey Systems 15

Strategically, the principle of non-uniqueness is realized through the concept of

grey target. This concept is a unification of the concept of non-unique target and
that of non-restrainable target. For example, on the one hand, if a high school
graduate does not plan to enroll in any university except for one specific institution,
then his chance of being accepted by a university is greatly limited. On the other
hand, if a high school graduate with similar qualifications as the one in the previous
example is willing to apply for several universities other than his preferred one, he
will be more likely to succeed in being accepted by a university because he has
multiple targets, which in turn leads to an improved chance of hitting one of the
targets.
The principle of non-uniqueness can be seen as a comprehensive realization that
each target can be approached, that any available information can be supplemented,
that each plan made earlier can be further modified and improved, that each rela-
tionship can be harmonized, that each thinking logic can be multi-directional, that
each understanding can be deepened, and that each path can be optimized. When
faced with the possibility of multiple solutions, one can locate one or several
satisfactory solutions through deterministic analysis and supplementation of
information. Therefore, the method of finding solutions on the basis of
“non-uniqueness” is one that combines both quantitative and qualitative analysis.
Axiom 1.7.3 Principle of Minimal Information One characteristic of grey system
theory is that it makes the most and best use of the “minimal amount of available
information.”
The “principle of minimal information” can be seen as a dialectic unification of
“a little” and “a lot.” One advantage of grey system theory is its ability to handle
such uncertain problems with “small data” and/or “poor information.” Its founda-
tion of study is the concept of “spaces of limited information.” “Minimal amount of
information” is the basic territory for grey system theory to show its power The
amount of acquirable information is the dividing line between “grey” and “not
grey”. Making sufficient discovery and application of any available “minimal
amount of information” is the basic thinking logic of problem-solving used in grey
system theory.
Axiom 1.7.4 Principle of Recognition Base Information is the foundation on
which people recognize and understand (nature).
This principle argues that all recognition must be based on information. Without
information, there is no way for people to know anything. With complete and
deterministic information, we can possibly gain firm understanding of nature. With
incomplete and non-deterministic information, it is only possible to obtain
incomplete and non-deterministic grey understanding of particular phenomena.
Axiom 1.7.5 Principle of New Information Priority The function of new pieces
of information is greater than that of old pieces of information.
The “principle of new information priority” is the key idea behind information
application in grey system theory. That is, by applying additional weights to new
16 1 Introduction to Grey Systems Research

information, one can achieve a better result from grey modeling, grey prediction,
grey analysis, grey evaluation, and grey decision making. The belief that “the new
replaces the old” reflects our “principle of new information priority.” With the
availability of new information, the motivation for whitening grey elements is
strengthened. The “principle of new information priority” reflects the fact that
information in general is time sensitive.
Axiom 1.7.6 Principle of Absolute Greyness “Incompleteness” of information is
absolute. Incompleteness and non-determinism of information have generality.
Completeness of information is relative and temporary. It is the moment when
the original non-determinism has just disappeared, and new non-determinism is
about to emerge. Human recognition and understanding of the objective world have
been improved over time through continued supplementation of information. With
endless supply of information, man’s recognition and understanding of the world
also become endless. That is, greyness of information is absolute and will never
disappear.
Chapter 2
The Grey Systems Theory Framework

Over the past 30 years, grey systems theory has developed as a scientific discipline
with its own theoretical structure consisting of systems analysis, evaluation, mod-
eling, prediction, decision-making, control, and techniques of optimization.

2.1 Grey Models and Framework

A grey number is a figure that represents a range of values rather than an exact
value when the exact value for the said figure is not known. The range of a grey
number can be an interval or a general number set. Grey numbers are usually
expressed as the symbol “⊗”, which is called grey. A grey number represents the
degree of information uncertainty in a given system. As the basis of grey systems
theory, research on grey numbers and grey measures has attracted increased
attention over the past years.
Professor Julong Deng presented the basic thinking and models of grey system
theory (GST) in his first paper, which was published in Systems and Control Letters
(Deng 1982). In 1985, he presented the GST framework in his two books, which
were published by National Defense Industry Press (Deng 1985a) and Huazhong
University of Science and Technology Press (Deng 1985b), respectively. Such
books addressed grey numbers and the operations of interval grey numbers, grey
incidence analysis, grey generation, grey clustering, grey forecasting model, grey
decision-making, and grey control. He confirmed the GST framework in his later
publications, such as “A course on Grey System Theory” (Deng 1990) and “The
Basis of Grey System Theory” (Deng 2002).

© Springer Science+Business Media Singapore 2017 17

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_2
18 2 The Grey Systems Theory Framework

2.2 The Thinking, Models and Framework of Grey

Systems Theory

2.2.1 Grey Numbers and Its Operations

In 2004, an axiomatic definition of the grey degree of a grey number was put
forward by Sifeng Liu and Yi Lin (Liu and Lin 2004), and such a definition was
based on a measure of grey number and its background or domain. This definition
of the grey degree of a grey number satisfied the requirement of standardability,
which provided the bedrock for us to cognize the uncertainty of grey information In
2010, the introduction of the unreduction axiom and a new definition of degree of
greyness of grey numbers were put forward. This allowed the operations of grey
numbers and grey algebraic system to be built based on the grey “kernel” and the
degree of greyness of grey numbers (Liu et al. 2010). On these grounds, the
operation of grey numbers has been transformed to the operation of real numbers.
Thus, to a certain extent the problem of setting up the operation of grey numbers
and the grey algebraic system has been solved.
In 2012, Sifeng Liu, Zhigeng Fang, Yingjie Yang and others developed the
concept of general grey number as follows:

[
n  
g
2 ai ;
ai
i¼1

S  
Among them, any interval grey numbers i 2 ½ai ; ai   ni¼1 ai ; ai , which
satisfy ai ; ai 2 < and ai1  ai  ai  ai þ 1 , g ¼ inf ai 2g
ai , g þ ¼ supai 2g
ai are
called the lower and upper limits of g
.
Such researchers also found that summation and subtraction operations about the
degree of greyness of grey numbers do not satisfy the introduction of the
un-reduction axiom. Then they verified summation and subtraction operations
about grey numbers and the grey synthesis axiom as specified below (Liu et al.
2012). The algebra system based on grey “kernel” and the degree of greyness of
grey numbers is shown in Fig. 2.1.
In addition, Zhigeng Fang and colleagues put forward the concept of standard
interval grey number and offered the algorithm of standard interval grey numbers
(Fang and Liu 2005), while Qiaoxing Li and colleagues raised grey number rules
based on numerical coverage (Li and Liu 2012). Such researchers carried out
beneficial exploration around grey numbers and related operations.
In the grey algebra system based on the grey “kernel” and the degree of
greyness of grey numbers, the principles of “take the bigger one” are still used to
calculate the degree of greyness of the “multiplication” and “division” operation
outcome, according to the introduction of the un-reduction axiom. To reveal the
inherent law of synthesis of degree of greyness in the process of operations of
2.2 The Thinking, Models and Framework … 19

 

Fig. 2.1 Algebra system based on grey “kernel” and degree of greyness of grey numbers

“multiplication” and “division” and structure a more exquisite principle for oper-
ations of “multiplication” and “division” is a critical problem waiting to be solved.

2.2.2 The Grey Sequence Operator

In order to solve the prediction problem of shock-disturbed systems, Sifeng Liu put
forward the concept of buffer operator developed the axioms for the buffer operator
system, and constructed several practical buffer operators (Liu 1991). One of most
used weakening buffer operators with acceptable properties is presented below:

1
xðkÞd ¼ ½xðkÞ þ xðk þ 1Þ þ    þ xðnÞ; k ¼ 1; 2;   ; n ð2:1Þ
n  kþ1

Research on buffer operators has become very active since 1991 and significant
new studies have emerged. For example, Yaoguo Dang (Dang et al. 2004), Zhengpeng
Wu (Wu et al. 2009), Jie Cui (Cui and Dang 2009), Lizhi Cui (Cui and Liu 2010),
Yeqing Guan (Guan and Liu 2008), Xiao-Li Hu (Hu et al. 2013), Yan Gao (Gao et al.
2013), Zhengxin Wang (Wang et al. 2009), Xuemei Li (Li et al. 2012) and Wenqiang
Dai (Dai and Su 2012), among others, have developed a variety of different weakening
and strengthening buffer operators based on three buffer operator axioms. In 2011,
Yong Wei brought forth the general formula of buffer operator:

1 X
n
xðkÞd ¼ xðkÞ  ½xðkÞ= Pn xi xðiÞa ð2:2Þ
i¼k xi i¼k
20 2 The Grey Systems Theory Framework

Yong Wei proved that the buffer operator given in Eq. (2.1) can express
weakening buffer operator, strengthening buffer operator and the identity operator,
respectively, according to the different values of a (Wei 2011).

2.2.3 The Grey Prediction Models

The grey prediction model is a type of grey model that has been widely researched
and used. In 2005, Naiming Xie built on Model GM (1, 1) by Professor Julong
Deng (Deng 1982), and proposed the discrete grey model as well as its properties
(Xie and Liu 2005). Later, Lifeng Wu developed the fractional accumulation dis-
crete grey model and completed the perturbation problem of grey model (Wu et al.
2013). Xiangdong Chen and Jun Xia set up the DHGM (2, 2) coupled equations,
combining grey differential equation and self-memory principle based on the power
system of self-memory principle (Chen and Xia 2009). Xiaojun Guo proposed the
interval grey number for the self-memory prediction model based on the degree of
greyness of synthesis grey numbers; he then studied the self-memory prediction
model from different perspectives (Guo et al. 2014).
Various developments and derived models have emerged in recent years. For
example, Yaoguo Dang came up with the GM (1, 1) model based on x(n) as the
initial condition (Dang and Liu 2004). Xican Li proposed the GM (1, 1, b) model,
studied the content type and parameter set form of the model, analyzed several
properties of the GM (1, 1, b) model, then developed its optimization algorithm (Li
et al. 2014). Zhengxin Wang provided several kinds of forms of GM (1, 1) power
model and studied the characteristics of their time response function (Wang 2008).
Xinping Xiao studied generalized accumulation grey model and proposed a com-
bined optimization method (Xiao 2000). Wuyong Qian developed the grey GM
(1, 1, ta)model with the time power item and studied the process of modeling and
parameter estimation (Qian and Dang 2000). Wanmei Tang proposed a new pre-
diction model based on the grey supporting vector machine (Tang 2006). Ke Zhang
put forward the multi variable discrete grey model based on driving control (Zhang
2014). Bo Zeng came up with the random oscillation sequence prediction model
taking the smooth operator compress random oscillation amplitude (Zeng et al.
2001). Qishan Zhang used the particle swarm algorithm and provided a new method
of increasing grey GM (1, 1) precision through the optimization of background
value interpolation coefficient and boundary value (Zhang 2001). Tianxiang Yao
studied the parameter characteristics of the new information discrete GM (1, 1)
model and fit properties of the geometric sequence, then put forward a new
information discrete GM (1, 1) model with sectional correction (Yao and Liu 2009).
Liyun Wu and Zhengpeng Wu constructed the twice time varying parameter dis-
crete grey model with features of the white index law coincidence, linear law
coincidence, twice law coincidence and stretching transformation consistency (Wu
et al. 2013). Carmona Benitez improved the GM model and forecasted long-term
trends in the passenger flow of the American air transport industry using the
2.2 The Thinking, Models and Framework … 21

improved model, which led to satisfying results (Benitez et al. 2013). Mark Evans
proposed a more general grey Verhulst model and forecasted changes in the
strength of British steel (Evans 2014). Jie Yang and Wenguo Weng made further
improvements to the unbiased grey model and forecasted the amount of gas supply
in a few cities (Yang and Weng 2014). Additionally, Naiming Xie studied the
prediction problems of grey number sequence (Xie and Liu 2004).
In 2015, Sifeng Liu and colleagues determined four kinds of GM (1, 1) basic
models including the even GM (1, 1) model, discrete GM (1, 1) model, even
difference GM (1, 1) model, and original difference GM (1, 1) model. Such models
were determined through simulation experiments, which helped to determine
suitable types of sequences for the different models (Liu 2015). The spectrum
diagram of grey prediction models is as follows (Fig. 2.2).
In the field of big data analysis, grey system prediction based on small data
mining is emerging as an effective tool to extract valuable information from masses
of data.

2.2.4 Grey Incidence Analysis Models

Grey incidence analysis models are used to assess whether different data sequences
are closely associated or not, according to the geometric shapes of their sequence
curves. Early grey incidence analysis models measured similarity based on prox-
imity. Examples include Deng’s grey incidence model, which is based on the point
incidence coefficients (Deng 1985), and Liu’s grey incidence model based on the
whole or global perspective (Liu and Guo 1991). Such models include absolute

Even differen al equa on's

Non-exponen al sequence,
solu on is me response formula
Even grey model fluctua ng sequence
Original difference equa on's
Homogenous exponen al sequence
solu on is me response formula
Original difference grey model Homogenous exponen al sequence
GM(1,1) Model Even difference equa on's
solu on is me response formula Approximate homogenous exponen al
(1 order equa on, Even difference grey model series, fluctua ng sequence
Single variable) Homogenous exponen al sequence
The discrete form of GM (1, 1)
Discrete grey model Approximate homogenous exponen al
series, fluctua ng sequence
Frac onal order cumula ve Sequence with memory and gene c
Frac onal order discrete
Basic idea: according to characteris cs
the informa on coverage Memoryless sequence
Self-memory principle
principle, sequence Memoryless grey model
operators are used to mine
the evalua on law of small
data. The sta c model, no deriva ve
GM (0, N) model
r order equa on, h variables System sequence influenced by
GM(r,h) model
mul factor
GM(1,N) Model The discrete form of GM (r, h) Mul variable discrete grey model
(1 order equa on,
mul variable) Variable-driving parameters Drive control mul variable Applicable scope
discrete grey model

Fig. 2.2 The spectrum diagram of grey prediction models

22 2 The Grey Systems Theory Framework

degree of grey incidence, relative degree of grey incidence and synthetic degree of
grey incidence, as follows:

1 þ jsi j þ jsj j
eij ¼ ð2:3Þ
1 þ jsi j þ jsj j þ jsi  sj j
0 0
1 þ jsi j þ jsj j
rij ¼ 0 0 0 0 ð2:4Þ
1 þ jsi j þ jsj j þ jsi  sj j

qij ¼ heij þ ð1  hÞrij ; h 2 ½0; 1 ð2:5Þ

In 2011, Sifeng Liu and Naiming Xie built a new grey incidence analysis model
based on the perspective of similarity and proximity, respectively, as follows (Xie
2011):

Zn
1
eij ¼ ; si  sj ¼ ðXi0  Xj0 Þdt ð2:6Þ
1 þ jsi  sj j
1

Zn
1
qij ¼ ; Si  Sj ¼ ðXi  Xj Þdt: ð2:7Þ
1 þ jSi  Sj j
1

Ke Zhang and colleagues proposed a two-dimensional grey incidence degree

model based on absolute incidence degree and double integral. With their new
model, the research object became the relationship between surface analysis and
curve analysis (Zhang and Liu 2010):

1 þ jsp j þ jsq j
epq ¼ ; ð2:8Þ
1 þ jsp j þ jsq jsp  sq j
ZZ ZZ ZZ
where sp ¼ Xp0 dxdy , sq ¼ Xq0 dxdy , sp  sq ¼ ðXp0  Xq0 Þdxdy .
The spectrum diagram of grey incidence analysis models is as follows (Fig. 2.3).

2.2.5 Grey Clustering Evaluation Models

In 1993, the fixed weight clustering model (Liu 1993) and grey clustering evalu-
ation model using end-point triangular possibility functions were proposed by
professor Sifeng Liu (Liu and Zhu 1993).
In 2011, Sifeng Liu and Naiming Xie improved the triangular possibility
function model proposed in 1993 and constructed the grey evaluation method based
2.2 The Thinking, Models and Framework … 23

Entropy degree of grey

incidence

Point degree of grey Deng’s grey incidence degree

incidence Weighted degree of grey
incidence
The basic idea of grey
incidence analysis: the
degree of similarity of the
geometric curves of the Absolute degree of grey Similitude degree of grey
available data sequences is incidence: reflects the incidence: reflects the
used to determine whether similarity degree of similarity of the shapes of
sequences sequences
or not their connec ons are
close.
SyntheƟc degree of grey Three-dimensional absolute
Grey incidence degree in incidence: describes rela vely degree of grey incidence:
completely the closeness measures the degree of the
the global perspec ve
between sequences behavior of matrix and matrix
sequence

RelaƟve degree of grey Closeness degree of grey

incidence: reflects the incidence: measures the
similarity degree of the rate spa al closeness of
of sequences change sequences

Fig. 2.3 The spectrum diagram of grey incidence analysis models

on center-point triangular possibility functions (Liu and Xie 2011). The new
method reduced the base of the triangular possibility function of class k, which
became the straight line joining the two center points of class k − 1 and class k + 1,
and replaced the straight line joining the left end-point of class k − 1 and the right
end-point of class k + 1. The multiple cross phenomenon existing in the original
triangular possibility function clustering model is avoided effectively, and the
clustering vector satisfied the requirement of normalization.
In 2014, Sifeng Liu and Zhigeng Fang set the weight function of grey class 1 to
the possibility function of lower measure, set the weight function of grey class s to
the possibility function of upper measure, and then put forward the grey clustering
evaluation model based on center-point and end-point mixed triangular possibility
functions (Liu and Fang 2014); see Figs. 2.4 and 2.5 for details. To extend the
value bound of each clustering index, the grey clustering evaluation model based on
mixed end-point triangular possibility function is suitable for situations in which all

y = f j1 ( x) y = f j2 ( x ) y = f j k ( x) y = f j s −1 ( x ) y = f js ( x )
1

o a1j λ j a 2j λ j2 a 3j λ 3j a 4j
1
a kj −1 λ jk −1a kj λ k a kj +1λ jk +1a kj + 2 a sj −2 λ js −2a sj −1λ js −1a sj λ js a sj +1 x
j

Fig. 2.4 End-point mixed possibility function

24 2 The Grey Systems Theory Framework

y
y = f j1 ( x ) y = f j 2 ( x ) y = f j k ( x) y = f j s −1 ( x) y = f j s ( x)
1

o a j λ j1 λ j 2 λ j 3 λ j k −1 λ j k λ j
k +1
λ j s −2 λ j s −1 λ j s bj x

Fig. 2.5 Center-point mixed possibility function

grey boundaries are clear, but the points belonging to each grey class are most
likely unknown. The grey clustering evaluation model based on mixed center-point
triangular possibility function is suitable for situations where the points belonging
to each grey class are most likely clear, but the grey boundaries are unknown. The
mixed triangular possibility functions are more suitable for solving problems of
clustering evaluation with poor information.
Grey clustering models based on mixed triangular possibility functions are
applicable to evaluation and classification of poor information objects, which have
broad application prospects.

2.2.6 Grey Decision-Making Models

In 2010, Sifeng Liu and colleagues put forward the multi-attribute intelligent grey
target decision model (Liu et al. 2010), based on Julong Deng’s (Deng 1986) grey
target model ideas.
As the basis of the new model, a grey target is defined as a region a decision
maker wants to reach, with an inside ideal point across multiple objectives. To
facilitate the uniform distance measure of a decision strategy to the pre-defined grey
target, four kinds of measure procedures are designed: the effect measures for
benefit type objectives and cost type objectives; the lower effect measure for
moderate type objectives; and the upper effect measure for moderate type objec-
tives. Such procedures are designed according to three types of decision objective
including benefit objective, cost objective, and non-monotonic objective with a
most preferred middle value. Then, a matrix of synthetic effect measures can be
easily obtained based on the uniform distance measure of a decision strategy to the
grey target over different objectives. Based on the obtained matrix information,
different decision strategies can be evaluated comprehensively The proposed
method has a clear physical meaning as missing target, hitting target, as well as
hitting performance.
Measurements with good properties satisfy the requirement of normalization and
dimensionless, and the greater the effect, the greater the value measure. In 2009,
2.2 The Thinking, Models and Framework … 25

Sifeng Liu first reported the initial idea of intelligent grey target and gained vali-
dation from Professor Deng Julong and other experts.
In 2014, Sifeng Liu and colleagues constructed a new decision model entailing a
two-stage grey comprehensive measure. The comparison between the maximum
components dki and dkj of decision coefficient vector di and dj may conflict with the
comparison between di and dj , which is decision paradox that has been solved (Liu
et al. 2014).
Indeed, in 2014, Sifeng Liu and colleagues defined a weight vector group of
kernel clustering and the weighted coefficient vector of kernel clustering for
decision-making. Then a novel two-stage decision model with weight vector group
of kernel clustering and the weighted coefficient vector of kernel clustering for
decision-making was put forward, and several functional weight vector group of
kernel clustering were developed. This method can effectively solve the decision
paradox and produce consistent results (Liu 2014).
Additionally, Dang Luo studied different types of grey decision models and
obtained a series of significant results (Luo and Wang 2012; Luo and Wang 2012).
Huan Guo and Xinping Xiao researched grey double layers and multi-objective
linear programming and also solved problems (Guo and Xiao 2014). Finally, Jie Cui
and colleagues developed the weighting formula evaluation value of each detection
stage based on the new information priority principle, which provided a new process
for solving grey decision problems consisting of multiple stages (Cui et al. 2012).

2.2.7 Combined Grey Models

Combined grey models include the Grey-Econometrics Combined Model (Liu and
Zhu 1996), Grey Cobb-Douglas Model (Liu et al. 2004), Grey DEA model (Wang
and Liu 2009), Grey Markov Model (Liu et al. 2014), Grey Rough Model (Jian
et al. 2011), among others. Zhigeng Fang and colleagues opened a new direction for
combined grey models based on grey game model. The researchers carried out
effective research on economic decision applications using grey game model (Fang
et al. 2006; Fang et al. 2010). In 2008, Qiao-Xing Li and Si-Feng Liu studied the
grey matrix and grey input-output models (Li and Liu 2008). Based on such
models, the authors put forward the Enterprise Grey Input-output analytical model
in 2012 (Li et al. 2012).

2.2.8 Grey Control Models

In the 1980s, Professor Julong Deng and Chaoshun Zhou (Deng and Zhou 1986;
Zhou and Deng 1986; Deng 1988) worked on grey forecasting control, and
addressed the stability of grey linear systems and sufficient conditions for the
stability of inter-connected dynamic systems. Furthermore, Chunhua Su and
26 2 The Grey Systems Theory Framework

colleagues studied the robust stability problem of grey stochastic time-delay sys-
tems, especially the distribution type, neutral type and neutral-distribution type
exponential robust stability problem of grey stochastic time-delay systems. In order
to do so, Chunhua Su and colleagues used several methods such as the Lyapuonv
function, Lyapunov-Krasovskii function, model transformation, combined Itô for-
mula, matrix inequality, Holder inequality, Schur complement, decomposition
technique of continuous grey matrix cover and other mathematical tools. As a
result, the researchers’ work provided effective criteria for robust stability, and
obtained several useful achievements (Su and Liu 2008, 2009).
It has been 31 years since the successful development of the first grey controller
in 1985. During the 1980s, the conditions to establish a more effective control
system seemed ripe, given the integration of a variety of grey control methods and
models. However, the promotion and establishment of a new control method is not
a short-term endeavor.

2.3 The New Framework and Main Components of Grey

Systems Theory

Since 1982, grey systems theory has matured and seen the development of a new,
generally accepted system structure. The new framework and main components of
GST are shown in Table 2.1.
The grey systems theory course offered at Nanjing University of Aeronautics and
Astronautics was selected as the national perfect curriculum and national excellent
resource sharing, and the book Grey System Theory and Its Application, 4th and
6th edition, has been selected as “Eleventh Five Years” and “Twelfth Five Years”
national programmes respectively.
A recent search through the China Knowledge Net (CNKI) using eight key terms
such as grey system, grey theory, grey model, GM (1, 1), grey incidence analysis,
grey clustering, grey prediction and grey decision making yielded 10,180 full text
PhD theses, and 48,185 Masters’ degree theses containing such terms. Also, a
search of PhD and Masters’ theses based on such search terms as key words
resulted in 2,873 and 13,463 thesis, respectively. Full text journals containing such
search terms amounted to 69,276 and journal papers based on those search terms as
key words totalled 39,544 papers. International publishers Springer-Verlag and
Taylor & Francis Group have launched a number of grey systems works in English,
and Science Press has officially approved a book series on Grey Systems; the first of
the 22 volumes is already on its way. Furthermore, the grey system modeling
software version 8.0 written by Professor Bo Zeng and Professor Yang Shen
contains applications of commonly used grey systems models. Interested readers
can go to the website of the Institute for Grey System Studies at Nanjing University
of Aeronautics and Astronautics (http://igss.nuaa.edu.cn/), to download the soft-
ware free of charge.
2.3 The New Framework and Main Components … 27

Table 2.1 The new GST framework

Main contents Detailed thinking, methods, and models of GST
Basic thinking Operations of grey Grey number, the algorithms of interval grey
and models number and grey algebra number, the kernel of grey number concept and
system general grey number, the operation axiom of
grey number, as well as the algebra system
based on grey “kernel” and degree of greyness
Sequence operator Average generation operator, accumulating
generation operator, inverse accumulating
generation operator, shock-disturbed system,
axiom system of buffer operator, series of
accumulating and inverse accumulating
generators
Grey incidence model Series grey incidence analysis model including
Deng’s grey incidence model, absolute degree of
incidence, relative degree of incidence, synthetic
degree of incidence, nearness degree of
incidence, similitude degree of incidence,
three-dimensional grey incidence degree, among
others
Grey clustering evaluation Variable weight grey clustering model, fixed
model weight grey clustering model, grey clustering
evaluation model based on mixed end-point, and
center-point triangular possibility function
Family model of GM and Even GM (1, 1) model, even difference GM
grey system forecasting (1, 1) model, original difference GM (1, 1)
model, discrete GM (1, 1) model, fractional
order grey model, memoryless grey model, grey
Verhulst model, multivariable discrete grey
model, discrete grey model with approximate
non-homogenous exponential law, sequence
grey forecasting, interval forecasting,
catastrophe forecasting, grey wave forecasting,
and system forecasting
Grey decision model Grey target decision, four kinds of uniform
effect measure function which are able to
characterize factors for positive point and
negative point, multi-attribute intelligent grey
target decision model, two-stage grey synthetic
measure decision model
Advanced Grey equation and grey Grey algebraic equation, grey differential
thinking and matrix equation, grey matrix, and matrix equation
models Combined grey model Grey-Econometrics Combined Model, grey
Cobb-Douglas Model, grey DEA model, grey
Markov Model, and grey rough model
Grey game model Grey matrix game model based on pure strategy,
grey matrix game model based on mixed
strategy, duopoly strategy output-making model
based on bounded knowledge and bounded
rationality, the paradox of centipede game model
(a new model of grey structured algorithm of
forwards induction)
(continued)
28 2 The Grey Systems Theory Framework

Table 2.1 (continued)

Main contents Detailed thinking, methods, and models of GST
Grey input and output P-F theorems of grey non-negative matrices,
regional input-output model, enterprise grey
input-output, grey input-output optimization
model
Grey programming Linear programming models with grey
parameters, grey linear programming of
prediction type, drift grey linear programming,
grey 0–1 programming, grey multiple objective
programming, grey non-linear programming
Grey control model Grey control model, controllability and
observability of grey systems, robust stability of
grey systems, grey linear time-delay systems,
grey stochastic linear time-delay systems

Although grey systems theory has been applied successfully to many contexts
and in many different countries, it is still in its infancy and, thus, has much scope for
further development and improvement. Colleagues who are interested in grey
systems research must welcome constructive criticism, and continuously explore
the potential to further advance grey systems theory.
Chapter 3
Grey Numbers and Their Operations

3.1 Grey Numbers

A grey system is described with grey numbers, grey sequences, grey equations, or
matrices. Here, grey numbers are the elementary “atoms” or “cells”, and their exact
values are unknown. In applications, a grey number stands for an indeterminate
number that takes its possible value within an interval or a general set of numbers.
A grey number is generally represented using the symbol “
.” There are several
types of grey numbers, as discussed below.
(1) Grey numbers with only a lower bound: This kind of grey number
is
represented as
2 ½a; 1Þ or
ðaÞ, where ðaÞ stands for the definite, known
lower bound of the grey number
. The interval ½a; 1Þ is referred to as the
field of
.
For example, the weight of a celestial body which is far away from the Earth is a
grey number containing only a lower bound, because the weight of the celestial
body must be greater than zero. However, the exact value of the weight cannot be
obtained through normal means. If we use the symbol
to represent the weight of
the celestial body, we then have that
2 ½0; 1Þ.
(2) Grey numbers with only an upper bound: This kind of grey number
is
written as
2 ð1;
a or
ð
aÞ, where
a stands for the definite, known upper
bound of
.
A grey number containing only an upper bound is a grey number with a negative
value, but its absolute value is infinitely great. For example, the opposite number of
the weight of the celestial body mentioned above is a grey number with only an
upper bound.
(3) Interval grey numbers: This kind of grey number
has both a lower bound a
and an upper bound
a, written
2 ½a;
a.

© Springer Science+Business Media Singapore 2017 29

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_3
30 3 Grey Numbers and Their Operations

For example, for an investment opportunity, there always exists an upper limit
representing the maximum amount of money that can be mobilized. For an elec-
trical equipment, there must be a maximum critical value for the equipment to
function normally. The critical value could be for a maximum voltage or for a
maximum amount of current allowed to be applied to the equipment. At the same
time, the values of investment, voltage, and current are all greater than zero.
Therefore, the amount of dollars that can be used for a specific investment
opportunity, and the voltage and the current requirements for the electrical equip-
ment are all examples of interval grey numbers.
(4) Continuous and discrete grey numbers: This kind of grey number takes only a
finite number or a countable number of potential values and is known as
discrete. If a grey number can potentially take any value within an interval,
then it is known as continuous.
For example, if a person’s age is between 30 and 35, his or her age could be one
of the values 30, 31, 32, 33, 34, 35. Thus, age is a discrete grey number. As for a
person’s height and weight, they are continuous grey numbers.
(5) Black and white numbers: Black numbers are represented as

2 ð1; þ 1Þ; that is, when
has neither an upper nor a lower bound,
then
is known as a black number. When
2 ½a;
a and a ¼
a,
is known
as a white number.
For the sake of parsimony, in our discussion we treat black and white numbers as
special grey numbers.
(6) Essential and non-essential grey numbers: The former stands for a grey number
that temporarily cannot be represented by a white number; the latter entails a
grey number that can be represented by a white number obtained either through
experience or through a certain method. The definite white number is referred
to as the whitenization (value) of the grey number, denoted
.~ Also, we use

ðaÞ to represent grey number(s) with a as its whitenization.
A grey number is an uncertain number with its value in a specific range. The
range can be regarded as a cover of the grey number. Therefore, an interval grey
number
2 ½a;
a; a\
a is very different from an interval number ½a;
a; a\
a. An
interval grey number
2 ½a;
a; a\
a is only one value in interval ½a;
a; a\
a.
However, an interval number ½a;
a; a\
a is the whole interval ½a;
a; a\
a.

3.2 The Whitenization of a Grey Number and Degree

of Greyness

When a type of grey number vibrates around a certain fixed value, the whitenization
of this kind of grey number is relatively easy. One can simply use that fixed value
as its whitenization. A grey number that vibrates around a can be written as
ðaÞ ¼
3.2 The Whitenization of a Grey Number and Degree of Greyness 31

a þ da or
ðaÞ 2 ð; a; þ Þ, where da stands for the vibration. In this case, the
~
whitenized value is
ðaÞ ¼ a:
For the general interval grey number
2 ½a; b, we can take its whitenization
value
~ as indicated in (3.1), based on the possible value information:

~ ¼ aa þ ð1  aÞb; a 2 ½0; 1

ð3:1Þ

Here, a is called the positioned coefficient of the interval grey number
2 ½a; b
(Liu 1989).
Definition 3.2.1 The whitenization of the form
~ ¼ aa þ ð1  aÞb; a 2 ½0; 1 is
called a whitenization with positioned coefficient a.
Definition 3.2.2 Mean whitenization occurs when a ¼ 12.
When the distribution of an interval grey number is unknown, mean whit-
enization is often employed.
Definition 3.2.3 Take the interval grey numbers
1 2 ½a; b,
2 2 ½a; b;

~ 1 ¼ a a þ ð1  aÞb; a 2 ½0; 1; and

~ 2 ¼ b a þ ð1  bÞb; b 2 ½0; 1:

If a ¼ b, we say that both
1 and
2 are synchronous. If a 6¼ b, we say that the

grey numbers
1 and
2 are non-synchronous. When two grey numbers
1 and
2
have the same value range in interval [a,b], it is only when they are synchronous
that it is possible to have
1 ¼
2 .
When the distribution of a grey number is known, mean whitenization is not
used. For instance, a certain person’s age is within the range of 30–45 years old.
Thus,
2 ½30; 45 is a grey number. It is also known that the person in question
finished their 12 years of pre-college education and entered college in the 1990s.
Hence, the chance of the person to be 38 years old or in the interval from 36 to
40 years of age in 2015 is quite good. For this grey number, it is not reasonable for
us to employ mean whitenization.
When the value information of a grey number is known to a certain extent, we
can use a possibility function to describe the possibility the grey number has of
taking its potential values.
The possibility function is different from the membership function in fuzzy
mathematics. The membership function describes the degree to which an object
belongs to a certain set. However, the possibility function describes the possibility
that a grey number can take a certain value, or the possibility that a certain value is
the true value of a grey number. The possibility function is similar to the density
function of probability distribution, but there are essential differences between the
two concepts. A grey number described by the possibility function is a number with
incomplete value information. Once a number with complete value information can
be treated as a random variable with a certain probability distribution, it is no longer
a grey number with poor value information:
32 3 Grey Numbers and Their Operations

Fig. 3.1 Different types of

possibility functions

For any conceptual type of grey number that represents wishes, its possibility
function generally increases monotonically. In Fig. 3.1, the possibility function f(x)
stands for, say, the grey number of a loan amount (in ten thousand dollars) and its
degree of preference. A straight line stands for the “normal desire,” that is, the
degree of preference is directly proportional to the amount of loan, with different
slopes representing different intensities of desire. In particular, f1 ðxÞ represents a
relatively mild intensity of desire, where a loan amount of $100,000 is not enough,
a loan in the amount of $200,000 will be more satisfying, and a loan of $300,000
will be quite adequate. f2 ðxÞ stands for a desire with more intensity, where a loan in
the amount of $350,000 is only about 40 % satisfactory. The curve of f3 ðxÞ means
that even for a loan in the amount of $400,000, the degree of satisfaction is only
about 20 %. To be satisfied, the loan amount has to be somewhere around
$800,000.
Generally speaking, the possibility function of a grey number is designed
according to what is known to the researcher. Therefore, it does not have a fixed
form. The start and end of the curve should have its significance. For instance, in a
trade negotiation, there is a process of changing from a grey state to a white state.
The eventual agreed upon deal will be somewhere between the ask and the bid.
Thus, the relevant possibility function should start at the level of the ask (or the bid)
and end at the level of bid (or the ask).
The typical possibility function is a continuous function with fixed starting and
ending points so that the left-hand side increases and the right-hand side decreases,
as seen in Fig. 3.2a, where:

Fig. 3.2 Typical possibility

3.2 The Whitenization of a Grey Number and Degree of Greyness 33

8
< LðxÞ; x 2 ½a1 ; b1 Þ
f1 ðxÞ ¼ 1; x 2 ½b1 ; b2 
:
RðxÞ; x 2 ðb2 ; a2 :

For the convenience of computer programming and computation, in practical

applications the left- and right-hand functions LðxÞ and RðxÞ are generally simpli-
fied into straight lines, as seen in Fig. 3.2b, where:
8
< LðxÞ ¼ x2 x11 ; x 2 ½x1 ; x2 Þ
xx

f2 ðxÞ ¼ 1; x 2 ½x2 ; x3 
: RðxÞ ¼ x4 x ; x 2 ðx ; x :
x4 x3 3 4

Definition 3.2.4 For the possibility function shown in Fig. 3.2a, the following
representation is referred to as the degree of greyness of
(Deng 1985):
 
 2j b1  b2 j j a1  b1 j j a2  b2 j
g ¼ þ max ; ð3:2Þ
b1 þ b2 b1 b2

The expression g is a sum of two parts. The first part represents the greyness of
the grey number as affected by the size of the peak area under the curve of the
possibility function, while the second part shows the effect of the size of the area
under the curves of LðxÞ and RðxÞ. Generally, the greater the peak
n area and theo area
ja1 b1 j ja2 b2 j
under LðxÞ and RðxÞ, the greater the value of g . When max b1 ; b2 ¼ 0,
g ¼ 2bjb1 1þbb22 j. In this case, the possibility function is a horizontal line. When
2jb1 b2 j
b1 þ b2 ¼ 0, grey number
is
a grey number with its basic value b ¼ b1 ¼ b2 .

When g = 0,
is a white number.

3.3 Degree of Greyness Defined by Axioms

Professor Julong Deng (1985) provided a definition of degree of greyness of a grey

number with a typical possibility function, as shown in Fig. 3.2a. However, in 1996
Sifeng Liu established an axiomatic definition of degree of greyness by using the
length l(
) of the grey number interval and its mean whitenization
^ (Liu 1996):

lð
Þ
g ð
Þ ¼ ð3:3Þ
^

Such a definition is valid on the basis of the postulates of non-negativity, zero

greyness, infinite greyness, and scalar multiplication. However, the concept of
greyness as defined in Eqs. (3.2) and (3.3) suffers from the following problems:
34 3 Grey Numbers and Their Operations

1) When the length l(
) of the grey interval approaches infinity, the degree of
greyness as defined in both (3.2) and (3.3) is likely to approach infinity.
2) Grey numbers centered at zero will not have greyness. In this case, in Eq. (3.2),
one has b1 ¼ b2 ¼ 0; and in Eq. (3.3), one faces
^ ¼ 0. That is, neither (3.2)
nor (3.3) is meaningful.
A grey number is a way to express the behavioral characteristics of a specific
grey system (Deng 1990). The greyness of grey numbers reflects the degree to
which the researcher understands the uncertainty involved in such numbers (Liu
1999; Chen 2001). Therefore, the magnitude of the greyness of a grey number
should be closely related to the background on which the grey number is created, or
to the field of discourse within which the said number becomes grey. If this
background, or field of discourse, and the characteristics of a grey system are not
detailed, there is no means through which to discuss the degree of greyness of a
given grey number. With this understanding in place, let X be the field of discourse
within which grey number
is created, and lð
Þ is the measure of the number
field from which
takes its value. Then, the degree of greyness g ð
Þ of grey
number
should satisfy the axioms below.
Axiom 3.3.1 0  g ð
Þ  1. That is, the degree of greyness of any grey number
has to be within the range of 0–1.
Axiom 3.3.2 Any
2 ½a;
a; a 
a; when a ¼ a, g ð
Þ ¼ 0. That is, white num-
bers contain no ambiguity, so their degree of greyness is 0.
Axiom 3.3.3 g ðXÞ ¼ 1. That is, because the background X within which grey
number
is created is generally known, it does not contain any useful information
leading to the greatest level of uncertainty.
Axiom 3.3.4 g ð
Þ is directly proportional to lð
Þ and inversely proportional to
lðXÞ.
Definition 3.3.1 The following equation is called the degree of greyness of grey
number
:

g ð
Þ ¼ lð
Þ=lðXÞ ð3:4Þ

X is the field of discourse of grey number
, and l is the measure of field X (Liu
et al. 2010).
Theorem 3.5.1. The degree of greyness of grey numbers satisfies the following
properties:
(1) If
1 
2 , then g ð
1 Þ  g ð
2 Þ.
(2) g ð
1 [
2 Þ  g ð
k Þ, k = 1, 2, where
1 [
2 ¼ fnjn 2 ½a; born 2 ½c; dg is
the union of grey numbers
1 2 ½a; b; a\b and
2 2 ½c; d; c\d.
(3) g ð
1 \
2 Þ  g ð
k Þ, k = 1, 2, where
1 \
2 ¼ fnjn 2 ½a; b and n 2
½c; dg is the interaction between grey numbers
1 2 ½a; b; a\b and

2 2 ½c; d; c\d.
3.3 Degree of Greyness Defined by Axioms 35

(4) If
1 
2 , then g ð
1 [
2 Þ ¼ g ð
2 Þ, g ð
1 \
2 Þ ¼ g ð
1 Þ.

(5) If lðXÞ ¼ 1 and the measures of
1 and
2 are independent of l, then
1 g ð
1 \
2 Þ ¼ g ð
1 Þ g ð
2 Þ; and
2 g ð
1 [
2 Þ ¼ g ð
1 Þ þ g ð
2 Þ  g ð
1 Þ g ð
2 Þ.

Proof All details for conclusions (1)–(4) are omitted. For (5 1 ), from lðXÞ ¼ 1
and the assumption that measures of
1 and
2 are independent of l, we have:

g ð
1 \
2 Þ ¼ lð
1 \
2 Þ ¼ lð
1 Þ lð
2 Þ ¼ g ð
1 Þ g ð
2 Þ:

Similarly, for 2 , we have:

g ð
1 [
2 Þ ¼lð
1 [
2 Þ ¼ lð
1 Þ þ lð
2 Þ  lð
1 Þ lð
2 Þ

¼ g ð
1 Þ þ g ð
2 Þ  g ð
1 Þ g ð
2 Þ: QED:

The way in which grey numbers are combined affects the degree of greyness and
the reliability of the resultant grey number. Generally, when grey numbers are
“unioned” together, the resultant degree of greyness and reliability of the new
information increase; when grey numbers are intersected together, the resultant
degree of greyness drops and the reliability of the combined information decreases.
When solving practical problems and processing a large amount of grey numbers, it
is advisable to combine the numbers at several different levels so that useful
information can be extracted at individual levels. Additionally, in the process of
combining grey numbers, “union” and “intersection” operations should be done at
individual and other levels in order to guarantee that the extracted information
satisfies pre-determined requirements in terms of reliability and degree of greyness.

3.4 The Operations of Interval Grey Numbers

In what follows, let us look at the operations of interval grey numbers. Given grey
numbers
1 2 ½a; b, a\b, and
2 2 ½c; d; c\d; let us use * to represent an
operation between
1 and
2 . If
3 ¼
1

2 , then
3 should also be an interval
grey number satisfying
3 2 ½e; f ; e\f ; and for any
~ 1 and
~ 2,
~1



~ 2 2 ½e; f . The operation rules of interval grey numbers are discussed below
(Deng 1985).
Rule 3.4.1 (Additive operation). Assume that
1 2 ½a; b, a\b;
2 2 ½c; d; c\d;
then the following equation is called the sum of
1 and
2 :

1 þ
2 2 ½a þ c; b þ d ð3:5Þ
36 3 Grey Numbers and Their Operations

Example 3.4.1 Assume that
1 2 ½3; 4,
2 2 ½5; 8, then
1 þ
2 2 ½8; 12.
Rule 3.4.2 (Additive inverse). Assume that
½a; b, a\b, then the additive inverse
of
is given by:


2 ½b; a ð3:6Þ

Example 3.4.2 Assume that
2 ½3; 4, then 
2 ½4; 3.

Rule 3.4.3 (Subtraction operation). Assume that
1 2 ½a; b, a < b;

2 2 ½c; d; c\d, then the following is called the deviation
1 minus
2 :

1 
2 ¼
1 þ ð
2 Þ 2 ½a  d; b  c ð3:7Þ

Example 3.4.3 Assume that
1 2 ½3; 4,
2 2 ½1; 2, then:

1 
2 2 ½3  2; 4  1 ¼ ½1; 3;
2 
1 2 ½1  4; 2  3 ¼ ½3; 1:

Rule 3.4.4 (Multiplication operation). Assume that
1 2 ½a; b, a\b;
2 2

½c; d; c\d then the following equation is called the product of
1 and
2 :

1
2 2 ½minfac; ad; bc; bdg; maxfac; ad; bc; bdg ð3:8Þ

Example 3.4.4 Assume that
1 2 ½3; 4;
2 2 ½5; 10, then:

1
2 2 ½minf15; 30; 20; 40g; maxf15; 30; 20; 40g ¼ ½15; 40:

Rule 3.4.5 (Reciprocal). Assume that
2 ½a; b, a\b, a 6¼ 0; b 6¼ 0, ab [ 0, then

the following equation is called the reciprocal of
:
 
1 1 1

2 ; ð3:9Þ
b a

Example 3.4.5 Assume that
2 ½2; 4, then
1 2 ½0:25; 0:5.
Rule 3.4.6 (Division). Assume that
1 2 ½a; b, a\b;
2 2 ½c; d; c\d; and
c 6¼ 0; d 6¼ 0; cd [ 0, then the following is called the quotient of
1 division by

2 :
    
a a b b a a b b

1 =
2 ¼
1 
1
2 2 min ; ; ; ; max ; ; ; ð3:10Þ
c d c d c d c d

Example 3.4.6 Assume that
1 2 ½3; 4;
2 2 ½5; 10, then:

3 3 4 4 3 3 4 4

1 =
2 2 ½minf ; ; ; g; maxf ; ; ; g ¼ ½0:3; 0:8:
5 10 5 10 5 10 5 10
3.4 The Operations of Interval Grey Numbers 37

Rule 3.4.7 (Scalar multiplication). Let
2 ½a; b, a\b, and k a positive real
number, then the following is called the product of scalar k with grey number
:

k
2 ½ka; kb ð3:11Þ

Example 3.4.7 Assume that
2 ½2; 4, and k=5, then 5 
2 ½10; 20.
Rule 3.4.8 (Power). Let
2 ½a; b, a\b, k k a positive real number, then the
following equation is called the kth power of the grey number
:

k 2 ½ak ; bk  ð3:12Þ

Example 3.4.8 Assume that
2 ½2; 4, and k = 5, then
5 2 ½32; 1024.

3.5 General Grey Numbers and Their Operations

3.5.1 Reduced Form of Interval Grey Numbers

As the basis of grey system theory, grey numbers, grey number operations and grey
algebraic systems have received much attention from grey system scholars over the
past years. In the 1980s, we put forward the concept of mean whitenization of grey
numbers (Liu 1989), and based on this concept we developed a new algebraic
system for grey numbers.
According to the standard definition of degree of greyness of grey numbers (Liu
1996, 2006; Yang et al. 2007, 2011), it is possible to address grey intervals after the
operation of grey numbers, with the help of the concept of degree of greyness.
In this section, a definition for grey “kernel” is put forward. The axioms for
operation of grey numbers and a grey algebraic system is built based on grey
“kernel” and the degree of greyness of grey numbers. Also, the properties of the
operation are discussed with regards to how the operation of grey numbers can be
transformed to the operation of real numbers. Thus, to a certain extent the problem
for setting up the operation of grey numbers and grey algebraic systems is solved.
Definition 3.5.1 (The “kernel” of grey number). (1) Suppose an interval grey
number
2 ½a;
a; a\
a. In case of a lack of distributing information of the values
of grey number
,
^ ¼ 12 ða þ
aÞ is called the “kernel” of grey number
.
(2) If a grey number
is a discrete number and ai 2 ½a;
aði ¼ 1; 2; nÞ are all
P
the possible values for grey number
, then
^ ¼ 1n ni¼1 ai is called the “kernel” of
grey number
.
(3) Suppose that grey number
2 ½a;
a; a\
a is a random grey number with
distribution information of its values. Then
^ ¼ Eð
Þ is called the “kernel” of
grey number
(Liu et al. 2010).
^ the “kernel” of grey number
, is the representation of grey numbers
,

,
which cannot be exchangeable in the course of transforming the operation of grey
38 3 Grey Numbers and Their Operations

numbers to operation of real numbers. In fact, the “kernel” of grey number
, as a

real number, can be completely operated by the operation of real numbers, such as
plus, minus, multiplication, division, power, extract, and so on. Also, it is rea-
sonable to take the operation results of the “kernels” as the “kernel” of operation
results of grey numbers.
Definition 3.5.2 Let
^ and g be the kernel and the degree of greyness of a grey
number
, respectively. Then
^ ðg Þ is called the reduced form of grey number
.
The reduced form
^ ðg Þ contains all the information of grey number

2 ½a;

a; a\

a.
Proposition 3.5.1 For interval grey numbers, there is an one-to-one correspon-
dence between the reduced form
^ ðg Þ and grey numbers
2 ½a;
a; a\
a.
In fact, for any chosen grey number
2 ½a;
a; a\
a, one can compute
^ ðg Þ
through both
^ and g . On the other hand, when
^ ðg Þ is given, one can deter-
mine the position of
from
. ^ Therefore, from the definition of degree of

greyness g , one can compute the measure of the grey number
and consequently
the upper and lower bounds
a and a, which provides detailed information for

2 ½a;

a; a\
a.
Example 3.5.1 Assume that the grey numbers
1 ¼ ½2; 1,
2 ¼ ½8; 18,
3 ¼
½2; 18 all on background X 2 ½2; 20. Take the length of grey interval as the
measure of grey numbers, and calculate the reduced forms of
1 ,
2 ,
3 .
Solution: The measures of X;
1 ;
2 ;
3 are lðXÞ ¼ 20  ð2Þ ¼ 22,
lð
1 Þ ¼ 1; lð
2 Þ ¼ 10; lð
3 Þ ¼ 20. Then we can get to the kernels and the
degree of greyness of
1 ,
2 ,
3 as follows:
^ 1 ¼ 1:5;


^ 2 ¼ 13;

^ 3 ¼ 8; g1 ð
1 Þ ¼ 0:045, g2 ð
2 Þ ¼ 0:45, g3 ð
3 Þ ¼ 0:91.
Therefore, we obtained:

1 ¼ 1:5ð0:045Þ ;
2 ¼ 13ð0:45Þ ;
3 ¼ 8ð0:91Þ :

3.5.2 General Grey Numbers and Their Reduced Form

Definition 3.5.3 (Basic element of grey number). Together, an interval grey

number and a white number are called the basic element of a grey number.
Definition 3.5.4 (General grey number). Let g 2 < be an unknown real number
within a union set of closed or open grey intervals, where:

g 2 [ ni¼1 ½ai ; ai  ð3:13Þ

If i = 1, 2, …, n, n is an integer and 0 < n < ∞, ai ; ai 2 < and

 
ai1  ai  ai  ai þ 1 , for any grey interval
i 2 ½ai ; ai   [ ni¼1 ai ; ai , then g is
3.5 General Grey Numbers and Their Operations 39

called a general grey number. g ¼ inf ai 2g ai and g þ ¼ supai 2g ai are called the
lower and upper limits of g (Liu et al. 2012).


 “kernel”
Definition 3.5.5 (The  of general grey number). (1) For a general grey
number g 2 [ i¼1 ai ; ai , the following is called the “kernel” of a general grey
n

number:

1X n
^g ¼ ^ai ð3:14Þ
n i¼1
 
(2) If the probability distribution of g 2 ai ; ai (i = 1, 2, …, n) is known, assume
 
that pi is the probability for g 2 ai ; ai (i = 1, 2, …, n), ^ai the “kernel” of grey
interval
i 2 ½ai ; ai , and the following conditions hold:

pi [ 0; i ¼ 1; 2; . . .; n; and
X n
pi ¼ 1:
i¼1

 
Then, the “kernel” ^g of general grey number g 2 [ ni¼1 ai ; ai can be defined
as follows:

X
n
^g ¼ pi ^ai ð3:15Þ
i¼1

Definition 3.5.6 (The degree of greyness of a general grey number).  Suppose

 that

the background which makes a general grey number g 2 [ i¼1 ai ; ai come into
n


i 2
being is X, l is the measure of X, and  ½ai ; ai , i = 1, 2, …, n are basic elements
of general grey number g 2 [ ni¼1 ai ; ai . Then the following is called the degree
 
of greyness of general grey number g 2 [ ni¼1 ai ; ai , also denoted as g for short
(Liu et al. 2012):

1X n
g ðg Þ ¼ ^ai lð
i Þ=lðXÞ ð3:16Þ
^g i¼1

Definition 3.5.7 (The reduced formof general  grey number). If g^ is the “kernel” of
a general grey number g 2 [ ni¼1 ai ; ai and g is the degree of greyness of this
general grey number, then, ^gðg Þ is called the reduced form of a general grey number.
The reduced form ^gðg Þ of a general grey number contains important information
 
regarding the values of general grey number g 2 [ ni¼1 ai ; ai . If all the ^ai and
lð
i Þ(i = 1, 2, …, n) are known, then the reduced form of grey number ^gðg Þ
contains all the  information regarding the values of general grey numbers
g 2 [ ni¼1 ai ; ai .
40 3 Grey Numbers and Their Operations

Example 3.5.2 Let us take a mixed general grey number g ¼
1 [
2 [ 2 [

4 [ 6, where
1 2 ½1; 3;
2 2 ½2; 4;
4 2 ½5; 9. Assume that the background or
field which makes general grey number g come into being is X=[0,32]. If we take
the length of the interval as the measure of these grey numbers, try and work out the
reduced forms of general grey number g .
Solution:
^ 1 ¼ 2;
^ 2 ¼ 3;
^ 4 ¼ 7, thus, the kernel of general grey number g is
as follows:

1 1
^ ^1 þ

g ¼ ð
^2 þ 2 þ

^ 4 þ 6Þ ¼ ð2 þ 3 þ 2 þ 7 þ 6Þ ¼ 4:
5 5

From that, lð
1 Þ ¼ 2; lð
2 Þ ¼ 2; lð
4 Þ ¼ 4; lð2Þ ¼ lð6Þ ¼ 0, we have:

1X 5
1
g ðg Þ ¼
^ i lð
i Þ=lðXÞ ¼ ð2  2 þ 3  2 þ 2  0 þ 7  4 þ 6  0Þ=32
^
g i¼1 4
0:297:

Therefore, the reduced forms of general grey number g is 4ð0:297Þ . When the
probability distribution of g is known, assume that:

p1 ¼ 0:1; p2 ¼ 0:2; p3 ¼ 0:3; p4 ¼ 0:3; p5 ¼ 0:1:

P
^ i ¼ ð0:1 2 þ 0:2 3 þ 0:3 2 þ 0:3 7 þ 0:1 6Þ ¼ 4:1.
g ¼ ni¼1 pi

Then: ^
Therefore, the reduced form of general grey number g is 4:1ð0:297Þ .

3.5.3 Synthesis of Degree of Greyness and Operations

of General Grey Numbers

Axiom 3.5.1 (The synthesis axiom of degree of greyness). When plus and minus
are operated on n general grey numbers of g  
1 , g2 , …, gn , then the degree of
greyness g of the operation results in g , which can be arrived at as follows:

1 X
n X
n
g ¼ Pn gi ^gi ¼ wi gi ð3:17Þ
i¼1 ^
gi i¼1 i¼1

^
where wi ¼ Pn i ; i ¼ 1; 2; ; n, are the weights of gi .
g
^
gi
i¼1
One can arrive at a conclusion as Proposition 3.5.1
Proposition 3.5.1 When plus and minus are operated on n general grey numbers
of g 
…, g
1 , g2 ,  , g is the degree of greyness of the operation result g ; if
n  
gm ¼ min gi , gM ¼ max gi , then:
 
1in 1in
3.5 General Grey Numbers and Their Operations 41

gm  g  gM ð3:18Þ

Axiom 3.5.2 (The unreduction axiom of degree of greyness). When division and
multiplication are operated on n general grey numbers, the degree of greyness g of
the operation result g is not less than gM , the maximum number of the degree of
greyness g1 ; g2 ; ; gn of n general grey numbers g  
1 , g2 , …, gn .

Usually, gM , the maximum number of the degree of greyness of n general grey
numbers is taken as the degree of greyness of the operation results.
One can arrive at a conclusion as Proposition 3.5.2 and Proposition 3.5.3.
Proposition 3.5.2 When division and multiplication are operated on n general
grey numbers with the same degree of greyness, then the degree of greyness of the
operation result holds the line.
Proposition 3.5.3 When division and multiplication are operated on a white
number and a general grey number, the degree of greyness of the result is equal to
the degree of greyness of the general grey number.
Suppose that g 
1 ; g2 are two general grey numbers; ^ g1 ; ^g2 are their kernels,
 
respectively, and g1 ; g2 are their degrees of greyness, respectively. Then, the fol-
lowing rules come into existence according to Axioms 3.5.1 and 3.5.2:

Rule 1 ^g1ðg1 Þ þ ^g2ðg2 Þ ¼ ð^g1 þ ^g2 Þðw1 g þ w2 g Þ ð3:19Þ

1 2

Rule 2  ^g1ðg1 Þ ¼ ð^g1 Þðg Þ ð3:20Þ

1

Rule 3 ^g1ðg1 Þ  ^g2ðg2 Þ ¼ ð^g1  ^g2 Þðw1 g þ w2 g Þ ð3:21Þ

1 2

Rule 4 ^g1ðg1 Þ  ^g2ðg2 Þ ¼ ð^g1  ^g2 Þðg _g Þ ð3:22Þ

1 2

Rule 5 If ^g1 6¼ 0; then 1=^g1ðg1 Þ ¼ ð1=^g1 Þðg Þ ð3:23Þ

1

Rule 6 If ^g2 6¼ 0; then ^g1ðg1 Þ  ^g2ðg2 Þ ¼ ð^g1  ^g2 Þðg _g Þ ð3:24Þ
1 2

Rule 7 If k is a real number, then k ^gðg1 Þ ¼ ðk ^gÞðg Þ ð3:25Þ

1

The operations of general grey numbers can be extended to cases where many
general grey numbers must be operated. In such cases, we can take the operation
results of the “kernels” as the “kernel” of operation results of general grey
numbers. We can then get the degree of greyness of the results according to axioms
1 or 2, and, thus, we can arrive at the reduced forms of the results.
Example 3.5.3 Take two mixed general grey numbers g 1 ¼
1 [
2 [ 2 [
4
[ 6 and g 2 ¼
6 [ 20 [
8 [
9 , where
1 2 ½1; 3;
2 2 ½2; 4;
4 2 ½5; 9,

6 2 ½12; 16;
8 2 ½11; 15;
9 2 ½15; 19. Assume that the background or field
which makes general grey number g 1 come into being is X = [0, 32], and the
42 3 Grey Numbers and Their Operations

background or field which makes general grey number g 2 come into being is
X = [10, 60]. Try and calculate the values of g3 ¼ g1 þ g
    
2 , g4 ¼ g 1  g2 ,
     
g5 ¼ g1  g2 , and g6 ¼ g1  g2 .
Solution: First, calculate the reduced forms of g 
1 and g2 . From Example 3.5.2, we

have g1 = 4ð0:297Þ . From that,
^ 6 ¼ 14;

^ 8 ¼ 13;
^ 9 ¼ 17, and lð
6 Þ ¼ 4;
lð
7 Þ ¼ 0; lð
8 Þ ¼ 4; lð
9 Þ ¼ 4, we have:

1 1
^ ^ 6 þ 20 þ

g2 ¼ ð
^8 þ

^ 9 Þ ¼ ð14 þ 20 þ 13 þ 17Þ ¼ 16; and
4 4

1X 4
1
g2 ðg Þ ¼ ^ i lð
i Þ=lðX2 Þ ¼ ð14  4 þ 20  0 þ 13  4 þ 17  4Þ=50


^
g2 i¼1 16
¼ 0:22:

Thus, the reduced form of general grey number g2 is 16ð0:22Þ . With the reduced
forms, as well as w1 ¼ 20 ¼ 0:2; w2 ¼ 20 ¼ 0:8, it is possible for us to get the
4 16

following results:

g  
3 ¼ g1 þ g2 ¼ ð^ g1 þ ^g2 Þðw1 g þ w2 g Þ ¼ ð4 þ 16Þð0:20:297 þ 0:80:22Þ ¼ 200:235
1 2
  
g4 ¼ g1  g2 ¼ ð^g1  ^g2 Þðg _g Þ ¼ ð4  16Þð0:20:297 þ 0:80:22Þ ð12Þ0:235
1 2
g  
5 ¼ g1  g2 ¼ ð^ g1  ^g2 Þðg _g Þ ¼ ð4  16Þð0:297_0:22Þ ¼ 640:297
1 2
g  
6 ¼ g1  g2 ¼ ð^ g1  ^g2 Þðg _g Þ ¼ ð4  16Þð0:297_0:22Þ ¼ ð14Þ0:297
1 2

Definition 3.5.8 Assume that Fðg Þ is a set of general grey numbers, and that
g           
i ; gj 2 Fðg Þ. If gi þ gj , gi  gj , gi gj ,and gi  gj all belong to Fðg Þ
(when division is considered, the conditions in rule 6 need to be satisfied), then
Fðg Þ is called a field of general grey numbers.
Theorem 3.5.1 The totality of all general grey numbers constitutes a field of
general grey numbers.
Definition 3.5.9 Assume that Rðg Þ is a set of general grey numbers. If for g 
i ; gj
and g 
k 2 Rðg Þ, the following hold true:

(1) g  
i þ gj = gj þ g i ;


(2) ðg    
i þ gj Þ þ gk = gi þ ðgj þ gk Þ;


(3) There exists a zero element 0 2 Rðg Þ, such that g 

i þ 0 ¼ gi ;
(4) For any gi 2 Rðg Þ, there exists a gi 2 Rðg Þ, such that gi + ðg
    
i Þ ¼ 0;
(5) ðg
i g 
j Þ g 
k = g 
i ðg 
j g 
k Þ;
(6) There exists a unit element 1 2 Rðg Þ, such that 1 g  
i ¼ gi 1 ¼ gi ;
3.5 General Grey Numbers and Their Operations 43

(7) ðg      
i þ gj Þ gk = gi gk þ gj gk ; and
(8) g      
i ðgj þ gk Þ ¼ gi gj þ gi gk .

Thus, Rðg Þ is called a linear space of general grey numbers.

Theorem 3.5.2 The totality of all synchronous general grey numbers constitutes a
linear space.
A grey number is the most elementary component of grey system theory and
forms the basis for studying the quantitative relations of a grey system. The
operation of grey numbers is the starting point for grey maths, and it has much
significance in the development of grey system theory. On the basis of intensifying
the effect and significance of the “kernel” of general grey numbers, and with the
degree of greyness of general grey numbers as a link, the operation of grey
numbers has been translated into the operation of real numbers. Therefore, to a
certain extent the problem of operation of grey numbers has been solved, and a
grey algebraic system based on this operation has been constructed. The operation
of grey numbers defined in this chapter can be extended to grey algebraic equa-
tions, grey differential equations and grey matrix operations. This is a development
of great significance to the study of grey input-output models and grey program-
ming, which have advanced slowly due to the restrictions imposed by previously
limited operations of general grey numbers.
The calculation of degree of greyness of general grey numbers relates to the
field X of general grey numbers. Thus, the field X must be considered in order to
translate the reduced form of general grey number to its common form.
Researchers tend to pay attention only to the operation of general grey numbers
and ignore the field of the results, which creates difficulties in reverting general
grey numbers. However, the reduced form of a general grey number provides
relevant information about the “kernel” and degree of greyness, which gives us
more confidence in the results. This is similar to the digital characteristics of a
random variable such as mean and variance, which hold the distribution infor-
mation of the random variable. The “kernel” and degree of greyness arising from
the reduced form are very important as they allow us to learn the value information
of a general grey number.
Chapter 4
Sequence Operators and Grey Data
Mining

4.1 Introduction

One of the main tasks of grey systems theory is to uncover the mathematical
relationships between different system variables and the laws of change of certain
system variables themselves based on the available data of characteristic behaviors
of social, economic and ecological systems, for example. Grey systems theory
looks at each stochastic variable as a grey quantity that varies within a fixed region
and within a certain time frame, and each stochastic process as a grey process.
When investigating the behavioral characteristics of a system, what is available
is often a sequence of definite white numbers. There is no substantial difference
between whether we treat the sequence as a trajectory or actualization of a
stochastic process, or as whitenized values of a grey process. However, to uncover
the laws of evolution of systems’ behavioral characteristics, different methods are
developed using different thinking logics. For instance, the theory of stochastic
investigates statistical laws on the basis of probabilities borrowed from prior
knowledge. This methodology generally requires large amounts of data. However,
even with large amounts of data there is no guarantee that any of the desired laws
can be successfully uncovered. That is because the number of basic forms of
distribution considered in this methodology is very limited. It is often extremely
difficult to deal with non-typical distribution processes. Nonetheless, grey systems
theory uncovers laws of change by excavating and organizing the available raw
data, representing an approach of finding data out of data through grey sequence
operators. Grey systems theory believes that a system possesses overall functions
and properties, even if the expression of such an objective system might be com-
plicated, and its data chaotic. Therefore, there must be internal laws governing the
existence of the system and its operation. The key is to choose an appropriate
method to excavate the internal laws and make use of such laws. For any given grey
sequence, its implicit pattern can always be revealed through weakening the explicit
randomness.

© Springer Science+Business Media Singapore 2017 45

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_4
46 4 Sequence Operators and Grey Data Mining

Fig. 4.1 The curve of X(0) X 10 thousands

15

10

2 4 6

For example, the following sequence does not clearly show any regularity or
pattern:


X ð0Þ ¼ ð1; 2; 1:5; 3Þ ¼ xð0Þ ð1Þ; xð0Þ ð2Þ; xð0Þ ð3Þ; xð0Þ ð4Þ :

Now, we depict the data set with the graph in Fig. 4.1. From this graph, it can be
seen that the curve of X(0) undulates with relatively large amplitude. If we apply the
accumulating operator once to the original data set X(0), and denote the resultant
sequence as X(1), then we have:


X ð1Þ ¼ ð1; 3; 4:5; 7:5Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ; xð1Þ ð3Þ; xð1Þ ð4Þ :

P
where for k = 1, 2, 3, 4, xð1Þ ðkÞ ¼ ki¼1 xð0Þ ðiÞ:
Now, the processed sequence X(1) clearly shows a growing tendency (see
Fig. 4.2 for more details).

Fig. 4.2 The curve of X(1) X 10 thousands

2 4 6
4.2 Systems Under Shocking Disturbances and Buffer Operators 47

4.2 Systems Under Shocking Disturbances and Buffer

Operators

4.2.1 The Trap for Shocking Disturbed System Forecasting

Behavioral prediction of problems under the influence of shocking disturbances has

always been a difficult problem. For such predictions, any theory on how to choose
models would lose its validity. This is because the problem to be address here is not
about which model is the best; instead, when a system is severely impacted by
shocks, the available behavioral data of the past no long represent the current state
of the system. In this case, the available data of the system’s behavior can no longer
truthfully reflect the law of change of the system.
Definition 4.2.1 Assume that


X ð0Þ ¼ xð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞ

stands for a sequence of a system’s true behaviors. If the observed behaviors of the
system are

X ¼ ðxð1Þ ; xð2Þ; . . .; xðnÞÞ


¼ xð0Þ ð1Þ þ e1 ; xð0Þ ð2Þ þ e2 ; . . .; xð0Þ ðnÞ þ en ¼ X ð0Þ þ e

where e = (e1, e2, …, en) is a term for the shocking disturbance, then X is called a
shock-disturbed sequence (Liu 1991).
To correctly uncover and recognize the true behavior sequence X(0) of the system
from the shock-disturbed sequence X, one first has to go over the hurdle e. If we
directly established our model and made our predictions using the severely affected
data X without first cleaning up the disturbance, then our predictions would most
likely fail. This is because the model would not have described the true state X(0) of
change of the underlying system.
The wide spread existence of severely shocked systems often causes quantitative
predictions to disagree with the outcomes of intuitive qualitative analyses. Hence,
there is a need to seek an organic equilibrium between quantitative predictions and
qualitative analyses, by eliminating shock wave disturbances in order to recover the
true state of the systems’ behavioral data. This way the accuracy of the consequent
predictions can be greatly improved, which is one of the most important tasks
performed by grey systems scientists. To this end, the discussion in this section is
centered around the overall goal of reaching X(0) from X.
48 4 Sequence Operators and Grey Data Mining

4.2.2 Axioms that Define Buffer Operators

Definition 4.2.2 Assume that X = (x(1), x(2), …, x(n)) is a system’s behavior data
sequence.
(1) If 8 k = 2, 3, …, n, x(k) − x(k − 1) > 0, then X is referred to as a monotonic
increasing sequence;
(2) If the inequality sign in (1) is inversed, then X is referred to as a monotonic
decreasing sequence;
(3) If there are k, k0 2 {2, 3, …, n} such that xðk Þ  xðk  1Þ [ 0; xðk0 Þ 
xðk0  1Þ\0; then X is referred to as a random vibrating or fluctuating
sequence. If M = maxfxðkÞjk ¼ 1; 2;   ; ng and m = minfxðkÞjk ¼ 1; 2;   ;
n:g; then M − m is referred to as the amplitude of sequence X.

Definition 4.2.3 Assume that X is a data sequence of a system’s behavior, D an

operator to work on X, and after being applied by the operator D, X becomes the
following sequence:

XD ¼ ðxð1Þd; xð2Þd;   ; xðnÞd Þ

where D is referred to as a sequence operator and XD the first order sequence of

operator D (Liu 1991). If D1, D2, and D3 are all sequence operators, then D1D2 is
referred to as a second order sequence operator, and

XD1 D2 ¼ ðxð1Þd1 d2 ; xð2Þd1 d2 ;   ; xðnÞd1 d2 Þ

a second order sequence of D1D2. Similarly, D1D2D3 is referred to as a third order

sequence operator and

XD1 D2 D3 ¼ ðxð1Þd1 d2 d3 ; xð2Þd1 d2 d3 ;   ; xðnÞd1 d2 d3 Þ

a third order sequence of D1D2D3.

Axiom 4.2.1 (Fixed Point) Assume that X is a data sequence of a system’s behavior
and D a sequence operator. Then D satisfies x(n)d = x(n) (Liu 1991).
This fixed point axiom means that under the effect of a sequence operator, data
point x(n) remains unchanged, and this is the last entry of the system’s behavior
data sequence. Based on the conclusions of relevant qualitative analysis, we can
also leave several of the last entries of the data unchanged by the operator D, say,

xð jÞd 6¼ xð jÞ and xðiÞd ¼ xðiÞ

for j = 1, 2, …, k − 1; i = k, k + 1, …, n.
Axiom 4.2.3 (In accordance with information) The sequence operator must be
defined in accordance with information in the data sequence X. That is, each entry
4.2 Systems Under Shocking Disturbances and Buffer Operators 49

value x(k), k = 1, 2, …, n, in the data sequence X of the system’s behavior should

sufficiently participate in the entire process of application of the operator (Liu 1991).
This axiom requires that any sequence operator be defined by using known
information of the given sequence. It cannot be produced without referencing
available raw data.
Axiom 4.2.3 (Expressed normality) Each x(k)d, k = 1, 2, …, n, is expressed by a
uniform, elementary analytic representation of x(1), x(2),…, x(n) (Liu 1991).
This last axiom requires that the procedure of applying sequence operators be
clear, normalized, and uniform, so that it can be conveniently carried out on
computers.
Definition 4.2.4 Any sequence operator satisfying these three axioms is referred to
as a buffer operator; the first order, second order, third order, …, sequences
obtained by applying a buffer operator are referred to as the first order, second
order, third order, …, buffered sequences.
Definition 4.2.5 For a raw data sequence X and a buffer operator D, when X is
respectively an increasing, decreasing, or fluctuating sequence:
(1) If the buffered sequence XD increases, decreases, or fluctuates slower or with
smaller amplitude, respectively, than the original sequence X, then D is
referred to as a weakening operator.
(2) If the buffered operator XD increases, decreases, or fluctuates faster or with
larger amplitude, respectively, than the original sequence X, then D is referred
to as a strengthening operator (Liu 1991).

4.2.3 Properties of Buffer Operators

Theorem 4.2.1 Assume that X is a monotonic increasing sequence then:

(1) If D is a weakening operator , x(k)d  x(k), k = 1, 2, …, n;
(2) If D is a strengthening operator , x(k)d  x(k), k = 1, 2, …, n (Liu 1991).

Proof Assume that

xðnÞ  xðkÞ
rðkÞ ¼ ; k ¼ 1; 2; 3;   
n  kþ1

is the average increasing rate from xðkÞ to xðnÞ in the sequence X of raw data, and

xðnÞd  xðkÞd
rðkÞd ¼ ; k ¼ 1; 2; 3;   
n  kþ1
50 4 Sequence Operators and Grey Data Mining

is the average increasing rate from xðkÞd to xðnÞd in the buffered sequence XD.
Given the condition that

xðnÞd ¼ xðnÞ

It follows that

½xðnÞ  xðkÞ  ½xðnÞd  xðkÞd xðkÞd  xðkÞ

rðkÞ  rðkÞd ¼ ¼
n  kþ1 n  kþ1

If D is a weakening operator, then, rðkÞ  rðkÞd; that is rðkÞ  rðkÞd  0:

Therefore xðkÞd  xðkÞ  0; that is, x(k)d  x(k) and vice versa.
If D is a strengthening operator, then rðkÞ  rðkÞd; that is rðkÞ  rðkÞd  0:
Therefore xðkÞd  xðkÞ  0; that is, xðkÞ  xðkÞd and vice versa.
Theorem 4.2.2 Assume that X is a monotonic decreasing sequence then:
(1) If D is a weakening operator , x(k)d  x(k), k = 1, 2, …, n;
(2) If D is a strengthening operator , x(k)d  x(k), k = 1, 2, …, n (Liu 1991).

Theorem 4.2.3 Assume that X is a fluctuating sequence and XD a buffered

sequence, then:
(1) If D is a weakening operator, then max1  k  n fxðk Þg  max1  k  n fxðkÞd g
and min1  k  n fxðkÞg  min1  k  n fxðkÞd g;
(2) If D is a strengthening operator, then max1  k  n fxðk Þg  max1  k  n fxðkÞd g
and min fxðkÞg  min fxðkÞd g:
1kn 1kn

For detailed proofs and relevant discussions of these theorems, please consult
Liu and Lin (2006, pp. 64–67). What theorem implies is that each monotonic
increasing sequence expands under the effect of a weakening operator and shrinks
under a strengthening operator. What theorem indicates is that each monotonic
decreasing sequence shrinks under the effect of a weakening operator and expands
under a strengthening operator.

4.3 Construction of Practically Useful Buffer Operators

4.3.1 Weakening Buffer Operators

Theorem 4.3.1 Given a raw data sequence X ¼ ðxð1Þ; xð2Þ;   ; xðnÞÞ; let XD ¼
ðxð1Þd; xð2Þd;   ; xðnÞd Þ; where
4.3 Construction of Practically Useful Buffer Operators 51

1
xðk Þd ¼ ½xðk Þ þ xðk þ 1Þ þ    þ xðnÞ; k ¼ 1; 2;   ; n ð4:1Þ
n  kþ1

Then D is always a weakening operator regardless of whether X is a monotonic

increasing, decreasing, or vibrating sequence.
This operator is referred to as an average weakening buffer operator (AWBO)
(Liu 1991).
The weakening operator D in Theorem 4.3.1 possesses some very good prop-
erties and has been applied widely in modeling and prediction of systems with
interference of uncontrollable shock waves.
Corollary 4.3.1 For the weakening operator D as defined in Theorem 4.3.1, let:

XD2 ¼ XDD ¼ ðxð1Þd 2 ; xð2Þd 2 ;   ; xðnÞd 2 Þ

1
xðkÞd 2 ¼ ½xðkÞd þ xðk þ 1Þd þ    þ xðnÞd ; k ¼ 1; 2;   ; n: ð4:2Þ
n  kþ1

Then D2 is always a second-order weakening operator for monotonic increasing,

monotonic decreasing, and fluctuating sequences.
Example 4.3.1 Let X = (36.5, 54.3, 80.1, 109.8, 143.2) and D and D2 as defined in
Theorem 4.3.1 and Corollary 4.3.1 respectively, calculate the buffered sequence XD
and XD2.
Solution: Here n ¼ 5; from formula 4.1, we have:

1 1
xð1Þd ¼ ½xðkÞ þ xðk þ 1Þ þ    þ xðnÞ ¼ ½xð1Þ þ xð2Þ þ    þ xð5Þ
n  kþ1 5  1þ1
1
¼ ½36:5 þ 54:3 þ 80:1 þ 109:8 þ 143:2 ¼ 84:78
5  1þ1

1 1
xð2Þd ¼ ½xðkÞ þ xðk þ 1Þ þ    þ xðnÞ ¼ ½xð2Þ þ    þ xð5Þ
n  kþ1 5  2þ1
1
¼ ½54:3 þ 80:1 þ 109:8 þ 143:2 ¼ 96:85
4

1 1
xð3Þd ¼ ½xð3Þ þ xð4Þ þ xð5Þ ¼ ½80:1 þ 109:8 þ 143:2 ¼ 111:03
5  3þ1 3
1 1
xð4Þd ¼ ½xð4Þ þ xð5Þ ¼ ½109:8 þ 143:2 ¼ 126:5
5  4þ1 2

xð5Þd ¼ 143:2
52 4 Sequence Operators and Grey Data Mining

Therefore:

XD ¼ ð84:78; 96:85; 111:03; 126:5; 143:2Þ:

Similarly, we can obtained the second-order buffered sequence XD2 as follows:

XD2 ¼ ð112:47; 119:4; 126:91; 134:85; 143:2Þ:

Theorem 4.3.2 Assume that X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ is a sequence of raw data,
x ¼ ðx1 ; x2 ;    ; xn Þ is a weight vector, and xi [ 0; i ¼ 1; 2; ;    ; n: Let:

XD ¼ ðxð1Þd; xð2Þd;    ; xðnÞdÞ

where

xk xðkÞ þ xk þ 1 xðk þ 1Þ þ    þ xn xðnÞ 1 X n

xðkÞd ¼ ¼P xi xðiÞ; ðk ¼ 1; 2;    ; nÞ:
xk þ x k þ 1 þ    þ x n n
xi i¼k
i¼k

ð4:2Þ

Then D is always a weakening operator regardless of whether X is a monotonic

increasing, decreasing, or vibrating sequence (Dang et al. 2004). This operator D is
called as a weighted average (or mean) weakening buffer operator ðWAWBOÞ.
Corollary 4.3.2 For the weighted average weakening operator D as defined in
Theorem 4.3.2, let:

x ¼ ð1; 1;    ; 1Þ:

Then:

1 X
n
1 Xn
P xi xðiÞ ¼ xðiÞ:
n
n  kþ1 i¼k
xi i¼k
i¼k

That is, the average weakening buffer operator (AWBO) is a special case of the
weighted average weakening buffer operator ðWAWBOÞ:
Theorem 4.3.3 Assume that X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ is a sequence of raw data,
x ¼ ðx1 ; x2 ;    ; xn Þ is a weight vector, and xi [ 0; i ¼ 1; 2; ;    ; n: Let:

XD ¼ ðxð1Þd; xð2Þd;    ; xðnÞdÞ

4.3 Construction of Practically Useful Buffer Operators 53

where

" #P
n
1

1 Y
n xi
xðkÞd ¼½xðkÞxk  xðk þ 1Þxk þ 1    xðnÞxn  xk þ xk þ 1 þ  þ xn
¼ xðiÞxi i¼k ; ð4:3Þ
i¼k

ðk ¼ 1; 2;    ; nÞ:

Then D is always a weakening operator, regardless of whether X is a monotonic

increasing, decreasing, or vibrating sequence (Dang et al. 2004).
This operator D is called as a weighted geometric average weakening buffer
operator ðWGAWBOÞ:
Example 4.3.1 From 1983 to 1986, the overall business revenue of private enter-
prises in Changge county, located in the Henan Province of The People’s Republic
of China, was recorded as:

X ¼ ð10;155; 12; 588; 23; 480; 35; 388Þ:

This showed a tendency of rapid growth. The average rate of revenue growth for
these years was 51.6 %, and the average rate of revenue growth from 1984 to 1986
was 67.7 %. The people involved in the economic planning of the county, including
politicians, scholars, policy makers, and residents, commonly believed that the
overall revenue of private enterprises in this county would not be able to keep up
with this record speed of growth in the coming years. If relevant data had been used
to build models and make predictions, nobody would have accepted the resultant
conclusions. After numerous rigorous analyses and discussions, all parties involved
recognized that the reason for such a high growth rate between 1983 and 1986 was
mainly a low baseline. Such a low baseline had been a consequence of the fact that,
in the past, policies relevant to private enterprises had been neither existent, nor
encouraged. To weaken the growth rate of the sequence of the raw data, it was
necessary to artificially add all favorable environmental factors to past years’ data,
and such environmental factors were created based on the introduction of relevant
policies for the development of private enterprise in recent years. With this goal in
mind, we introduced the second-order weakening operator, as defined in
Theorem 4.3.1, and obtained the following second-order buffered sequence:

XD2 ¼ ð27,260; 29,547; 32,411; 35,388Þ:

As a result, the consequent modeling based on XD2 produced credible predic-

tions for the county’s business revenue of private enterprises growth between 1987
and 2000.
54 4 Sequence Operators and Grey Data Mining

4.3.2 Strengthening Buffer Operators

Theorem 4.3.4 Assume that X ¼ ðxð1Þ; xð2Þ;   ; xðnÞÞ is a sequence of raw data,
and XD ¼ ðxð1Þd; xð2Þd;   ; xðnÞd Þ; where D is defined as follows:

xð1Þ þ xð2Þ þ    þ xðk  1Þ þ kxðk Þ

xðkÞd ¼ ; k ¼ 1; 2;   ; n  1 ð4:4Þ
2k  1

If xðnÞd ¼ xðnÞ; then D is a strengthening buffer operator regardless of whether

the raw data sequence X is a monotonic, increasing or decreasing sequence (Liu
1991).
Theorem 4.3.5 Assume that X ¼ ðxð1Þ; xð2Þ;   ; xðnÞÞ is a sequence of raw data,
and Di is a sequence operator defined by:

x ð k  1Þ þ x ð k Þ
xðkÞdi ¼ ; k ¼ 2; 3;   ; n; i ¼ 1; 2: ð4:5Þ
2

If xð1Þd1 ¼ axð1Þ; a 2 ½0; 1; xð1Þd2 ¼ ð1 þ aÞxð1Þ; a 2 ½0; 1; and xðnÞdi ¼
xðnÞ; i ¼ 1; 2; then D1 is a strengthening buffer operator for monotonic increasing
sequences and D2 a weakening buffer operator for monotonic decreasing sequences
(Liu 1991).
Both D1 and D2 are called even strengthening buffer operators (ESBO).
Theorem 4.3.6 For a given increasing or decreasing sequence X of raw data, the
operator D is defined as follows:

½xðkÞ þ xðk þ 1Þ þ    xðnÞ=ðn  k þ 1Þ

xðkÞd ¼  xðkÞ; k ¼ 1; 2;    n: ð4:6Þ
xðnÞ

D is a strengthening buffer operator (Xie and Liu 2006), and is called average
strengthening buffer operator (ASBO).
Theorem 4.3.7 Assume that X ¼ ðxð1Þ; xð2Þ;   ; xðnÞÞ is a sequence of raw data,
x ¼ ðx1 ; x2 ;    ; xn Þ is a weight vector, and xi [ 0; i ¼ 1; 2; ;    ; n: Let XD ¼
ðxð1Þd; xð2Þd;   ; xðnÞd Þ; where D is defined as follows:
Pn
ðxk þ xk þ 1 þ    þ xn ÞðxðkÞÞ2 i¼k xi ðxðkÞÞ
2
xðkÞd ¼ ¼ P n ; ðk ¼ 1; 2;    ; nÞ:
xk xðkÞ þ xk þ 1 xðk þ 1Þ þ    þ xn xðnÞ i¼k xi xðiÞ
ð4:7Þ

D is a strengthening buffer operator regardless of whether the raw data sequence

X is a monotonic increasing, decreasing, or vibrating sequence (Dang et al. 2005).
D is called a weighted average strengthening buffer operator (WASBO).
4.3 Construction of Practically Useful Buffer Operators 55

4.3.3 The General Form of Buffer Operator

Theorem 4.3.8 Assume that X ¼ ðxð1Þ; xð2Þ;   ; xðnÞÞ is a sequence of raw data,
x ¼ ðx1 ; x2 ;    ; xn Þ is a weight vector, and xi [ 0; i ¼ 1; 2; ;    ; n: Let XD ¼
ðxð1Þd; xð2Þd;   ; xðnÞd Þ; where D is defined as follows:

xk xðkÞ þ xk þ 1 xðk þ 1Þ þ    þ xn xðnÞ a

xðkÞd ¼ xðkÞ  ½xðkÞ= 
xk þ xk þ 1 þ    þ xn
1 Xn
¼ xðkÞ  ½xðkÞ= Pn xi xðiÞa : ð4:8Þ
i¼k xi i¼k

Then:
(1) When a\0; D is a weakening operator regardless of whether X is a mono-
tonic increasing or decreasing sequence.
(2) When a [ 0; D is a strengthening buffer operator regardless of whether the
raw data sequence X is a monotonic increasing or decreasing sequence.
(3) When a ¼ 0; D is an identical operator (Wei and Kong, 2010).
D is called the general form of buffer operator (GFBO).
Corollary 4.3.3 Take a ¼ 1 in Theorem 4.3.8, then formula (4.8) changes to
(4.2). That is, the weighted average weakening buffer operator (WAWBO) is a
special case of the general form of buffer operator (GFBO).
Corollary 4.3.4 Take a ¼ 1 in Theorem 4.3.8, then formula (4.8) changes to (4.7).
That is, the weighted average strengthening buffer operator (WASBO) is a special
case of the general form of buffer operator (GFBO).
The buffer operator concept has been employed not only in grey systems
modeling but also in other kinds of model building. Generally, before building a
mathematical model based on qualitative analysis and its conclusions, one applies a
buffer operator on the original data sequence. This is done to soften or eliminate the
effects of shock-disturbances on the behavior sequence of a given system. By doing
so, expected results are often obtained.
Example 4.3.2 From 1996 to 1999, the annual gross revenues produced by the
agricultural, forestry, animal husbandry, and fishery sectors in the area of Nanjing
were (in 0.1 billion yuan):

X ¼ ð91:9895; 94:2439; 96:9644; 98:9199Þ:

The growth rate shown in X is very slow, as it represents an average of about

2.4 % annually. Such a slow growth rate was not aligned with the fast advances of
the overall annual economic development of the area. If such a slow growth con-
tinued in these economic sectors, it would have caused imbalances in the develop-
ment of the overall economic structure of the region and sustained regional economic
56 4 Sequence Operators and Grey Data Mining

growth would have been adversely affected. In 2000, Nanjing City gradually
adjusted the economic structure of the countryside to counteract slow economic
growth. In order to accurately control that economic development tendency in a
timely fashion, there was a need to produce scientifically reasonable economic
forecasts. To achieve this goal we had to address available data where slow growth
was recorded. This would allow the resultant predictions to possess practical value in
the realm of economic forecast and pro-growth government intervention. By
applying the strengthening operator in Theorem 4.3.6 twice on the available data
sequence, we obtained the following second order buffered data sequence:

XD2 ¼ ð79:5513; 85:5446; 93:1686; 98:9199Þ:

A GM(1, 1) model (for details, see Liu and Lin (2006), or Sect. 4.1 in this book)
based on this buffered sequence provided:

dX ð1Þ
 0:0720X ð1Þ ¼ 77:1389:
dt

The time response function was as follows:

^
X ð1Þ ðk þ 1Þ ¼ 1150:7003e0:0720k  1071:1503:

Based on this equation, the computational simulation results, effectiveness of the

data fit, and prediction efficacy are given in Tables 4.1 and 4.2.
Tables 4.1 and 4.2 show that by employing the buffered data using a
strengthening operator to establish our model, the simulated results and corre-
sponding predictions are quite good. In particular, for 2000 and 2001, predicted
values reached an accuracy rate of over 98 % compared to the actual data for those
years.

Table 4.1 The effectiveness of the simulation results

Year Strengthened Simulated data Error Relative error
data xð0Þ ðkÞ ^xð0Þ ðkÞ eðk Þ ¼ ^xð0Þ ðkÞ  xð0Þ ðkÞ Dk ¼ xjð0Þ
eðk Þj
ðkÞ

1997 85.5446 85.9245 0.3799 0.4441 %

1998 93.1686 92.3407 −0.8279 0.8886 %
1999 98.9199 99.2359 0.316 0.3195 %

Table 4.2 The efficacy of the predictions

Year Actual data Predictions Error Relative error
^xð0Þ ðkÞ ^xð0Þ ðkÞ eðk Þ ¼ ^xð0Þ ðkÞ  xð0Þ ðkÞ Dk ¼ xjð0Þ
eðk Þj
ðkÞ

2000 106.3412 106.6460 0.3048 0.2866 %

2001 113.29 114.6094 1.3194 1.1646 %
2005 152.8703
4.4 Average Operator 57

4.4 Average Operator

Due to various obstacles that are difficult to overcome, available data sequences
may or may not contain missing entries. Nevertheless, even if data sequences are
complete without any missing entries, systems’ behaviors can change suddenly at
any point in time, and corresponding entries in data sequences can become out of
the ordinary. This can create great difficulties for the researcher. For example, if
abnormal entries are removed, blank entries are created. Hence, how to effectively
fill blanks in data sequences naturally becomes one of the first questions one has to
address when processing available data. Data generation using averages is another
frequently used method to create new data, fill a vacant entry in the available data
sequence, and construct new sequences.
Assume that X = (x(1), x(2), …, x(k), x(k + 1), …, x(n)) is a sequence of raw
data. Then, entry x(k) is referred to the preceding value and x(k + 1) the succeeding
value. If x(n) stands for a piece of new information, then for any k  n − 1,
x(k) will be seen as a piece of old information. If the sequence X has a blank entry at
location k, denoted ∅(k), then the entries x(k − 1) and x(k + 1) will be referred to as
∅(k)’s boundary values, with x(k − 1) being the preceding boundary and x(k + 1)
the succeeding boundary. If a value x(k) at the location of ∅(k) is generated on the
basis of x(k − 1) and x(k + 1), then the established value x(k) is referred to as an
internal point of the interval [x(k − 1), x(k + 1)].
Definition 4.4.1 Assume that x(k − 1) and x(k + 1) are two entries in a data
sequence X. If x(k − 1) stands for a piece of old information and x(k + 1) a piece of
new information, the sequence operator D is defined as:

xðkÞd ¼ x ðkÞ ¼ axðk þ 1Þ þ ð1  aÞxðk  1Þ; for a 2 ½0; 1: ð4:9Þ

D is called a non-adjacent neighbor generating operator. The new value x*(k) is

referred to as generated by the new and old information under the generation
coefficient (weight) a. When a > 0.5, the generation of x*(k) is seen with more
weight placed on the new information than the old information. When a < 0.5, the
generation of x*(k) is seen with more weight placed on the old information than the
new information. If a = 0.5, then the value x*(k) is seen as generated without
preference.
In terms of stable time series, the exponential smoothing method employed in
smooth prediction, focuses on the generation of predictions with more preference
given to old information than new information. This is because the smoothing value

ð1Þ ð1Þ
sk ¼ axk þ ð1  aÞsk1

stands for a weighted sum of old and new information, with the weight a taking
value from the range of 0.1–0.3.
58 4 Sequence Operators and Grey Data Mining

Definition 4.4.2 Assume that sequence X has a blank entry ∅(k) at location k. This
blank entry ∅(k) is filled by using the sequence operator D, which is defined as
follows:

xðkÞd ¼ x ðkÞ ¼ 0:5xðk  1Þ þ 0:5xðk þ 1Þ:

If x(k − 1) and x(k + 1) are the adjacent neighbors of the location k, then D will
be referred to as mean generation operator by using the non-adjacent neighbor. If x
(k + 1) stands for a piece of new information, then the non-adjacent neighbor mean
generation operator is an equal weight generation operator of new and old infor-
mation. This kind of operator is employed when it is difficult to determine the
degree of influence of new and old information on the missing value x(k).
Definition 4.4.3 For a given sequence X = (x(1), x(2), …, x(n)), the sequence
operator D is defined as:

xðkÞd ¼ x ðkÞ ¼ 0:5xðkÞ þ 0:5xðk1Þ: ð4:10Þ

In this case, D is referred to as even generation operator by adjacent neighbor.

The sequence worked by even generation operator by adjacent neighbor is
referred to as a sequence of even generated by adjacent neighbor. In grey systems
modeling the sequence of even generated by adjacent neighbor is often employed. It
provides a method of constructing new sequences based on available time series
data.
For the sequence X of length n, as given above, if Z stands for the sequence of
even generated by adjacent neighbor, then the length of Z = (z(2), z(3), …, z(n)) is
n − 1, where z(1) cannot be generated based on what is given in X.

4.5 The Quasi-Smooth Sequence and Stepwise Ratio

Operator

Definition 4.5.1 Assume that X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ , xðkÞ  0; k ¼ 1; 2;    ; n;

then the following is referred to as the smoothness ratio of the sequence X (Deng
1985):

xð k Þ
qðkÞ ¼ Pk1 ; k ¼ 2; 3;   ; n: ð4:11Þ
i¼1 xðiÞ

The concept of smoothness ratio reflects the smoothness of a sequence from a

special angle. In particular, it uses the ratio qðkÞ of the kth data value x(k) over the
4.5 The Quasi-Smooth Sequence and Stepwise Ratio Operator 59

P
sum k1i¼1 xðiÞ of the previous values to check whether or not the changes in the
data points of X are stable. The more stable the changes of the data points in
sequence X are, the smaller the smoothness ratio qðkÞ:
Definition 4.5.2 If a sequence X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ , xðkÞ  0; k ¼ 1; 2;    ; n
satisfies the following, then X is referred to as a quasi-smooth sequence:

(1) qðqkðþkÞ1Þ \1; k ¼ 2; 3;   ; n  1;

(2) qðkÞ 2 ½0; e; k ¼ 3; 4;   ; n; and
(3) e\0:5:
Quasi-smooth conditions are very important criteria, which are employed to
check whether a sequence can be used to build a grey model.
If the first entry x(1) or the last entry x(n) of a sequence are blank, that is, x
(1) = ∅(1) or x(n) = ∅(n), we cannot fill these missing entries by using the method
of adjacent neighbor mean generation operator. In this case, the operator of step-
wise ratio is often employed.
Definition 4.5.3 Assume that a sequence X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ , xðkÞ  0;
k ¼ 1; 2;    ; n; then the following is referred to as the operator of stepwise ratios of
X (Deng 1985):

xð k Þ
xðk Þd ¼ rðk Þ ¼ ; k ¼ 2; 3;   ; n: ð4:12Þ
x ð k  1Þ

The missing entry x(1) = ∅(1) can be generated by using the operator of step-
wise ratio of its right-hand side neighbors, and x(n) = ∅(n) its left-hand side
neighbors. The sequence obtained by filling all its missing entries using the oper-
ators of stepwise ratio is referred to as stepwise ratio generated.
Proposition 4.5.1 Assume that a sequence X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ,
xðkÞ  0; k ¼ 1; 2;    ; n; and x(1) = ∅(1) or x(n) = ∅(n). If both x(1) and x(n) are
generated by operator of stepwise ratio, then:

xð1Þ ¼ xð2Þ=rð3Þ; xðnÞ ¼ xðn  1Þrðn  1Þ:

Proposition 4.5.2 Stepwise ratio rðk þ 1Þ and smoothness ratio as defined in

formulas (4.11) and (4.12), respectively, satisfy the relation as follows:

qðk þ 1Þ
rðk þ 1Þ ¼ ð1 þ qðk ÞÞ; k ¼ 2; 3;   ; n: ð4:13Þ
qðk Þ

Proposition 4.5.3 If X = (x(1), x(2), …, x(n)) is an increasing sequence, and

satisfies the following conditions:
60 4 Sequence Operators and Grey Data Mining

(1) For any k ¼ 2; 3;   ; n; rðkÞ\2; and

(2) qðqkðþkÞ1Þ \1;
then for any e 2 ½0; 1 and k ¼ 2; 3;    ; n; when qðkÞ 2 ½0; e; we have
rðk þ 1Þ 2 ½1; 1 þ e:

4.6 Accumulating and Inverse Accumulating Operators

Accumulating operator is a method employed to mine the law implied in a grey data
sequence. It plays an extremely important role in grey system modelling Through
the accumulating operator method, one can potentially uncover a development
tendency existing in the process of accumulated grey quantities. This allows the
characteristics and laws of integration hidden in chaotic original data to be suffi-
ciently revealed. For instance, when looking at the financial outflows of a family, if
we do our computations on a daily basis, we may not see obvious patterns.
However, if our calculations are done on a monthly basis, some patterns of
spending, which are somehow related to the monthly income of the family, will
likely emerge.
The inverse accumulating operator is often employed to acquire additional
insights from a small amount of available information. It plays the role of recovery
from the acts of the accumulating operator and is its inverse operation. In particular,
Definition 4.6.1 For an original sequence X(0) = (x(0)(1), x(0)(2),…, x(0)(n)), D is a
sequence operator defined as follows:
X(0)D = (x(0)(1)d, x(0)(2)d,…, x(0)(n)d), where

X
k
xð0Þ ðkÞd ¼ xð0Þ ðiÞ; k ¼ 1; 2;   ; n: ð4:14Þ
i¼1

Here, D is called a once accumulating generation operator of X ð0Þ , denoted as

1-AGO. And X ð0Þ D; the sequence worked by accumulating operator D on X(0), is
denoted as X ð1Þ for parsimony:


X ð0Þ D ¼ X ð1Þ ¼ xð0Þ ð1Þd; xð0Þ ð2Þd; . . .; xð0Þ ðnÞd :

If the accumulating operator D is applied r times on X(0), we obtain:


X ð0Þ Dr ¼ X ðrÞ ¼ xðrÞ ð1Þ; xðrÞ ð2Þ;   ; xðrÞ ðnÞ
4.6 Accumulating and Inverse Accumulating Operators 61

where

X
k
xðrÞ ðk Þ ¼ xðr1Þ ðiÞ; k ¼ 1; 2;   n: ð4:15Þ
i¼1

Dr is denoted as r-AGO (Deng 1985). Corresponding to the accumulating

operator, the inverse accumulating operator D is defined below.
Definition 4.6.2 For an original sequence X(0) = (x(0)(1), x(0)(2),…, x(0)(n)), D is a
sequence operator defined as follows:
X(0)D = (x(0)(1)d, x(0)(2)d,…, x(0)(n)d), where

xð0Þ ðkÞd ¼ xð0Þ ðkÞ  xð0Þ ðk  1Þ; k ¼ 2;   ; n: ð4:16Þ

D is called an inverse accumulating generation operator of X ð0Þ , denoted as

1-IAGO. In X ð0Þ D; the sequence works by inverse accumulating operator D on X(0),
and is denoted as að1Þ X ð0Þ .
If the inverse accumulating operator D is applied r times on X(0), we write
conventionally:


X ð0Þ Dr ¼ aðrÞ X ð0Þ ¼ aðrÞ xð0Þ ð1Þ; aðrÞ xð0Þ ð2Þ;   ; aðrÞ xð0Þ ðnÞ

where aðrÞ xð0Þ ðk Þ ¼ aðr1Þ xð0Þ ðkÞ  aðr1Þ xð0Þ ðk  1Þ; k ¼ 1; 2;   ; n (Deng 1985).
Proposition 4.6.1 For an original sequence X(0) = (x(0)(1), x(0)(2),…, x(0)(n)), if
both X ðrÞ and aðrÞ are defined according to Definitions 4.6.1 and 4.6.2, then:

aðrÞ X ðrÞ ¼ X ð0Þ :

Example 3.6.1 If X = (5.3, 7.6, 10.4, 13.8, 18.1), calculate the 1-AGO X ð1Þ ,
2-AGO X ð2Þ and 1-IAGO að1Þ X ð0Þ .
Solution: The results are shown in Table 4.3.

Table 4.3 The 1-AGO,

X ð0Þ 5.3 7.6 10.4 13.8 18.1
2-AGO and 1-IAGO of X ð0Þ
ð1Þ 5.3 12.9 23.3 37.1 55.2
X
ð2Þ 5.3 18.2 41.5 78.6 133.8
X
að1Þ X ð0Þ 5.3 2.3 2.8 3.4 4.3
62 4 Sequence Operators and Grey Data Mining

4.7 Exponentiality of Accumulating Generation

After applying the accumulating operator a few times, the general non-negative
quasi-smooth sequence will show the pattern of exponential growth with decreased
randomness. The smoother the original sequence is, the more obvious an expo-
nential growth pattern in the first order accumulation generated sequence will
appear.
Example 4.7.1 The sales quantity of cars from 2010 to 2015 in a city located in
southeast of China is as follows:
n o6
X ð0Þ ¼ xð0Þ ðkÞ ¼ ð50810; 46110; 51177; 93775; 110574; 110524Þ:
1

The 1-AGO sequence of X ð0Þ is:

n o6
X ð1Þ ¼ xð1Þ ðkÞ ¼ ð50810; 96920; 148097; 241872; 352446; 462970Þ:
1

The Figures of X ð0Þ and X ð1Þ are shown in Figs. 4.3 and 4.4, respectively.
For the curve shown in Fig. 4.3, it is difficult to find a simple curve as the
approximation of X ð0Þ . However, the curve shown in Fig. 4.4 is very close to an
exponential growth curve. X ð1Þ can be fitted with an exponential curve.
Definition 4.7.1 Assume that X ðtÞ ¼ ceat þ b; c; a 6¼ 0 is a continuous exponential
function, then:
(1) XðtÞ is referred to as homogeneous exponential function, if b = 0;
(2) XðtÞ is referred to as non-homogeneous exponential function, if b 6¼ 0.

Definition 4.7.2 If a sequence X = (x(1), x(2), …, x(n)) satisfies:

Fig. 4.3 The curve of X(0)

0
4.7 Exponentiality of Accumulating Generation 63

Fig. 4.4 The curve of X(1)

(1) x(k) = ceak, c, a 6¼ 0, for k = 1, 2, …, n, then X is referred to as a homoge-

neous exponential sequence; and
(2) x(k) = ceak + b, c, a, b 6¼ 0, for k = 1, 2, …, n, then X is referred to as a
non-homogeneous sequence.

Theorem 4.7.1 A sequence X = (x(1), x(2), …, x(n)) is a homogeneous exponential

sequence if, and only if, for k = 1, 2, …, n, r(k) is a constant.
Proof (1) Assume that 8 k = 1, 2, …, n, xðkÞ ¼ ceak , c; a 6¼ 0; then:

xðkÞ ceak
rðkÞ ¼ ¼ aðk1Þ ¼ ea ¼ const:
xðk  1Þ ce

(2) Assume that 8 k = 1, 2, …, n, rðkÞ ¼ const ¼ ea , then:

xðkÞ ¼ ea xðk  1Þ ¼ e2a xðk  2Þ ¼    ¼ xð1Þeaðk1Þ :

Definition 4.7.3 For the given sequence X = (x(1), x(2), …, x(n)),

(1) if 8 k, r(k) 2 (0, 1), then X is referred to as satisfying the law of negative grey
exponent
(2) if 8 k, r(k) 2 (1, b), for some b > 1, then X is referred to as satisfying the law
of positive grey exponent
(3) if 8 k, r(k) 2 [a, b], b − a = d, then X is referred to as satisfying the law of
grey exponent with the absolute degree of greyness d; and
(4) if d < 0.5, then X is referred to as satisfying the law of quasi-exponent

Theorem 4.7.2 Assume that X(0) is a non-negative quasi-smooth sequence. Then,

the sequence X(1), generated by applying accumulating generation once on X(0),
satisfies the law of quasi-exponent.
64 4 Sequence Operators and Grey Data Mining

Proof According to the definition of quasi-smooth sequence and

xð1Þ ðkÞ xð0Þ ðkÞ þ xð1Þ ðk  1Þ

rð1Þ ðkÞ ¼ ¼ ¼ 1 þ qðkÞ
xð1Þ ðk  1Þ xð1Þ ðk  1Þ

We have
8k; qðkÞ\0:5

Therefore
rð1Þ ðkÞ 2 ½1; 1:5Þ; d\0:5:

Thus, X ð1Þ is a sequence that satisfies the law of quasi-exponent.

Theorem 4.7.2 is the theoretical foundation of grey systems modeling. In fact,
because economic, ecological and agricultural systems (among others) can be seen
as energy systems, and given that the accumulation and release of energy generally
satisfy an exponential law, this explains why exponential modeling of grey systems
theory has found an extremely wide range of applications.
Theorem 4.7.3 Assume that X(0) is a non-negative sequence. If X(r) satisfies a law
of exponent, and the stepwise ratio of X(r) is given by r(r)(k) = r, then according to
Deng (1985):
1  rk
(1) rðr þ 1Þ ðkÞ ¼ ;
1  rk  1
(2) When r 2 (0, 1), lim rðr þ 1Þ ðkÞ ¼ 1; and for each k, rðr þ 1Þ ðkÞ2 ð1; 1 þ r;
k!1
(3) When r [ 1; lim rðr þ 1Þ ðkÞ ¼ r ; and for each k, rðr þ 1Þ ðkÞ2 ðr; 1 þ r:
k!1

Proof (1) Assume that X ðrÞ satisfies a law of exponent, and 8 k,

ðrÞ xðrÞ ðkÞ
r ðkÞ ¼ xðrÞ ðk1Þ
¼ r, then 8 k,

xðrÞ ðkÞ ¼ rxðrÞ ðk  1Þ ¼ r2 xðrÞ ðk  2Þ ¼    ¼ rðk1Þ xðrÞ ð1Þ

X ðrÞ ¼ ðxðrÞ ð1Þ; rxðrÞ ð1Þ; r2 xðrÞ ð1Þ;   ; rðn1Þ xðrÞ ð1ÞÞ

X ðr þ 1Þ ¼ ðxðrÞ ð1Þ; ð1 þ rÞxðrÞ ð1Þ; ð1 þ r þ r2 ÞxðrÞ ð1Þ;   ;

ð1 þ r þ    þ rðn1Þ ÞxðrÞ ð1ÞÞ

Therefore
xðr þ 1Þ ðkÞ ð1 þ r þ    þ rk1 ÞxðrÞ ð1Þ
k
1r
1  rk
rðr þ 1Þ ðkÞ ¼ ¼ ¼ 1r
1r
k1 ¼
x ðr þ 1Þ ðk  1Þ ð1 þ r þ    þ r Þx ð1Þ
k2 ðrÞ
1r
1  rk1
4.7 Exponentiality of Accumulating Generation 65

(2) When r 2 ð0; 1Þ; rðr þ 1Þ ðkÞ will decrease as k increases.

k¼2

xðr þ 1Þ ð2Þ
rðr þ 1Þ ð2Þ ¼ ¼ 1þr
xðr þ 1Þ ð1Þ

k!1

1  rk
rðr þ 1Þ ðkÞ ¼ !1
1  rk1

Therefore 8 k,

rðr þ 1Þ ðkÞ 2 ½1; 1 þ r

(3) When r [ 1; rðr þ 1Þ ðkÞ will decrease as k increases.

k¼2

rðr þ 1Þ ð2Þ ¼ 1 þ r

k!1

1  rk
rðr þ 1Þ ðkÞ ¼ !r
1  rk1

Therefore 8 k,

rðr þ 1Þ ðkÞ 2 ðr; 1 þ r

The Theorem 4.7.3 says that if the rth accumulating generation sequence of X(0)
satisfies an obvious law of exponent, additional application of the accumulating
generation operator will destroy the pattern of exponent. In practical applications, if
the rth accumulating generation sequence of X(0) satisfies the law of
quasi-exponent, we generally stop applying the accumulating generation operator.
To this end, Theorem 4.7.2 implies that only one application of the accumulating
generation operator is needed for a non-negative quasi-smooth sequence before
establishing an exponential model.
Chapter 5
Grey Incidence Analysis Models

5.1 Introduction

Any given system, such as a social, economic, agricultural, ecological, and edu-
cational system, will encompass different kinds of factors. It is the result of the
mutual interactions of these factors that determines the development tendency and
behavior of the system. It is often the case that, among all the factors, investigators
will need to know which ones are primary and which ones are secondary. Primary
factors have dominant effects on the development of systems. Such factors drive the
development of systems positively and must be strengthened. Conversely, sec-
ondary factors exert less influence on the development of systems. They tend to
pose obstacles for the development of systems and, therefore, must be weakened.
For instance, there are generally many influencing factors on the overall perfor-
mance of an economic system. In order to realize the production of additional
output with less input, systems analysis must be conducted prudently and a key part
of this analysis is the identification of primary and secondary factors. Regression
analysis, variance analysis, and main component analysis are the most commonly
employed methods for conducting systems analysis. However, these methods suffer
from the following weaknesses:
(1) Large samples are needed in order to produce reliable conclusions.
(2) Available data need to satisfy some typical types of probability distribution;
linear relationships between factors and system behaviors are assumed, while
no interactions can be found between factors. Generally, these requirements
are difficult to satisfy.
(3) The amount of computation is large and generally done by using computers.
(4) At times quantitative conclusions do not resonate with qualitative analysis
outcomes so that the laws governing system development are distorted or
misunderstood.

© Springer Science+Business Media Singapore 2017 67

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_5
68 5 Grey Incidence Analysis Models

In fact, when available data are small it is extremely difficult to apply such
traditional methods of statistics to analyze such data. This is because small data do
not satisfy the modelling conditions of traditional methods; they contain relatively
large amounts of grey information and do not follow any conventional probability
distribution.
The Grey Incidence Analysis (GIA) model is a new method to analyze systems
where statistical methods do not seem appropriate. It can be applied to large or
small samples and does not have conventional distribution requirements.
Additionally, the amount of computation involved is small and can be carried out
conveniently, without issues of disagreement between quantitative and qualitative
conclusions.
The basic idea of grey incidence analysis is to use the degree of similarity of the
geometric curves of available data sequences to determine whether or not their
connections are close. The more similar the curves, the closer the incidence between
sequences, and vice versa.
A number of scholars have conducted meaningful research focused on the
construction and properties of GIA models, and such researchers have achieved
valuable results. For example, Zhang (1999) has analyzed the predominant point
trend of Deng’s (1985) GIA model. Zhang has introduced grey relation entropy to
improve the traditional model, and has proposed a new method to calculate degree
of grey incidence. Xiao and colleagues (Xiao et al. 1995; Xiao and Guo 2010) have
constructed a weighted degree of grey incidence through the weighted compound of
incidence coefficient of each point. Zhao et al. (1998) have introduced Euclid
nearness into grey incidence analysis, and have established the Euclid incidence
degree model based on the measurement of nearness of factor points through cal-
culating nearness. Furthermore, Zhao and Wei (1998) have defined a GIA model
according to upper and lower boundaries of distances between grey factor points.
The authors have also demonstrated that their GIA model as well as Deng’s (1985)
GIA model through weighted incidence analysis and the Euclid incidence degree
model are three special types of GIA model. Shi (1995) has proposed extreme
difference relation according to the difference between distance of maximum value
and distance of sequences, complementing Deng’s (1985) incidence coefficient.
Zhang et al. (1997) have integrated the method of discrimination coefficient cor-
rection, the entropy weight method and the projection method to advance Deng’s
(1985) GIA model. Zhao, Ma, and Jia (2007) has introduced variant coefficient to
incidence analysis, and improved Deng’s GIA model through weighted values of
variant coefficient and incidence coefficient. Further, Zhou and Chen (2005) defined
incidence coefficient with the application of generalized distance in fuzzy math to
measure the difference between reference sequence and compared sequence. Peng
(2008) has extended Deng’s (1985) GIA model to second-order trend incidence
analysis model through second-order difference. Finally, Wang (1989) has pro-
posed the B-type incidence degree model, Tang (1995) has developed the T-type
incidence degree model, and Dang et al. (2004) has proposed the gradient incidence
degree model as well as its improved version. Among these models, the GIA model
proposed by Professor Deng Julong (1985) is the most influential one.
5.1 Introduction 69

Thus, research based on early GIA models relies on incidence coefficients of

particular points to the absolute degree, relative degree, and synthetic degree of the
original GIA model, which in turn is based on integral or overall perspectives. Such
research also includes GIA models that measure similarity based on nearness to the
models, which consider similarity and nearness, respectively. Additionally, research
objects extend from the analysis of relationship among curves to those of rela-
tionships among curved surfaces, analysis of relationships in three-dimensional
space and even the analysis of relationships among super surfaces in n-dimensional
space. However, the study of high-dimensional models is still in its infancy. Indeed,
many practical and scientific problems are yet to receive research attention and
there is a need to focus on analysis methods based on panel data, matrix data,
matrix sequence data and high-dimensional data. The absolute degree of GIA
model, which extends definite integral models to multiple integral ones, can be used
for incidence analysis of high-dimensional data. However, the testing and specific
quantitative standards of GIA models require additional research.

5.2 Grey Incidence Factors and Set of Grey Incidence

Operators

When analyzing a system, one must choose the quantity of factors to reflect the
characteristics of such a system, and determine the factors that influence the
behavior of the system. If a quantitative analysis is considered, one needs to process
the chosen characteristics quantity and the effective factors using sequence opera-
tors so that the available data are converted to their relevant non-dimensional values
of roughly equal magnitudes.
Definition 5.2.1 Assume that Xi is a system factor and its observation value at the
ordinal position k is xi ðkÞ; k ¼ 1; 2;    ; n; then Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is
referred to as the behavioral sequence of factor Xi .
If k stands for the time order, then xi ðkÞ is referred to as the observational value
of factor Xi at time moment k, and Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is the behavioral
time sequence (or series) of Xi .
If k stands for an index ordinal number and xi ðkÞ the observational value of the
kth index of factor Xi , then Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is referred to as the
behavioral index sequence of factor Xi .
If k stands for the ordinal number of the observed object and xi ðkÞ is the
observed value of the kth object of factor Xi , then Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is
referred to as the horizontal sequence of factor Xi ’s behavior.
For example, if Xi represents an economic factor, k time, and xi ðkÞ the observed
value of factor Xi at time moment k, then Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is a time
70 5 Grey Incidence Analysis Models

series of economic behaviors. If k is the ordinal number of an index, then

Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is the index sequence of an economic behavior. If
k represents the ordinal number of different economic regions or departments, then
Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ is a horizontal sequence of an economic behavior. No
matter what kinds of sequence data are available, they can be employed in inci-
dence analysis.
Definition 5.2.2 Let Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the behavioral sequence of
factor Xi , and D1 a sequence operator such that Xi D1 = ðxi ð1Þd1 ; xi ð2Þd1 ;    ;
xi ðnÞd1 Þ; where:

xi ðkÞd1 ¼ xi ðkÞ=xi ð1Þ; xi ð1Þ 6¼ 0; k ¼ 1; 2;    ; n: ð5:1Þ

Then D1 is referred to as an initialing operator and Xi D1 is its image, called

initial image of Xi (Deng 1985).
Example 5.2.1 Let X ¼ ð3:2; 3:7; 4:5; 4:9; 5:6Þ; and calculate the initial image of X:
Solution: From formula (5.1), we have:

xð1Þd1 ¼ xð1Þ=xð1Þ ¼ 1; xð2Þd1 ¼ xð2Þ=xð1Þ ¼ 3:7  3:2 ¼ 1:15625:

Similarly,

xð3Þd1 ¼ 1:40625; xð4Þd1 ¼ 1:53125; xð5Þd1 ¼ 1:75:

Therefore:

XD1 ¼ ðxð1Þd1 ; xð2Þd1 ; xð3Þd1 ; xð4Þd1 ; xð5Þd1 Þ

¼ ð1; 1:15625; 1:40625; 1:53125; 1:75Þ:

Definition 5.2.3 Let Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the behavioral sequence of
factor Xi . Sequence operator D2 satisfies Xi D2 = ðxi ð1Þd2 ; xi ð2Þd2 ;    ; xi ðnÞd2 Þ;
and:

xi ðkÞ 1X n
xi ðkÞd2 ¼ ; Xi ¼ xi ðkÞ; k ¼ 1; 2;    ; n: ð5:2Þ
Xi n k¼1

Here, D2 is referred to as an averaging operator and Xi D2 is its image, called the

average image of Xi (Deng 1985).
Example 5.2.2 Let X be the same as Example 5.2.1 and calculate the average image
of X:
5.2 Grey Incidence Factors and Set of Grey Incidence Operators 71

Solution: From formula 5.2, we have:

1X 5

¼ 0:73; xð2Þd2 ¼ xð2Þ=X

¼ 0:84:
X¼ xðkÞ ¼ 4:38; xð1Þd2 ¼ xð1Þ=X
5 k¼1

Similarly:

xð3Þd2 ¼ 1:03; xð4Þd2 ¼ 1:12; xð5Þd2 ¼ 1:28:

Therefore:

XD2 ¼ ðxð1Þd2 ; xð2Þd2 ; xð3Þd2 ; xð4Þd2 ; xð5Þd2 Þ ¼ ð0:73; 0:84; 1:03; 1:12; 1:28Þ:

Definition 5.2.4 Let Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the behavioral sequence of
factor Xi . Sequence operator D3 satisfies Xi D3 ¼ ðxi ð1Þd3 ; xi ð2Þd3 ;    ; xi ðnÞd3 Þ;
and:

xi ðkÞ  mink xi ðkÞ

xi ðkÞd3 ¼ ; k ¼ 1; 2;    ; n: ð5:3Þ
maxk xi ðkÞ  mink xi ðkÞ

D3 is referred to as an interval operator and Xi D3 is its image, called the interval

image of Xi (Deng 1985).
Example 5.2.3 Let X be the same as Example 5.2.1, and calculate the interval
image of X:
Solution: min xðkÞ ¼ 3:2, max xðkÞ ¼ 5:6: From formula (5.3), we have:
k k

xð1Þd3 ¼ 0; xð2Þd3 ¼ 0:208

xð3Þd3 ¼ 0:542; xð4Þd3 ¼ 0:708; xð5Þd3 ¼ 1:

Therefore:

XD3 ¼ ðxð1Þd3 ; xð2Þd3 ; xð3Þd3 ; xð4Þd3 ; xð5Þd3 Þ ¼ ð0; 0:208; 0:542; 0:708; 1Þ:

As usual, D1 , D2 , D3 should not be mixed or overlapped. Only one of them can

be selected according to a particular situation.
Definition 5.2.5 Let Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the behavioral sequence of
factor Xi . The behavioral sequence of factor Xi satisfies xi ðkÞ 2 ½0; 1; i ¼ 1; 2;    ; n;
sequence operator D4 satisfies Xi D4 ¼ ðxi ð1Þd4 ; xi ð2Þd4 ;    ; xi ðnÞd4 Þ; and:
72 5 Grey Incidence Analysis Models

xi ðkÞd4 ¼ 1  xi ðkÞ; k ¼ 1; 2;    ; n: ð5:4Þ

Then D4 is referred to as a reversing operator and Xi D4 is its image, called the

reverse image of Xi (Deng 1985).
Definition 5.2.6 Let Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the behavioral sequence of
factor Xi . Sequence operator D5 satisfies Xi D5 ¼ ðxi ð1Þd5 ; xi ð2Þd5 ;    ; xi ðnÞd5 Þ;
and:

xi ðkÞd5 ¼ 1=xi ðkÞ; xi ðkÞ 6¼ 0; k ¼ 1; 2;    ; n: ð5:5Þ

Here, D5 is referred to as a reciprocating operator with Xi D5 as its image, called

the reciprocal image of Xi (Deng 1985).
Let X0 be the sequence of a system’s behavioral characteristics, which is
increasing, and Xi the behavioral sequence of a relevant factor. If Xi is also an
increasing sequence then both Xi and X0 have a positive or direct incidence
relationship. If Xi is a decreasing sequence then both Xi and X0 have a negative or
inverse incidence relationship.
The negative relationship will be transformed to a positive relationship if
affected by reversing operator D4 or reciprocating operator D5 . Here, D4 and D5
should not be mixed or overlapped either.
Definition 5.2.7 The set D ¼ fDi ji ¼ 1; 2; 3; 4; 5g is referred to as the set of grey
incidence operators
Definition 5.2.8 If X stands for the set of all system factors and D the set of grey
incidence operators, then ðX; DÞ is referred to as the space of grey incidence factors
of a system.

5.3 Degrees of Grey Incidences Model

Given the sequence X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ, we can image the corresponding
zigzagged line of the plane X = fxðkÞ þ ðt  kÞðxðk þ 1Þ  xðkÞÞjk ¼ 1; 2;    ;
n  1; t 2 ½k; k þ 1g. Without causing confusion, the same symbol is used for both
the sequence and its zigzagged line. For parsimony, we will not distinguish between
the two in our discussions.
Because sequences of inverse incidence relationships can be transformed into
those of direct incidence relationships by using either reversing or reciprocating
operators, we will focus our attention on the study of positive incidence
relationships.

Definition 5.3.1 The given sequence X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ; a ¼ xðsÞxðkÞ sk ;
s [ k; k ¼ 1; 2;    ; n  1; is referred to as the slope of X on interval ½k; s; and
a ¼ n1
1
ðxðnÞ  xð1ÞÞ the average slope of X.
5.3 Degrees of Grey Incidences Model 73

Theorem 5.3.1 Assume that Xi and Xj are non-negative increasing sequences such
that Xj ¼ Xi þ c; where c is a nonzero constant. Let D1 be an initialing operator,
Yi ¼ Xi D1 and Yj ¼ Xj D1 . If ai and aj are respectively the average slopes of Xi and
Xj , and bi and bj the average slopes of Yi and Yj , then, the following must be true:
ai ¼ aj ; when c\0; bi \bj ; and when c [ 0; bi [ bj .
What is meant here is that when the absolute amount of increase of two
increasing sequences are the same, the sequence with the smallest initial value will
increase faster than the other. To maintain the same relative rate of increase, the
absolute amount of increase of the sequence with the greatest initial value must be
greater than that of the sequence with the smallest initial value.
Definition 5.3.2 Let X0 ¼ ðx0 ð1Þ; x0 ð2Þ;    ; x0 ðnÞÞ be a data sequence of a sys-
tem’s behavioral characteristic and the following are relevant factor sequences:

X1 ¼ ðx1 ð1Þ; x1 ð2Þ;    ; x1 ðnÞÞ

. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .

Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ

. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .

Xm ¼ ðxm ð1Þ; xm ð2Þ;    ; xm ðnÞÞ

Given real numbers cðx0 ðkÞ; xi ðkÞÞ; i ¼ 1; 2; . . .; m; and k ¼ 1; 2; . . .; n; if the

following
1X n
cðX0 ; Xi Þ ¼ cðx0 ðkÞ; xi ðkÞÞ:
n k¼1

satisfies conditions of normality (1) and closeness (2) below:

(1) Normality: 0\cðX0 ; Xi Þ  1;cðX0 ; Xi Þ ¼ 1 , X0 ¼ Xi ; and
(2) Closeness: the smaller jx0 ðkÞ  xi ðkÞj; the greater cðx0 ðkÞ; xi ðkÞÞ:
In this case, cðX0 ; Xi Þ is referred to as the degree of grey incidence between Xi
and X0 , cðx0 ðkÞ; xi ðkÞÞ as the incidence coefficient of Xi and X0 at point k (Deng
1985).
Theorem 5.3.2 Given a system’s behavioral sequences X0 ¼ ðx0 ð1Þ; x0 ð2Þ;    ;
x0 ðnÞÞ and Xi = ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ; i ¼ 1; 2; . . .; m; for n 2 ð0; 1Þ; it is pos-
sible to define:

mini mink jx0 ðkÞ  xi ðkÞj þ n maxi maxk jx0 ðkÞ  xi ðkÞj
cðx0 ðkÞ; xi ðkÞÞ ¼ : ð5:6Þ
jx0 ðkÞ  xi ðkÞj þ n maxi maxk jx0 ðkÞ  xi ðkÞj
74 5 Grey Incidence Analysis Models

And:

1X n
cðX0 ; Xi Þ ¼ cðx0 ðkÞ; xi ðkÞÞ: ð5:7Þ
n k¼1

In this case, cðX0 ; Xi Þ is a degree of grey incidence between X0 and Xi , where n is

known as the distinguishing coefficient (Deng 1985).
The degree of grey incidence cðX0 ; Xi Þ is commonly written as c0i , and the
incidence coefficient cðx0 ðkÞ; xi ðkÞÞ as c0i ðkÞ: c0i is also referred to as Deng’s degree
of grey incidence.
Based on Theorem 5.3.1, the computation steps of the degree of grey incidence
can be accomplished as explained below.
Step 1 Calculate the initial image (or average image) of X0 and Xi , i ¼ 1; 2; . . .; m;
where:

Xi0 ¼ Xi =xi ð1Þ ¼ ðx0i ð1Þ; x0i ð2Þ;    ; x0i ðnÞÞ i ¼ 0; 1; 2;    ; m:

Step 2 Compute the difference sequences of X00 and Xi0 , i ¼ 1; 2; . . .; m; and write
as:
 
Di ðkÞ ¼ x00 ðkÞ  x0i ðkÞ; D ¼ ðDi ð1Þ; Di ð2Þ;    ; Di ðnÞÞ i ¼ 1; 2;    ; m:

Step 3 Find the maximum and minimum differences, and denote as:

M ¼ max max Di ðkÞ; m ¼ min min Di ðkÞ

i k i k

Step 4 Calculate the incidence coefficients:

m þ nM
c0i ðkÞ ¼ ; n 2 ð0; 1Þ k ¼ 1; 2;    ; n; i ¼ 1; 2;    ; m:
Di ðkÞ þ nM

Step 5 Compute the degree of grey incidence:

1X n
c0i ¼ c ðkÞ; i ¼ 1; 2;    ; m:
n k¼1 0i

Example 5.3.1 The Gross Domestic Product (GDP) of China and the value-added
contributions of the primary, secondary and tertiary industries are as follows:
GDP: X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5ÞÞ ¼ ð109:7; 120:3; 135:8; 159:9; 183:1Þ:
Value-added contribution of the primary industry: X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ;
x2 ð4Þ; x2 ð5ÞÞ ¼ ð15:5; 16:2; 17:1; 21:0; 23:1Þ:
Value-added contribution of the secondary industry: X3 ¼ ðx3 ð1Þ; x3 ð2Þ; x3 ð3Þ;
x3 ð4Þ; x3 ð5ÞÞ ¼ ð49:5; 53:9; 62:4; 73:9; 87:0Þ:
5.3 Degrees of Grey Incidences Model 75

Value-added contribution of the tertiary industry: X4 ¼ ðx4 ð1Þ; x4 ð2Þ; x4 ð3Þ;

x4 ð4Þ; x4 ð5ÞÞ ¼ ð44:6; 50:2; 56:3; 65:0; 73:0Þ:
Unit: One hundred billion; data sources: China Statistical Yearbook 2006.
Calculate the degree of grey incidence between X0 and Xi , i = 1, 2, 3.
Solution: Take X1 as the system’s behavioral characteristics sequence.
Step 1
Calculate the initial image of Xi , i ¼ 1; 2; 33; 4

From Xi0 ¼ Xi =xi ð1Þ ¼ ðx0i ð1Þ; x0i ð2Þ; x0i ð3Þ; x0i ð4Þ; x0i ð5ÞÞ; i ¼ 1; 2; 3; 4; we have

X10 ¼ ð1; 1:0966; 1:2379; 1:4576; 1:6691Þ

X20 ¼ ð1; 1:0452; 1:1032; 1:3548; 1:4903Þ

X30 ¼ ð1; 1:0889; 1:2606; 1:4929; 1:7576Þ

X40 ¼ ð1; 1:1256; 1:2623; 1:4574; 1:6368Þ

Step 2
Compute the difference sequences
 
From Di ðkÞ ¼ x01 ðkÞ  x0i ðkÞ ; i ¼ 2; 3; 4, it follows that

D2 ¼ ð0; 0:0515; 0:1347; 0:1028; 0:1788Þ

D3 ¼ ð0; 0:0077; 0:0227; 0:0353; 0:0885Þ

D4 ¼ ð0; 0:0289; 0:0244; 0:0002; 0:0323Þ

Step 3
Find the maximum and minimum differences

M ¼ max max Di ðkÞ ¼ 0:1788

i k

m ¼ min min Di ðkÞ ¼ 0

i k

Step 4
Calculate the incidence coefficients

Let n ¼ 0:5, it follows that:

m þ nM 0:0894
c1i ðkÞ ¼ ¼ ; i ¼ 2; 3; 4; k ¼ 1; 2; 3; 4; 5:
Di ðkÞ þ nM Di ðkÞ þ 0:0894
76 5 Grey Incidence Analysis Models

Therefore

r12 ð1Þ ¼ 1; r12 ð2Þ ¼ 0:6346; r12 ð3Þ ¼ 0:3989; r12 ð4Þ ¼ 0:4652; r12 ð5Þ ¼ 0:3333

r13 ð1Þ ¼ 1; r13 ð2Þ ¼ 0:9203; r13 ð3Þ ¼ 0:7976; r13 ð4Þ ¼ 0:7168; r13 ð5Þ ¼ 0:5026

r14 ð1Þ ¼ 1; r14 ð2Þ ¼ 0:7555; r14 ð3Þ ¼ 0:7855; r14 ð4Þ ¼ 0:9976; r14 ð5Þ ¼ 0:7344

Step 5
Compute the degree of grey incidence

1X 5
c12 ¼ c ðkÞ ¼ 0:5664
5 k¼1 12

1X 5
c13 ¼ c ðkÞ ¼ 0:7875
5 k¼1 13

1X 5
c14 ¼ c ðkÞ ¼ 0:8546
5 k¼1 14

The maximum value is the degree of grey incidence between GDP and the
tertiary industry. The intermediate value is the degree of grey incidence between
GDP and the secondary industry. Finally, the minimum value is the degree of grey
incidence between GDP and the primary industry.

5.4 Absolute Degree of Grey Incidence Model

Proposition 5.4.1 Let Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the data sequence of a
system’s behavior, Xi  xi ð1Þ denote the zigzagged line ðxi ð1Þ  xi ð1Þ; xi ð2Þ 
xi ð1Þ;    ; xi ðnÞ  xi ð1ÞÞ; and

Zn
si ¼ ðXi  xi ð1ÞÞdt: ð5:8Þ
1

Then, when Xi increases, si  0; when Xi decreases, si  0; and when Xi vibrates,

the sign of si varies.
The results of Proposition 5.4.1 are represented in Fig. 5.1, where (a) shows the
case where the sequence increases; (b) the situation where Xi decreases; and (c) the
scenario where Xi vibrates.
5.4 Absolute Degree of Grey Incidence Model 77

Fig. 5.1 The zigzagged line of proposition 5.1

Definition 5.4.1 Let Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ be the data sequence of a sys-
tem’s behavior and D the sequence operator which satisfies Xi D ¼
ðxi ð1Þd; xi ð2Þd;    ; xi ðnÞdÞ and xi ðkÞd ¼ xi ðkÞ  xi ð1Þ; k ¼ 1; 2;    ; n: Then D is
referred to as a zero-starting point operator and Xi D is the image of Xi . Xi D is often
written as Xi D = Xi0 = ðx0i ð1Þ; x0i ð2Þ;    ; x0i ðnÞÞ:
Proposition 5.4.2 Assume that the images of the zero-starting point of two
behavioral sequences Xi and Xj are respectively Xi0 ¼ ðx0i ð1Þ; x0i ð2Þ;    ; x0i ðnÞÞ and
Xj0 ¼ ðx0j ð1Þ; x0j ð2Þ;    ; x0j ðnÞÞ: Let:

Zn
si  sj ¼ ðXi0  Xj0 Þdt; and ð5:9Þ
1

Zn
Si  S j ¼ ðXi  Xj Þdt: ð5:10Þ
1

Then, when Xi0 is entirely located above Xj0 , si  sj  0; when Xi0 is entirely
underneath Xj0 , si  sj  0; and when Xi0 and Xj0 alternate their positions, the sign of
si  sj is not fixed.
As shown in Fig. 5.2, when Xi0 is entirely located above Xj0 (Fig. 5.2a), the
shaded area is positive so that si  sj  0: When Xi0 and Xj0 alternate their positions
(Fig. 5.2b), the sign of si  sj is not fixed. Similarly, We can discuss the sign of
Si  Sj as si  sj .
Definition 5.4.2 The sum of time intervals between consecutive observation values
of a sequence Xi is called the length of Xi . It should be noted that two sequences
with the same length may not have the same number of data. For example:

X1 ¼ ðx1 ð1Þ; x1 ð3Þ; x1 ð6ÞÞ

X2 ¼ ðx2 ð1Þ; x2 ð3Þ; x3 ð5Þ; x2 ð6ÞÞ

78 5 Grey Incidence Analysis Models

Fig. 5.2 A description of the relationship between Xi0 and Xj0

X3 ¼ ðx3 ð1Þ; x3 ð2Þ; x3 ð3Þ; x3 ð4Þ; x3 ð5Þ; x3 ð6ÞÞ

The lengths of X1 ; X2 ; X3 are all 5, but X1 has 3 data, X2 has 4 data, and X3 has 6
data.
Definition 5.4.3 Let Xi and Xj be two sequences of the same length, and si and sj
are defined as above. Then, the following is referred to as the absolute degree of
grey incidence between Xi and Xj , or absolute degree of incidence for short (Liu
1991):
1 þ jsi j þ jsj j
eij ¼ : ð5:11Þ
1 þ jsi j þ jsj j þ jsi  sj j

As for sequences of different lengths, the concept of absolute degree of incidence

can be defined by either shortening the longest sequence or by prolonging the
shortest sequence using appropriate methods. This procedure will ensure that the
sequences have the same length. However, by doing so, the ultimate value of the
absolute degree of incidence will be affected.
Proposition 5.4.3 Assume that Xi and Xj are two sequences with the same length.
Let Xi0 ¼ Xi  a; Xj0 ¼ Xj  b; where a; b are real numbers. Denote e00i as the
absolute degree of grey incidence between Xi0 and Xj0 , then e00i ¼ e0i . In fact, when Xi
and Xj have been transformed, the values of si ; sj , and si  sj are not changed.
Therefore, the value of absolute degree of incidence does not change.
Definition 5.4.4 If the time intervals of any two consecutive observation values of
a sequence Xi with the same length, then Xi is called an equal-time-interval
sequence.
Lemma 5.4.1 Assume that Xi is an equal-time-interval sequence. If the length of
time-interval l 6¼ 1; then following can transform Xi into an 1-time-interval
sequence:
5.4 Absolute Degree of Grey Incidence Model 79

t:T!T

t 7! t=l

Lemma 5.4.2 Assume that Xi and Xj are 1-time-interval sequences of the same
length, and the following are zero-starting point images of Xi and Xj :

Xi0 ¼ ðx0i ð1Þ; x0i ð2Þ;    ; x0i ðnÞÞ

Xj0 ¼ ðx0j ð1Þ; x0j ð2Þ;    ; x0j ðnÞÞ

Then, according to Liu (1991):

 
Xn1 
 1 
j si j ¼  x0i ðkÞ þ x0i ðnÞ
 k¼2 2 
 
  X n1
1 

sj  ¼  xj ðkÞ þ xj ðnÞ
0 0
 k¼2 2 
 
  X n1
1 

si  sj  ¼  ðx0i ðkÞ  x0j ðkÞÞ þ ðx0i ðnÞ  x0j ðnÞÞ:
 k¼2 2 

Theorem 5.4.1 Assume that Xi and Xj are two sequences with the same length,
same time distances from one moment to another, and equal time moment intervals.
Then, the absolute degree of grey incidence can also be computed as follows (Liu
1991):
   
Xn1  X n1 
 1   1 
eij ¼ ½1 þ  x0i ðkÞ þ x0j ðnÞ þ  x0j ðkÞ þ x0j ðnÞ
 k¼2 2   k¼2 2 
   
Xn1
1 0  X n1
1 0 

 ½1 þ  x ðkÞ þ xi ðnÞ þ 
0
x ðkÞ þ xj ðnÞ
0
 k¼2 i 2   k¼2 j 2 
 
Xn1 
 1 
þ ðx0i ðkÞ  x0j ðkÞÞ þ ðx0i ðnÞ  x0j ðnÞÞ1
 k¼2 2 

Example 5.4.1 Calculate the absolute degree of grey incidence e01 of sequences X0
and X1 . Let sequences X0 and X1 be as follows:
80 5 Grey Incidence Analysis Models

X0 ¼ ðx0 ð1Þ; x0 ð2Þ; x0 ð3Þ; x0 ð4Þ; x0 ð5Þ; x0 ð7ÞÞ ¼ ð10; 9; 15; 14; 14; 16Þ

X1 ¼ ðx1 ð1Þ; x1 ð3Þ; x1 ð7ÞÞ ¼ ð46; 70; 98Þ

Solution:
Step 1 Transform X1 into a sequence with the same corresponding time-intervals
as X0

1 1
x1 ð2Þ ¼ ðx1 ð1Þ þ x1 ð3Þ ¼ ð46 þ 70Þ ¼ 58
2 2
1 1
x1 ð5Þ ¼ ðx1 ð3Þ þ x1 ð7ÞÞ ¼ ð70 þ 98Þ ¼ 84
2 2
1 1
x1 ð4Þ ¼ ðx1 ð3Þ þ x1 ð5ÞÞ ¼ ð70 þ 84Þ ¼ 77
2 2

Thus, we have a new sequence X1 in place of the original X1 :

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð7ÞÞ ¼ ð46; 58; 70; 77; 84; 98Þ

Step 2 Transform X0 and X1 into equal-time-interval sequences:

1 1
x0 ð6Þ ¼ ðx0 ð5Þ þ x0 ð7ÞÞ ¼ ð14 þ 16Þ ¼ 15
2 2
1 1
x1 ð6Þ ¼ ðx1 ð5Þ þ x1 ð7ÞÞ ¼ ð84 þ 98Þ ¼ 91
2 2

We have:

X0 ¼ ðx0 ð1Þ; x0 ð2Þ; x0 ð3Þ; x0 ð4Þ; x0 ð5Þ; x0 ð6Þ; x0 ð7ÞÞ ¼ ð10; 9; 15; 14; 14; 15; 16Þ

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð6Þ; x1 ð7ÞÞ ¼ ð46; 58; 70; 77; 84; 91; 98Þ

Where X0 and X1 are 1-time-interval sequences.

Step 3 Compute the zero-starting point images of sequences X0 and X1 :

X00 ¼ ðx00 ð1Þ; x00 ð2Þ; x00 ð3Þ; x00 ð4Þ; x00 ð5Þ; x00 ð6Þ; ; x00 ð7ÞÞ ¼ ð0; 1; 5; 4; 4; 5; 6Þ

X10 ¼ ðx01 ð1Þ; x01 ð2Þ; x01 ð3Þ; x01 ð4Þ; x01 ð5Þ; x01 ð6Þ; ; x01 ð7ÞÞ ¼ ð0; 12; 24; 31; 38; 45; 52Þ

Step 4 Calculate js0 j, js1 j, js1  s0 j

5.4 Absolute Degree of Grey Incidence Model 81

 
X 6
1 0 

j s0 j ¼  x ðkÞ þ x0 ð7Þ ¼ 20
0
 k¼2 0 2 
 
X 6
1 0 

j s1 j ¼  x ðkÞ þ x1 ð7Þ ¼ 176
0
 k¼2 1 2 
 
X 6 
 1 0 
j s1  s0 j ¼  ðx1 ðkÞ  x0 ðkÞÞ þ ðx1 ð7Þ  x0 ð7ÞÞ ¼ 156
0 0 0
 k¼2 2 

Step 5 Compute the absolute degree of grey incidence e01 of sequences X0 and X1

1 þ js0 j þ js1 j 197

e01 ¼ ¼ 0:5581:
1 þ js0 j þ js1 j þ js1  s0 j 353

Theorem 5.4.2 The absolute degree of grey incidence eij satisfies the following
properties:
(1) 0\eij  1;
(2) eij is only related to the geometric shapes of Xi and Xj , and has no rela-
tionship with the spatial positions of these sequences;
(3) Any two sequences are not absolutely unrelated. That is, eij never equals zero;
(4) The more Xi and Xj are geometrically similar, the greater eij is;
(5) If Xi and Xj are parallel or Xi0 fluctuates around Xj0 , with the area of the parts
of Xi0 located above Xj0 equal to that of the parts with Xi0 located underneath
Xj0 , then eij ¼ 1;
(6) When one of the observed values of Xi and Xj change, eij also changes
accordingly;
(7) When the lengths of Xi and Xj change, eij also changes;
(8) ejj ¼ eii ¼ 1; and
(9) eij ¼ eji .

5.4.1 Relative and Synthetic Degree of Grey Incidence

Models

5.4.1.1 Relative Degree of Grey Incidence Model

Definition 5.5.1 Let Xi and Xj be sequences of the same length with non-zero initial
values, and Xi0 and Xj0 the initial images of Xi and Xj , respectively. The absolute
degree of grey incidence of Xi0 and Xj0 is referred to as the relative degree of grey
incidence of Xi and Xj , denoted rij (Liu 1991).
82 5 Grey Incidence Analysis Models

This relative degree of incidence is a quantitative representation of the relationship

between the rates of change of sequences Xi and Xj , relative to their initial values. The
closer the rates of change of Xi and Xj are, the greater rij is, and vice versa.
Proposition 5.5.1 Let Xi be a sequence with a non-zero initial value. If Xj ¼ cXi . If
c [ 0 is a constant, then rij ¼ 1:
Proof Assume that Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ; then:

Xj ¼ ðxj ð1Þ; xj ð2Þ;    ; xj ðnÞÞ ¼ ðcxi ð1Þ; cxi ð2Þ;    ; cxi ðnÞÞ:

The initial images of Xi and Xj are as follows:

xi ð1Þ xi ð2Þ xi ðnÞ

Xi0 ¼ Xi =xi ð1Þ ¼ ð ; ;; Þ
xi ð1Þ xi ð1Þ xi ð1Þ

xj ð1Þ xj ð2Þ xj ðnÞ

Xj0 ¼ Xj =xj ð1Þ ¼ ð ; ;; Þ
xj ð1Þ xj ð1Þ xj ð1Þ

cxi ð1Þ cxi ð2Þ cxi ðnÞ xi ð1Þ xi ð2Þ xi ðnÞ

¼ð ; ;; Þ¼ð ; ;; Þ:
cxi ð1Þ cxi ð1Þ cxi ð1Þ xi ð1Þ xi ð1Þ xi ð1Þ

Therefore, Xj0 ¼ Xi0 , so rij ¼ 1:

Proposition 5.5.2 Let Xi and Xj be two sequences of the same length with non-zero
initial values. Additionally, the relative degree of grey incidence rij and the absolute
degree of grey incidence do not have any connections. When eij is relatively large,
rij can be very small; when eij is very small, rij can also be very large.
Proposition 5.5.3 Let Xi and Xj be two sequences of the same length with non-zero
initial values. Then, for any non-zero constants a and b, the relative degree rij0 of
incidence between aXi and bXj is the same as the rij of Xi and Xj .
In fact, the initial images of a Xi and b Xj are equal to those of Xi and Xj ,
respectively. Thus, scalar multiplication does not act in any way under the function
of initialing operators. Hence, rij0 = rij .
Example 5.4.2 Calculate the relative degree of grey incidence r01 for sequences X0
and X1 of Example 5.4.1.
Solution:
Step1 Transform X1 and X0 into the same 1-time-interval sequences

X0 ¼ ðx0 ð1Þ; x0 ð2Þ; x0 ð3Þ; x0 ð4Þ; x0 ð5Þ; x0 ð6Þ; x0 ð7ÞÞ ¼ ð10; 9; 15; 14; 14; 15; 16Þ

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð6Þ; x1 ð7ÞÞ ¼ ð46; 58; 70; 77; 84; 91; 98Þ
5.4 Absolute Degree of Grey Incidence Model 83

Step 2 Calculate the initial images of sequences X0 and X1

X00 ¼ ð1; 0:9; 1:5; 1:4; 1:4; 1:5; 1:6Þ

X10 ¼ ð1; 1:26; 1:52; 1:67; 1:83; 1:98; 2:13Þ

Step 3 Compute the zero-starting point images of sequences X00 and X10

X000 ¼ ðx00 00 00 00 00 00 00
0 ð1Þ; x0 ð2Þ; x0 ð3Þ; x0 ð4Þ; x0 ð5Þ; x0 ð6Þ; x0 ð7ÞÞ
¼ ð0; 0:1; 0:5; 0:4; 0:4; 0:5; 0:6Þ

X100 ¼ ðx00 00 00 00 00 00 00
1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð6Þ; x1 ð7ÞÞ
¼ ð0; 0:26; 0:52; 0:67; 0:83; 0:98; 1:13Þ
     
Step 4 Calculate s00  , s01  , s01  s00 :
 
 0  X 6
1 

s  ¼  00 00
x0 ðkÞ þ x0 ð7Þ ¼ 2
0
 k¼2 2 
 
 0  X 6
1 00 
s  ¼  00
x ðkÞ þ x1 ð7Þ ¼ 3:828
1
 k¼2 1 2 
 
0  X 6
1 

s  s  ¼ 
0
ðx 00
ðkÞ  x 00
ðkÞÞ þ ðx 00
ð7Þ  x 00
ð7ÞÞ ¼ 1:925
1 0
 k¼2 1 0
2 1 0


Step 5 Calculate the relative degree of grey incidence r01 :

1 þ js00 j þ js01 j 6:825

r01 ¼ ¼ 0:78
1 þ js00 j þ js01 j þ js01  s00 j 8:75

Theorem 5.5.1 The relative degree rij of grey incidence satisfies the following
properties:
(1) 0\rij  1;
(2) The value of rij relates only the rates of change of the sequences Xi and Xj with
respect to their individual initial values. It does not relate to the magnitudes of
other entries. In other words, scalar multiplication does not change the rel-
ative degree of grey incidence;
(3) The rates of change of any two sequences are somehow related. That is, rij is
never zero;
84 5 Grey Incidence Analysis Models

(4) The closer the individual rates of change of Xi and Xj with respect to their
initial values, the greater the rij ;
(5) If Xj ¼ aXi , or when the images of zero initial points of the initial images of Xi
and Xj satisfy that Xi00 fluctuates around Xj00 , and if the area of the parts where
Xi00 is located above Xj00 equals that of the parts where Xi00 is located
underneath Xj00 , then rij ¼ 1;
(6) When an entry in Xi or Xj is changed, rij will change accordingly;
(7) When the length of Xi or Xj is changed, rij also changes;
(8) rjj ¼ rii ¼ 1; and
(9) rij ¼ rji .

5.4.1.2 Synthetic Degree of Grey Incidence Model

Definition 5.5.2 Let Xi and Xj be sequences of the same length with non-zero initial
entries, eij and rij be respectively the absolute and relative degrees of incidence
between Xi and Xj , and h 2 ½0; 1: Then the following is referred to as the synthetic
degree of grey incidence between Xi and Xj (Liu 1991):

qij ¼ heij þ ð1  hÞrij ð5:12Þ

The concept of synthetic degree of incidence reflects the degree of similarity

between the zigzagged lines of Xi and Xj , and the closeness between the rates of
change of Xi and Xj with respect to their individual initial values. It is an index that
describes relatively completely the closeness relationship between sequences. In
general, we take h = 0.5. If the focus of a study is the relationship between relevant
absolute quantities, h can take a greater value than 0.5. On the other hand, if the
focus is more on comparison between rates of change, then h can take a smaller
value than 0.5.
Example 5.4.3 Calculate the synthetic degree of grey incidence q01 for sequences
X0 and X1 of Example 5.4.1.
Solution: From Examples 5.4.1 and 5.4.2, we have e01 ¼ 0:5581 and r01 ¼ 0:78: If
h ¼ 0:5:

q01 ¼ he01 þ ð1  hÞr01 ¼ 0:5  0:5581 þ 0:5  0:78 0:669:

We can obtain different q01 values if we take h ¼ 0:2; 0:3; 0:4; 0:6; 0:8;
respectively (see Table 5.1).
5.4 Absolute Degree of Grey Incidence Model 85

Table 5.1 The values of q01 with different h

h 0.2 0.3 0.4 0.6 0.8
q01 0.73562 0.71343 0.69124 0.64686 0.60248

Theorem 5.5.2 The synthetic degree of incidence qij satisfies the following
properties:
(1) 0\qij  1;
(2) The value of qij relates to the individual observed values of sequences Xi and
Xj , as well as to the rates of change of these values with respect to their initial
values;
(3) qij will never be zero;
(4) qij changes along with the values in Xi and Xj ;
(5) When the lengths of Xi and Xj change, so does qij ;
(6) With different h value, qij also varies;
(7) When h ¼ 1; q0i ¼ e0i ; when h ¼ 0; qij ¼ rij ;
(8) qjj ¼ qii ¼ 1; and
(9) qij ¼ qji .

5.4.2 Similarity, Closeness and Three-Dimensional Degree

of Grey Incidence Models

This section focuses on the new models which measure mutual influences and
connections between sequences from two different angles: similarity and closeness.
These new models are much easier to apply to practical problems than traditional
models. Also, three-dimensional degree of grey incidence models can be used to
analyze the relationship among curved surfaces in three-dimensional space and this
is discussed next.

5.4.2.1 Grey Incidence Models Based on Similarity and Closeness

Definition 5.6.1 Let Xi and Xj be sequences of the same length, and si  sj the
same as defined in Proposition 5.4.2. Then, the following equation is referred to as
the similitude degree of grey incidence between Xi and Xj (Liu and Xie 2011):

1
eij ¼ : ð5:13Þ
1 þ jsi  sj j

The concept of similitude degree of incidence is employed to measure the

geometric similarity of the shapes of sequences Xi and Xj . The more similar the
geometric shapes of Xi and Xj , the greater the value of eij , and vice versa.
86 5 Grey Incidence Analysis Models

Definition 5.6.2 Let Xi and Xj be sequences of the same length, and Si  Sj the
same as defined in Proposition 5.4.2. Then, the following expression is referred to
as the closeness degree of grey incidence between Xi and Xj (Liu and Xie 2011):

1
qij ¼ : ð5:14Þ
1 þ jSi  Sj j

The concept of closeness degree of incidence is employed to measure the spatial

closeness of sequences Xi and Xj . The closer the Xi and Xj sequences, the greater the
value of qij , and vice versa.
Proposition 5.6.1 Let Xi and Xj be sequences of 1-time-intervals with the same
length. Then:
 
  1 X
n1
1 

Si  Sj  ¼  ½xi ð1Þ  xj ð1Þ þ ½xi ðkÞ  xj ðkÞ þ ½xi ðnÞ  xj ðnÞ: ð5:15Þ
2 k¼2
2 

It should be noted that the concept of closeness degree of incidence is only

meaningful when sequences Xi and Xj possess similar meanings and identical units.
Otherwise, it does not stand for any practical significance.
Theorem 5.6.1 The similitude degree of incidence eij satisfies the following
properties:
(1) 0\eij  1;
(2) The value of eij is determined only by the geometric shape of sequences Xi and
Xj without any relationship with their relative spatial positions. In other
words, the transform translation of Xi and Xj will not change the value of eij ;
(3) The more geometrically similar the sequences Xi and Xj , the greater the value
of eij , and vice versa;
(4) If Xi and Xj are parallel, or when Xi0 fluctuates around Xj0 , and the area of the
parts where Xi0 is located above Xj0 equals that of the parts where Xi0 is
located beneath Xj0 , then eij ¼ 1;
(5) eii ¼ 1; ejj ¼ 1; and
(6) eij ¼ eji .

Theorem 5.6.2 The closeness degree of incidence qij satisfies the following
properties:
(1) 0\qij  1;
(2) The value of qij is determined not only by the geometric shape of sequences Xi
and Xj , but also by their relative spatial positions. In other words, the
transform translation of Xi and Xj will change the value of qij ;
5.4 Absolute Degree of Grey Incidence Model 87

(3) The closer the sequences Xi and Xj , the greater the qij value, and vice versa;
(4) If Xi and Xj coincide, or Xi fluctuates around Xj , and the area of the parts
where Xi is located above Xj equals that of the parts where Xi is located
beneath Xj , then qij ¼ 1;
(5) qii ¼ 1; qjj ¼ 1; and
(6) qij ¼ qji .

Example 5.6.1 Compute the similitude degrees e12 ; e13 and the closeness degrees of
incidence q12 ; q13 between X1 and X2 , X3 , respectively, given the sequences below:

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð7ÞÞ ¼ ð0:91; 0:97; 0:90; 0:93; 0:91; 0:95Þ

X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð5Þ; x2 ð7ÞÞ ¼ ð0:60; 0:68; 0:61; 0:63; 0:65Þ

X3 ¼ ðx3 ð1Þ; x3 ð3Þ; x3 ð7ÞÞ ¼ ð0:82; 0:90; 0:86Þ

Solution:
Step 1 Let us translate both X2 and X3 into sequences with the same time intervals
as X1 . To this end, consider the following:

1 1
x2 ð4Þ ¼ ðx2 ð3Þ þ x2 ð5ÞÞ ¼ ð0:61 þ 0:63Þ ¼ 0:62
2 2
1 1
x3 ð2Þ ¼ ðx3 ð1Þ þ x3 ð3Þ ¼ ð0:82 þ 0:90Þ ¼ 0:86
2 2
1 1
x3 ð5Þ ¼ ðx3 ð3Þ þ x3 ð7ÞÞ ¼ ð0:90 þ 0:86Þ ¼ 0:88
2 2
1 1
x3 ð4Þ ¼ ðx3 ð3Þ þ x3 ð5ÞÞ ¼ ð0:90 þ 0:88Þ ¼ 0:89
2 2

Thus, we have:

X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð4Þ; x2 ð5Þ; x2 ð7ÞÞ ¼ ð0:60; 0:68; 0:61; 0:62; 0:63; 0:65Þ

X3 ¼ ðx3 ð1Þ; x3 ð2Þ; x3 ð3Þ; x3 ð4Þ; x3 ð5Þ; x3 ð7ÞÞ ¼ ð0:82; 0:86; 0:90; 0:89; 0:88; 0:86Þ

Step 2 Let us translate X1 , X2 , and X3 into sequences of equal time distance. To

this end:

1 1
x1 ð6Þ ¼ ðx1 ð5Þ þ x1 ð7ÞÞ ¼ ð0:91 þ 0:95Þ ¼ 0:93
2 2
88 5 Grey Incidence Analysis Models

1 1
x2 ð6Þ ¼ ðx2 ð5Þ þ x2 ð7ÞÞ ¼ ð0:63 þ 0:65Þ ¼ 0:64
2 2
1 1
x3 ð6Þ ¼ ðx3 ð5Þ þ x3 ð7ÞÞ ¼ ð0:88 þ 0:86Þ ¼ 0:87
2 2

Therefore, the following sequences are all 1-time distance, which means that the
time distances between consecutive entries are all 1.

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð6Þ; x1 ð7ÞÞ

¼ ð0:91; 0:97; 0:90; 0:93; 0:91; 0:93; 0:95Þ

X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð4Þ; x2 ð5Þ; x2 ð6Þ; x2 ð7ÞÞ

¼ ð0:60; 0:68; 0:61; 0:62; 0:63; 0:64; 0:65Þ

X3 ¼ ðx3 ð1Þ; x3 ð2Þ; x3 ð3Þ; x3 ð4Þ; x3 ð5Þ; x3 ð6Þ; x3 ð7ÞÞ

¼ ð0:82; 0:86; 0:90; 0:89; 0:88; 0:87; 0:86Þ

Step 3 Compute the images of zero-starting points provided below

X10 ¼ ðx01 ð1Þ; x01 ð2Þ; x01 ð3Þ; x01 ð4Þ; x01 ð5Þ; x01 ð6Þ; x01 ð7ÞÞ
¼ ð0; 0:06; 0:01; 0:02; 0; 0:02; 0:04Þ

X20 ¼ ðx02 ð1Þ; x02 ð2Þ; x02 ð3Þ; x02 ð4Þ; x02 ð5Þ; x02 ð6Þ; x02 ð7ÞÞ
¼ ð0; 0:08; 0:01; 0:02; 0:03; 0:04; 0:05Þ

X30 ¼ ðx03 ð1Þ; x03 ð2Þ; x03 ð3Þ; x03 ð4Þ; x03 ð5Þ; x03 ð6Þ; x03 ð7ÞÞ
¼ ð0; 0:04; 0:08; 0:07; 0:06; 0:05; 0:04Þ

Step 4 Compute js1  s2 j; js1  s3 j and jS1  S2 j; jS1  S3 j as follows

 
X 6 
 1 0 
j s1  s2 j ¼  ðx1 ðkÞ  x2 ðkÞÞ þ ðx1 ð7Þ  x2 ð7ÞÞ ¼ 0:095
0 0 0
 k¼2 2 
 
X 6 
 1 
j s1  s3 j ¼  ðx01 ðkÞ  x03 ðkÞÞ þ ðx01 ð7Þ  x03 ð7ÞÞ ¼ 0:21
 k¼2 2 
 
X 6 
 1 
j S1  S 2 j ¼  ðx1 ðkÞ  x2 ðkÞÞ þ ðx1 ð7Þ  x2 ð7ÞÞ ¼ 1:91
 k¼2 2 
 
X 6 
 1 
j S1  S3 j ¼  ðx1 ðkÞ  x3 ðkÞÞ þ ðx1 ð7Þ  x3 ð7ÞÞ ¼ 0:375
 k¼2 2 
5.4 Absolute Degree of Grey Incidence Model 89

Step 5 Calculate the similitude degrees e12 ; e13 and closeness degrees q12 ; q13 :

1 1
e12 ¼ ¼ 0:91; e13 ¼ ¼ 0:83
1 þ js1  s2 j 1 þ js1  s3 j

1 1
q12 ¼ ¼ 0:34; q13 ¼ ¼ 0:73
1 þ jS1  S2 j 1 þ jS1  S3 j

Because e12 > e13 , it follows that X2 is more similar to X1 than X3 . Because
q12 < q13 , it follows that X3 is closer to X1 than X2 .
Please note that the focus is on relevant order relations instead of the specific
magnitudes of the values of the degree of grey incidence. This is done through the
degree of GIA model to investigate the mutual influences and connections between
sequences. For instance, let us assume that one must compute the similitude degrees
or closeness degrees of incidence based on Eqs. (5.13) or (5.14). When the absolute

values of the sequence data are relatively large, the values of both si  sj  and
 
Si  Sj  might be large, too, which in turn leads to the resultant similitude and
closeness degrees of incidence being relatively small. This scenario does not have
any substantial impact on the analysis of order relationships. If a particular problem
demands relatively large numerical magnitudes in the degree of incidence, one can
replace the number 1 appearing in the numerators or denominators of Eqs. (5.13)
and (5.14) by a relevant constant, use the grey absolute degree of incidence model,
or use other appropriate models.

5.4.2.2 Three-Dimension Degree of Grey Incidence Models

The degree of GIA model can be generalized to three-dimensional space based on

geometric descriptions of a behavior matrix.
Definition 5.6.3 Assume that X is a two-dimensional system factor, and aij is an
observation value of the system’s behavior at two-dimensional point (i, j), where
1  i  m; 1  j  n. Then the following expression is called the behavior
matrix of system factor X:
2 3
a11 a12    a1n
6 a21 a22    a1n 7
A ¼ ðaij Þmn ¼6
4 
7
   5
am1 am2    amn

For example, if the prices (e.g., opening prices, closing prices, maximum prices,
or minimum prices) of a share have been recorded on different dates, we can obtain
the behavior matrix of the different prices X of the share. The behavior matrix will
90 5 Grey Incidence Analysis Models

Fig. 5.3 The scatter diagram

as behavior matrix

Fig. 5.4 The corresponding

behavioral curved surface

shrink to a behavior sequence if only one share price has been recorded on different
dates.
The scatter diagram in behavior matrix A and the corresponding behavioral
curved surface in three-dimensional space are shown in Figs. 5.3 and 5.4.
Definition 5.6.4 Assume the following behavior matrix of system factor X:

2 3
a11 a12    a1n
6 a21 a22    a1n 7
A ¼ ðaij Þmn ¼6
4 
7
   5
am1 am2    amn

AD ¼ ðaij dÞmn , where D is a matrix operator, aij d ¼ aij  a1j , then D is called a
zero-starting edge operator, AD is called the zero-starting edge image of A, and they
are denoted as AD ¼ A0 ¼ ða0ij Þmn .
5.4 Absolute Degree of Grey Incidence Model 91

Fig. 5.5 The zero-starting

edge image of a behavioral
curved surface

The zero-starting edge curved surface of A is shown in Fig. 5.5.

Definition 5.6.5 Assume that behavior matrices A ¼ ðaij Þmn , B ¼ ðbij Þmn are
matrices of the same type. Then the following formula is called the
three-dimensional absolute degree of grey incidence:

1 þ jsa j þ jsb j
eab ¼ : ð5:16Þ
1 þ jsa j þ jsb j þ jsa  sb j
RR RR RR
This occurs when sa ¼ Da A0 dxdy, sb ¼ Db B0 dxdy, sa  sb ¼ Dab ðA0 
B0 Þdxdy:
Formula (5.16) looks similar to the absolute degree of GIA model shown in
formula (5.11). However, the meaning is different. The meaning of jsi j; jsj j; jsi  sj j
in formula (5.11) is the area of curved edge trapezoids surrounded by axis and Xi0 ,
Xj0 , the zero-starting point curves, and the area of curved edge trapezoid surrounded
by Xi0 and Xj0 . However, the meaning of jsp j; jsq j; jsp  sq j in formula (5.16) is the
volume of curved roof cylinders surrounded by the axis plane and A0, B0, the
curved surface of zero-starting edge, and the volume of curved roof cylinders
surrounded by A0 and B0.
It reflects the relations among behavior matrices by the three-dimensional GIA
model and can easily be calculated by a computer. The three-dimensional GIA
model is seen to have expansive application prospects in many fields such as
multi-criterion decision-making, panel data analysis, image processing, among
others, which include matrices as objects of study.
92 5 Grey Incidence Analysis Models

5.5 Superiority Analysis

Definition 5.7.1 Assume that Y1 ; Y2 ;    ; Ys are a system’s characteristic behavioral

sequences, and X1 ; X2    ; Xm are behavioral sequences of relevant factors with the
same length. Let cij be the degree of grey incidence between Yi and Xj , i ¼ 1; 2;    ; s;
and j ¼ 1; 2;    ; m: Then:

2 3
c11 c12  c1m
6 c21 c22  c2m 7
C ¼ ðcij Þsm ¼6
4
7:
   5
cs1 cs2  csm

This formula is referred to as the grey incidence matrix of the system, where the
ith row is made up of the degrees of grey incidence between the characteristic
sequence Yi ði ¼ 1; 2;    ; sÞ and each of the factor sequences X1 ; X2    ; Xm ; and the
jth column consists of the degrees of grey incidence between each of the charac-
teristic sequences Y1 ; Y2 ;    ; Ys and Xj ðj ¼ 1; 2;    ; mÞ: We can analyze both the
superiority of a system’s characteristic behavioral variables or the behavioral
variables of relevant factors.
Definition 5.7.2 Assume that Y1 ; Y2 ;    ; Ys are a system’s characteristic behavioral
sequences, X1 ; X2    ; Xm are behavioral sequences of relevant factors, and
C ¼ ðcij Þsm is the grey incidence matrix. If there are k; i 2 f1; 2;    ; sg such that
ckj  cij , j ¼ 1; 2;    ; m; then the system’s characteristic variable Yk is said to be
more favorable than the system’s characteristic variable Yi , written as Yk
Yi .
If 8 i ¼ 1; 2;    ; s; i 6¼ k; Yk
Yi always holds true, then Yk is said to be the
most favorable characteristic variable.
Definition 5.7.3 Assume that Y1 ; Y2 ;    ; Ys are a system’s characteristic behavioral
sequences, X1 ; X2    ; Xm are behavioral sequences of relevant factors, and
C ¼ ðcij Þsm is the grey incidence matrix. If there are l; j 2 f1; 2;    ; mg such that
cil  cil , i ¼ 1; 2;    ; s; then we say that the system’s factor Xl is more favorable
than factor Xj , written as Xk
Xi .
If 8 j ¼ 1; 2;    ; m; j 6¼ l; Xl
Xj always holds true, then Xl is said to be the
most favorable factor.
Definition 5.7.4 Assume that Y1 ; Y2 ;    ; Ys are a system’s characteristic behavioral
sequences, X1 ; X2    ; Xm are behavioral sequences of relevant factors, and
C ¼ ðcij Þsm is the grey incidence matrix
Pm Pm
(1) If there are k; i 2 f1; 2;    ; sg satisfying j¼1 cij  j¼1 cij , then the sys-
tem’s characteristic variable Yk is said to be more quasi-favorable than Yi ,
which is denoted as Yk
Yi .
5.5 Superiority Analysis 93
Pm Pm
(2) If there are l; j 2 f1; 2;    ; mg satisfying i¼1 cil  i¼1 cij , then the sys-
tem’s factor Xl is more quasi-favorable than Xj , which is denoted as Xl
Xj .

Definition 5.7.5 Assume that Y1 ; Y2 ;    ; Ys are a system’s characteristic behavioral

sequences, X1 ; X2    ; Xm are behavioral sequences of relevant factors, and
C ¼ ðcij Þsm is the grey incidence matrix.
(1) If there is k 2 f1; 2;    ; sg such that 8 i ¼ 1; 2;    ; s; i 6¼ k; Yk
Yi , then
the system’s characteristic variable Yk is said to be quasi-preferred.
(2) If there is l 2 f1; 2;    ; mg such that 8 j ¼ 1; 2;    ; m; j 6¼ l; Xl
Xj , then
the system’s factor Xl is said to be quasi-preferred.

Proposition 5.7.1 In a system of S characteristic variables and m relevant factors,

there may not be a most favorable characteristic variable and a most favorable
factor. However, there must be quasi-preferred characteristic variable and factor.
Example 5.7.1 The formulas below are system’s characteristic behavioral sequences.

Y1 ¼ ð170; 174; 197; 216:4; 235:8Þ

Y2 ¼ ð57:55; 70:74; 76:8; 80:7; 89:85Þ

Y3 ¼ ð68:56; 70; 85:38; 99:83; 103:4Þ

The formulas below are behavioral sequences of relevant factors.

X1 ¼ ð308:58; 310; 295; 346; 367Þ

X2 ¼ ð195:4; 189:9; 189:2; 205; 222:7Þ

X3 ¼ ð24:6; 21; 12:2; 15:1; 14:57Þ

X4 ¼ ð20; 25:6; 23:3; 29:2; 30Þ

X5 ¼ ð18:98; 19; 22:3; 23:5; 27:655Þ

Try and analyze the superiority of the system’s characteristic behavioral vari-
ables and the superiority of the behavioral variables of relevant factors.
Solution: We analyze the superiority of the system’s characteristic behavioral
sequences and the behavioral sequences of relevant factors by absolute degree of
GIA model.
(1) Find the matrix of absolute degree of grey incidence, calculate the images of
zero-starting point for all the system’s characteristic behavioral sequences as
well as the behavioral sequences of relevant factors as follows:
94 5 Grey Incidence Analysis Models

Y10 ¼ ð0; 4; 27; 46:4; 65:8Þ

Y20 ¼ ð0; 13:19; 19:25; 23:15; 32:3Þ

Y30 ¼ ð0; 1:44; 16:82; 31:27; 34:84Þ

X10 ¼ ð0; 1:42; 13:58; 37:42; 58:42Þ

X20 ¼ ð0; 5:5; 8:2; 9:6; 27:3Þ

X30 ¼ ð0; 3:6; 12:4; ; 9:5; 10:03Þ

X40 ¼ ð0; 5:6; 3:3; 9:2; 10Þ

X50 ¼ ð0; 0:02; 3:32; 4:52; 8:675Þ

For the system’s characteristic behavioral variable Y1, we have:

   
  X 4
1  
  1 
 sy  ¼  y 0
ðkÞ þ y 0
ð5Þ  ¼ 4 þ 27 þ 46:4 þ  65:8  ¼ 110:3
1
 k¼2 1
2 1
  2 

   
X 4   
 1   1
jsx1 j ¼  x ðkÞ þ x1 ð5Þ ¼ 1:42 þ ð13:58Þ þ 37:42 þ  58:42 ¼ 54:47
0 0
 k¼2 1 2  2
 
  X 4
1 

sy  sx  ¼  ðy 0
ðkÞ  x 0
ðkÞÞ þ ðy 0
ð5Þ  x 0
ð5ÞÞ  ¼ 55:9
1 1
 k¼2 1 1
2 1 1

 
1 þ sy1  þ jsx1 j 1 þ 110:3 þ 54:47
e11 ¼    ¼ ¼ 0:748
1 þ sy  þ jsx j þ sy  sx  1 þ 110:3 þ 54:47 þ 55:9
1 1 1 1

   
X 4
1 0   
 1
j s x2 j ¼  x2 ðkÞ þ x2 ð5Þ ¼ ð5:5Þ þ ð8:2Þ þ 9:6 þ  27:3 ¼ 9:55
0
 k¼2 2  2
 
  X 4
1 

sy  sx  ¼  ðy01 ðkÞ  x02 ðkÞÞ þ ðy01 ð5Þ  x02 ð5ÞÞ ¼ 100:75
1 2
 k¼2 2 
 
1 þ sy1  þ jsx2 j 1 þ 110:3 þ 9:55
e12 ¼    ¼ ¼ 0:545
  
1 þ sy þ j sx j þ sy  sx  1 þ 110:3 þ 9:55 þ 100:75
1 2 1 2
5.5 Superiority Analysis 95

Similarly:
 
1 þ sy1  þ jsx3 j
e13 ¼     ¼ 0:502
1 þ sy  þ j sx j þ sy  sx 
1 3 1 3

 
1 þ sy1  þ jsx4 j
e14 ¼     ¼ 0:606
1 þ sy  þ j sx j þ sy  sx 
1 4 1 4

 
1 þ sy1  þ jsx5 j
e15 ¼     ¼ 0:557
1 þ sy  þ j sx j þ sy  sx 
1 5 1 5

For the system’s characteristic behavioral variable Y2, Y3, we have:

e21 ¼ 0:880; e22 ¼ 0:570; e23 ¼ 0:502; e24 ¼ 0:663; e25 ¼ 0:588

e31 ¼ 0:907; e32 ¼ 0:574; e33 ¼ 0:503; e34 ¼ 0:675; e35 ¼ 0:594

Therefore, we have the matrix of absolute degree of grey incidence as follows:

2 3
e11 e12 e13 e14 e15
A ¼ ðeij Þ ¼ 4 e21 e22 e23 e24 e25 5
2 e31 e32 e33 e34 e35 3
0:748 0:545 0:502 0:606 0:557
¼ 4 0:880 0:570 0:502 0:663 0:588 5
0:907 0:574 0:503 0:675 0:594

(2) Calculate the matrix of relative degree of grey incidence. Calculate the initial
images:

Yi0 ði ¼ 1; 2; 3Þ and Xj0 ðj ¼ 1; 2; 3; 4; 5Þ of Yi ði ¼ 1; 2; 3Þ and Xj ðj ¼ 1; 2; 3; 4; 5Þ:

Then find the images of zero-starting point for all system’s characteristic
behavioral sequences Yi ði ¼ 1; 2; 3Þ and the behavioral sequences of relevant fac-
tors Xj ðj ¼ 1; 2; 3; 4; 5Þ:

Yi00 ði ¼ 1; 2; 3Þ and Xj00 ðj ¼ 1; 2; 3; 4; 5Þ of Yi0 ði ¼ 1; 2; 3Þ and Xj0 ðj ¼ 1; 2; 3; 4; 5Þ

From:
 
  X 4
1 00 
0  00
syi  ¼  y ðkÞ þ yi ð5Þ; i ¼ 1; 2; 3
 k¼2 i 2 
96 5 Grey Incidence Analysis Models

 
  X 4
1 00 
0  00
sxj  ¼  x ðkÞ þ xj ð5Þ; j ¼ 1; 2; 3; 4; 5
 k¼2 j 2 
 
  X 4 
0 0   00 00 1 00 00 
 syi  s xj  ¼  ðyi ðkÞ  xj ðkÞÞ þ ðyi ð5Þ  xj ð5ÞÞ; i ¼ 1; 2; 3; ; j
 k¼2 2 
¼ 1; 2; 3; 4; 5
   
   
1 þ s0yi  þ s0xj 
rij ¼       ; i ¼ 1; 2; 3; j ¼ 1; 2; 3; 4; 5;
     
1 þ s0yi  þ s0xj  þ s0yi  s0xj 

we have:

r11 ¼ 0:7945; r12 ¼ 0:7389; r13 ¼ 0:6046; r14 ¼ 0:8471; r15 ¼ 0:9973

r21 ¼ 0:6937; r22 ¼ 0:6571; r23 ¼ 0:5837; r24 ¼ 0:9738; r25 ¼ 0:8271

r31 ¼ 0:7300; r32 ¼ 0:6866; r33 ¼ 0:6101; r34 ¼ 0:9444; r35 ¼ 0:8884

Therefore, we have the matrix of relative degree of grey incidence as follows:

2 3
r11 r12 r13 r14 r15
B ¼ 4 r21 r22 r23 r24 r25 5
2 r31 r32 r33 r34 r35 3
0:7945 0:7389 0:6046 0:8471 0:9973
¼ 4 0:6937 0:6571 0:5837 0:9738 0:8271 5
0:7300 0:6866 0:6101 0:9444 0:8884

(3) Compute the matrix of synthetic degree of grey incidence. If h ¼ 0:5; we

have:

C ¼ hA þ ð1  hÞB ¼ ðheij þ ð1  hÞrij Þ ¼ ðqij Þ

2 3
q11 q12 q13 q14 q15
6 7
¼ 4 q21 q22 q23 q24 q25 5
q31 q32 q33 q34 q35
2 3
0:7713 0:6420 0:5533 0:7266 0:7772
6 7
¼ 4 0:7869 0:6136 0:5429 0:8184 0:7076 5
0:8185 0:6303 0:5566 0:8097 0:7412
5.5 Superiority Analysis 97

(4) Analysis and discussion. In matrix A of absolute degree of incidence, the rows
of A satisfy the following formula:

e3j [ e2j  e1j ; j ¼ 1; 2; 3; 4; 5:

Therefore, we have Y3
Y2
Y1 . That is, Y3 is the most favorable characteristic
variable, Y2 is the second, and Y1 the least favorable characteristic variable. All
columns of A satisfy:

ei1 [ ei4 [ ei5 [ ei2 [ ei3 ; i ¼ 1; 2; 3:

Therefore, we have:

X1
X4
X5
X2
X3 :

That is, X1 is the most favorable factor, X4 the second, X5 the third, X2 the fourth,
and X3 the least.
From the matrix B of relative degree of incidence, it can be seen that because the
elements of B satisfy
ri4 [ ri1 [ ri2 [ ri3 ; i ¼ 1; 2; 3
ri5 [ ri1 [ ri2 [ ri3 ; i ¼ 1; 2; 3

Thus, we can conclude that:

X4
X1
X2
X3 ; X5
X1
X2
X3 :

Hence, X3 is the most unfavorable factor of the system. Further, let us consider
the following:

X
5 X
5 X
5
r1j ¼ 3:9824 [ r3j ¼ 3:8595 [ r2j ¼ 3:7354:
j¼1 j¼1 j¼1

Thus, we can conclude that Y1

Y3
Y2 , that is, Y1 is the quasi-preferred char-
acteristic. Also, given that:

X
3 X
3 X
3 X
3 X
3
ri4 ¼ 2:7653 [ ri5 ¼ 2:7128 [ ri1 ¼ 2:2182 [ ri2 ¼ 2:0826 [ ri3 ¼ 1:7984;
i¼1 i¼1 i¼1 i¼1 i¼1

we have:

X4
X5
X1
X2
X3 :

That is, X4 is the quasi-preferred factor, X5 the next, and X3 the most unfavorable
factor.
98 5 Grey Incidence Analysis Models

On matrix C of synthetic degree of incidence, it can be seen that the elements of

C satisfy:

qi1 [ qi2 [ qi3 ; qi4 [ qi2 [ qi3 ; qi5 [ qi2 [ qi3 ; i ¼ 1; 2; 3:

Therefore, we have:

X1
X2
X3 ; X4
X2
X3 ; X5
X2
X3 :

That is, X3 is the least preferred factor. We further consider the following:

X
5 X
5 X
5
q3j ¼ 3:5563 [ q1j ¼ 3:4704 [ q2j ¼ 3:4694:
j¼1 j¼1 j¼1

Thus,

Y3
Y1
Y2 :

That is, Y3 is the quasi-preferred characteristic variable. Also, based on:

X
3 X
3 X
3 X
3 X
3
qi1 ¼ 2:3767 [ qi4 ¼ 2:3547 [ qi5 ¼ 2:226 [ qi2 ¼ 1:8859 [ qi3 ¼ 1:6528;
i¼1 i¼1 i¼1 i¼1 i¼1

it follows that:

X1
X4
X5
X2
X3 :

Therefore, X1 is the quasi-preferred factor, X4 the next, X5 is more favorable than

X2 , and X3 is the most unfavorable factor.
When investigating practical problems, the analyses of the three incidence orders
may not provide cohesive conclusions. This is because the absolute incidence order
looks at the relationship between absolute quantities, the relative incidence order
focuses on the rates of change with respect to the initial values of the observed
sequences, while the synthetic incidence order combines both the relationships
between absolute quantities and rates of change. When considering the background
of the problem of concern, we can choose one of the incidence orders. For parsi-
mony purposes, after a particular grey incidence operator is applied to the system’s
characteristic behavioral sequences and relevant factor sequences, one only needs to
employ the absolute incidence order to the processed data.
5.5 Superiority Analysis 99

5.5.1 Practical Application

Through the example below (Chen and Liu 2007), we look at how to apply GIA
models to analyze the time difference of economic indices.
Example 5.8.1 In order to effectively monitor the performance of macro-economic
systems and provide timely warnings, there is a need to investigate the time rela-
tionship of various economic indices with respect to economic cycles in terms of
their peaks and valleys. In order to do so, questions such as the following must be
addressed: Which indices can provide warning ahead of time? Which indices would
be synchronic with the evolution of economic systems? And which indices tend to
lag behind economic development? In other words, there is a need to divide eco-
nomic indices into three classes: leading indicators, synchronic indices, and stag-
nant representations. To this end, grey incidence analysis is an effective method for
classifying economic indices.
Through careful research and analysis, we selected the following 8 major classes
and 17 criteria as indices for economic performance:
(1) The Energy and raw materials class: the total production of energy;
(2) The investments class: the total investment in real estate;
(3) The production class: increase in industry output, increase in light industry
output, increase in heavy industry output;
(4) The revenue class: national income, national expenditure;
(5) The currency and credit class: currency in circulation, savings at various
financial institutions, amount of loans issued by financial institutions, cash
payout in the form of salary and wages, net amount of currency in circulation;
(6) The consumption class: the gross retail amount of the society;
(7) The foreign trade class: gross amount of imports, gross amount of exports,
direct investments by foreign entities; and
(8) The commodity prices class: the consumer price index.
By applying the following standards, we classify the previous criteria into three
classes: leading indicators, synchronic indices, and stagnant representations. The
standards for determining leading indicators are as follows:
(1) The indicated appearance of economic cyclic peaks needs to be at least three
months ahead of their actual occurrence. Such leading relationship must be
relatively stable with few exceptions;
(2) Indicated cycles and historical cycles are nearly one-to-one corresponded to
each other. Also, for the most recent three economic cycles, the indicated
cycles must be at least two times ahead of the actual occurrences with at least
3 months of lead time; and
(3) The economic characteristics of the indices provide relatively definite and
clear leading relationships with respect to the background economic cycles.
100 5 Grey Incidence Analysis Models

The standards for determining both synchronic indices and stagnant represen-
tations are similar to those outlined above. However, for synchronic indices the
time differences between the indicated appearances and the actual occurrences of
economic cycles must be within plus and minus 3 months, while for stagnant
representations the indicated appearances of economic cycles are behind the actual
occurrences by at least 3 months.
As expected, in practice it is almost impossible to find an index that meets all the
stated standards. Therefore, based on the recorded reference cycles, we look for the
statistical indices that meet the previously stated standards as closely as possible. In
reality, a leading indicator can sometimes lag behind actual economic development,
while an identified stagnant representation can also provide good lead-time in its
forecast of a specific economic evolution. Similar scenarios also occur with regard
to synchronic indices. However, theoretically, if the index is leading the actual
occurrences among the one-to-one correspondences between an index and the
actually recorded cycles over 2/3 of times, then we treat such an index as leading.
Similar treatments are applied to synchronic indices and stagnant representations.
Given that the increase in industry output has played a significant role in the
Chinese economy, as a synchronic index it has high quality. Therefore, it can be
employed as the basic index in our grey incidence analysis. We will compute not
only the absolute degree of grey incidence between each criterion and the increase
in industry output, but also the absolute degrees of grey incidence of the other 16
criteria with their data translated 1–12 months along the time axis either left or
right. When data are translated to the left, the months will take negative values;
when translated to the right, the months will take positive values. The amount of
horizontal translation is denoted by L. That is, we compute the absolute degrees of
grey incidence between all 16 individual criteria, excluding that of increase in
industry output, and that of increase in industry output for L = −12, …, 12. For
each L-value, we order the obtained absolute degrees of grey incidence from the
smallest to the largest, with the criterion listed in the front chosen as candidate
criterion for that specific L-value. For instance, when L = 0, the absolute degrees of
grey incidence of the criteria are listed in Table 5.2.
Synchronic indices should be selected from those with large absolute degrees of
grey incidence, because large degrees of incidence indicate that these criteria have
greater similarities in comparison with that of increase in industry output, which we
employ as the basic standard of the Chinese economic cycles. However, we still do
not have theoretical evidence to support that an index with large absolute degree of
grey incidence must be synchronic. To this end, we also need to consider whether
or not the related absolute degrees of grey incidence will be even greater when
L 6¼ 0. If when L = 0 the value of the absolute degree of grey incidence of a certain
index is ranked in the front, and if when L = −4 its value is even greater, it means
that after this index is translated to 4 months earlier, it is more similar to the pattern
of the increase in industry output. Thus, in this case, this specific index can be seen
as one leading the economic cycle by as much as about 4 months. By using these
two standards, we can not only classify indices as synchronic, leading, or stagnant,
but also specify the amount of leading or staggering time.
5.5 Superiority Analysis 101

Table 5.2 The absolute degrees of grey incidence of the criteria when L = 0
Index Absolute degree of Index Absolute degree of
grey incidence grey incidence
Increase in heavy 0.979810 National income 0.559540
industry output
Increase in light 0.972655 Gross amount of 0.544870
industry output exports
Gross retail amount 0.862105 Total production of 0.541044
energy
Cash payout as 0.789278 Net amount of currency 0.525936
salaries in circulation
Currency in 0.753681 Loans issued by 0.507958
circulation financial institutions
Total investment in 0.726366 Savings at financial 0.505226
real estate institutions
GROSS amount of 0.598248 Consumer price index 0.500173
imports
National expenditure 0.566914 Direct investments by 0.500002
foreign entities

Table 5.3 The absolute degrees of “cash payout as salaries” when L 6¼ 0

L Absolute degree of grey incidence L Absolute degree of grey incidence
−12 0.664615 1 0.877090
−11 0.705983 2 0.867859
−10 0.733564 3 0.857366
−9 0.752740 4 0.832260
−8 0.753598 5 0.825027
−7 0.732221 6 0.806787
−6 0.723942 7 0.806782
−5 0.731232 8 0.820384
−4 0.742249 9 0.803771
−3 0.752628 10 0.806649
−2 0.770216 11 0.805679
−1 0.800838 12 0.836308
0 0.789278

When L = 0, the index of “cash payout as salaries” is ranked relatively in the

front. Therefore, it is a natural candidate for being a synchronic indicator. When the
L–value changes, the relevant changes in its absolute degrees of grey incidence are
given in Table 5.3.
From Table 5.3, it follows that when L = 1, the absolute degree of grey incidence
reaches its maximum. Therefore, this specific index should be seen as one that is
lagging the economic cycle by as much as 1 month. An index which is leading or
102 5 Grey Incidence Analysis Models

Table 5.4 The absolute degrees of “gross retail amount”

L Absolute degree of grey incidence L Absolute degree of grey incidence
−12 0.914466 1 0.856944
−11 0.915117 2 0.866789
−10 0.918527 3 0.876758
−9 0.887243 4 0.882430
−8 0.888258 5 0.889590
−7 0.928151 6 0.895899
−6 0.948684 7 0.900899
−5 0.939351 8 0.900130
−4 0.923900 9 0.895977
−3 0.909621 10 0.894374
−2 0.884610 11 0.892662
−1 0.846814 12 0.889532
0 0.862105

lagging no more than 2 months is usually seen as synchronic. However, if it exceeds

this range of time it will be treated as either a leading or staggering index.
As a second example, the computational results for the index of “gross retail
amount” are provided in Table 5.4.
From Table 5.4, it can be seen that when L = −6 the absolute degree of grey
incidence of the particular index reaches its maximum. Therefore, it can be seen as
a leading indicator. By using this method, we can compute the L–values corre-
sponding to the maximum absolute degrees of grey incidence of each of the indices
of our interest. The results are listed in Table 5.5.

Table 5.5 L–values corresponding to maximum absolute degrees of incidence of the indices of
our interest
Index L Absolute Index L Absolute
degree degree
Currency in circulation −6 0.983452 National income +12 0.718998
Increase in heavy 0 0.979810 Gross amount of −9 0.606556
industry output imports
Increase in light 0 0.972655 Gross amount of +10 0.560054
industry output exports
gross retail amount −6 0.948684 Total production of −6 0.555035
energy
Cash payout as salaries +1 0.877090 Direct investments by −11 0.510016
foreign entities
National expenditure +12 0.800533 Loans issued by −5 0.508375
financial institutions
Net amount of +8 0.796688 Savings at financial −6 0.505588
currency in circulation institutions
Total investment in −11 0.769778 Consumer price index +11 0.503235
real estate
5.5 Superiority Analysis 103

Table 5.6 Classifications of leading, synchronic, and stagnant indices

Leading index Synchronic index Stagnant index
Energy and Total production of
raw materials energy (−6)a
Investment Total investment in real
estate (−11)
Production Increase in light
industry output (0)
Increase in heavy
industry output (0)
Finance National income (+12)
National expenditure
(+12)
Currency and Currency in circulation Cash payout as Net amount of
credit (−6) salaries (+1) currency in circulation
Savings at financial (+8)
institutions (−6)
Loans issued by financial
institutions (−5)
Consumption Gross retail amount (−6)
Foreign trade Gross amount of imports Gross amount of
(−9) exports (+10)
Direct investments by
foreign entities (−11)
Commodity Consumer price index
price (+11)
a
Numbers in parentheses stand for the time difference between indicated cycles and reference
cycles

Table 5.5 indicates that we can classify the 16 indices of the eight major classes
into three classes as leading, synchronic, and stagnant indices, as shown in
Table 5.6.
Chapter 6
Grey Clustering Evaluation Models

6.1 Introduction

Grey clustering is a method developed for classifying observation indices or

observation objects into definable classes using grey incidence matrices or grey
possibility functions. Each cluster can be seen as a set consisting of all the
observational objects of a kind. When investigating practical problems, it is often
the case that each observational object possesses quite a few characteristic indices,
which are difficult to accurately classify. Depending on the objects to be clustered,
grey clustering can be based on two methods: clustering using GIA models, and
clustering using grey possibility functions. The first method is mainly applied to
group the same kinds of factors into their individual categories, so that a compli-
cated system can be simplified. By using the clustering method of grey incidence,
we can examine whether or not some of the factors under consideration really
belong to the same kind. This allows a synthetic index of these factors, or one of
these factors, to be used to represent all factors without losing any part of the
available information carried by such factors. This problem regards the selection of
variables to be used in the study of a system. Before conducting a large-scale
survey, which generally costs a lot of money and man power, by using the clus-
tering method of grey incidence on a typical sample data, one can reduce the
amount of data collection to a minimal level by eliminating the unnecessary vari-
ables so that tangible savings can be achieved.
The clustering method based on grey possibility functions is mainly used for
checking whether or not the observational objects belong to pre-determined classes
so that they can be treated differently. In practice, we need to set the possibility
functions and the weights for different criterion according to the corresponding
clustering index and the grey classes we intend to partition if using the clustering
method based on grey possibility functions.
Grey clustering evaluation models using possibility functions are used widely for
uncertain systems analysis. For the past three decades, much research on modeling

© Springer Science+Business Media Singapore 2017 105

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_6
106 6 Grey Clustering Evaluation Models

techniques has been done, and new research results emerge constantly. For
example, Professor Julong Deng (1985) has proposed the variable weight grey
clustering model, while Professor Liu (1993) has proposed the fixed weight grey
clustering evaluation model, the grey clustering evaluation model using end-point
triangular possibility functions, the grey cluster evaluation model using center-point
triangular possibility functions, among others. These models are all used widely.
Grey variable weight clustering model is applicable to the problems with criteria
that have the same meanings and dimensions. When the criteria for clustering
involve different meanings and dimensions, the fixed weight grey clustering eval-
uation model and grey clustering evaluation model using triangular possibility
functions are suitable. In particular, compared with the variable weight grey clus-
tering and fixed weight grey clustering models, the grey clustering evaluation
model using triangular possibility functions is more suitable for problems of poor
information clustering evaluation. The grey cluster evaluation model using
end-point triangular possibility functions is suitable for situations where all grey
boundaries are clear, but where the most likely points belonging to each grey class
are unknown. Conversely, the grey clustering evaluation model using center-point
triangular possibility functions is suitable for problems where it is easy to judge the
most likely points belonging to each grey class, but where the grey boundaries are
unclear.
Additionally, two evaluation models are based on triangular possibility function
for moderate measure. Further, Xinping Xiao, Hejin Xiong, Fenyi Dong, Lingling
Pei, Weiguo Xu, and others have improved and optimized grey clustering evalu-
ation models from different perspectives. Furthermore, Zhang Qishan has studied
the measurement problem of Grey Characteristics of Grey Clustering Result. The
author has investigated the relation between a grey clustering analysis result and the
entropy of the weight sequence, and proposed a measure method for the grey
characteristics of a grey clustering analysis result. Weijie Zhou and Yaoguo Dang
have constructed the integral mean value function of an interval grey number set,
and given the grey interval number variable and certain weight grey clustering
assessment models. Yong Liu has combined the grey cluster model based on
center-point triangular possibility function with the variable precision dominance
based on rough set approach; through this work the author has built up a new mixed
method. Fang Peng and Guoping Wu have developed a new method for cover layer
quantitative evaluation based on grey clustering analysis. The authors have applied
this method to evaluate 3 main exploration areas and 4 sets of mudstone in the
southeast basin of Hainan Province, totaling 12 kinds of cap rock objects. Their
conclusions met with the exploration results.
In this chapter, two novel grey cluster evaluation models based on mixed
center-point triangular possibility functions and mixed end-point triangular possi-
bility functions are put forward. These new grey clustering models based on mixed
possibility functions are especially applicable to evaluation and classification of
poor information objects, and have broad application prospects.
6.2 Grey Incidence Clustering Model 107

6.2 Grey Incidence Clustering Model

Definition 6.2.1 Assume that there are n observational objects. For each object the
data of m attribute indexes are collected, producing the following sequences:

X1 ¼ ðx1 ð1Þ ; x1 ð2Þ;    ; x1 ðnÞÞ

X2 ¼ ðx2 ð1Þ ; x2 ð2Þ ;    ; x2 ðnÞÞ



Xm ¼ ðxm ð1Þ ; xm ð2Þ ;    ; xm ðnÞÞ

Then, for all i\j; i; j ¼ 1; 2; . . .; m; calculate eij , the absolute degree of grey
incidence between Xi and Xj , so that we have the following upper triangular matrix A:
0 1
e11 e12  e11
B e22  e2m C
B C
A¼B .. .. C
@ . . A
emm

A is referred to as the incidence matrix of the attribute indexes, where

eii ¼ 1;i ¼ 1; 2; . . .; m: For a chosen threshold value r 2 ½0; 1; which in general
satisfies r [ 0:5; if eij  r; i 6¼ j; the variables Xj and Xi are seen as the same
attribute.
Definition 6.2.2 The classification of the attribute indexes with the chosen value
r is referred to as the r-classification by degree of grey incidence.
When studying a specific problem, the particular value r is determined based on
the circumstances involved. The closer the r is to 1, the finer the classification and
the fewer the variables in each class. Conversely, the smaller the r, the coarser the
classification and the greater the number of variables in each class.
Example 6.2.1 The talent search committee of a firm has proposed 15 candidate
recruitment criteria as follows:
1. Impression of overall application package;
2. Academic abilities;
3. Likability by others;
4. Level of self-confidence;
5. Intelligence;
6. Honesty;
7. Ability to sell;
8. Experience;
9. Motivation;
10. Ambition;
108 6 Grey Clustering Evaluation Models

Table 6.1 The Scores of 9 observational objects

Objects attributes 1 2 3 4 5 6 7 8 9
X1 6 9 7 5 6 7 9 9 9
X2 2 5 3 8 8 7 8 9 7
X3 5 8 6 5 8 6 8 8 8
X4 8 10 9 6 4 8 8 9 8
X5 7 9 8 5 4 7 8 9 8
X6 8 9 9 9 9 10 8 8 8
X7 8 10 7 2 2 5 8 8 5
X8 3 5 4 8 8 9 10 10 9
X9 8 9 9 4 5 6 8 9 8
X10 9 9 9 5 5 5 10 10 9
X11 7 10 8 6 8 7 9 9 9
X12 7 8 8 8 8 8 8 9 8
X13 5 8 6 7 8 6 9 9 8
X14 7 8 8 6 7 6 8 9 8
X15 10 10 10 5 7 6 10 10 10

11. Presentation skills;

12. Ability to comprehend instructions;
13. Potential for future growth;
14. Interpersonal skills; and
15. Adaptability.
Members of the committee admit that some of these 15 criteria can overlap and
hope that through the study of a sample of a few data points, these 15 criteria can be
classified into fewer categories. By using the scoring method to quantify the cri-
teria, 9 observational objects have been scored according to each of the criteria.
Table 6.1 gives the scores, where Oi stands for the ith object, i = 1, 2, …, 9.
To calculate the absolute degree of grey incidence eij of Xi and Xj for all
i  j; i;j ¼ 1; 2    ; 15; we obtained the upper triangular matrix A as shown in
Table 6.2.
We divided the 15 criteria into different classes based on Table 6.2, where the
value of threshold r can be different based on the requirements involved. For
example, if we take r = 1, all 15 criteria above belong to their own classes with
each in its own class. If we take r = 0.80, then we check the values in Table 6.2,
row by row, and pick out all the values of eij which are greater than 0.80. Thus, we
have:

e1;3 ¼ 0:88; e1;11 ¼ 0:90; e1;12 ¼ 0:88; e1;13 ¼ 0:80; e2;8 ¼ 0:99

e3;11 ¼ 0:80; e3;13 ¼ 0:90; e6;11 ¼ 0:84; e6;12 ¼ 0:86; e6;14 ¼ 0:81
Table 6.2 The matrix of grey incidence of attribute indexes
X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15
X1 1 0.66 0.88 0.52 0.58 0.77 0.51 0.66 0.51 0.51 0.9 0.88 0.8 0.67 0.51
X2 1 0.72 0.51 0.53 0.59 0.5 0.99 0.51 0.51 0.63 0.62 0.77 0.55 0.51
X3 1 0.56 0.7 0.51 0.72 0.51 0.51 0.51 0.8 0.78 0.9 0.63 0.51
X4 1 0.56 0.53 0.58 0.51 0.69 0.62 0.52 0.52 0.51 0.54 0.6
6.2 Grey Incidence Clustering Model

X5 1 0.65 0.51 0.53 0.53 0.52 0.61 0.61 0.55 0.75 0.52
X6 1 0.51 0.59 0.52 0.52 0.84 0.86 0.66 0.81 0.51
X7 1 0.5 0.7 0.83 0.51 0.51 0.51 0.51 0.89
X8 1 0.51 0.51 0.63 0.62 0.77 0.55 0.51
X9 1 0.81 0.52 0.52 0.51 0.53 0.76
X10 1 0.51 0.51 0.51 0.52 0.92
X11 1 0.97 0.74 0.71 0.51
X12 1 0.73 0.72 0.51
X13 1 0.6 0.51
X14 1 0.52
X15 1
109
110 6 Grey Clustering Evaluation Models

e7;10 ¼ 0:83; e7;15 ¼ 0:89; e9;10 ¼ 0:81; e10;15 ¼ 0:92; e11;12 ¼ 0:97

Therefore, we know that X3, X11, X12, and X13 belong to the same class as X1;
X8 belong to the same class as X2; X11 and X13 belong to the same class as X3; X11,
X12, and X14 belong to the same class as X6; X10 and X15 belong to the same class
as X7; X10 belong to the same class as X9; X15 belong to the same class as X10; and
X12 belong to the same class as X11.
Let each class be represented with the criterion with the minimum index con-
tained in the class, and combine the classes containing X6 and X11, respectively,
with the class containing X1. Put X9 and X10 into the class containing X7, and treat
X4 and X5 as individual classes. Then, we have obtained a classification of the 15
attribute criteria for our shortened list as follows:

fX1 ; X3 ; X6 ; X11 ; X12 ; X13 ; X14 g fX2 ; X8 g; fX4 g; fX5 g;

fX7 ; X9 ; X10 ; X15 g

Here, the class of {X1 , X3 , X6 , X11 , X12 , X13 , X14 } including the attribute criteria
such as impression of overall application package, likability by others, honesty,
presentation skills, ability to comprehend instructions, potential for future growth,
and interpersonal skills, all of which direct impression, can be obtained through the
application form or interviews. These attribute criteria can be replaced by one
synthetic impression attribute criterion because all these attribute criteria correlate
and it is difficult to be separate them completely. The class of {X2 , X8 } includes two
attribute criteria, namely academic abilities and experience, which can be evaluated
through investigation and understanding of the academic research and practical
work accomplished by the candidate. The class of {X7 , X9 , X10 , X15 } includes four
attribute criteria, namely ability to sell, motivation, ambition, and adaptability,
which can be judged synthetically by investigating the learning and working
background of the candidate. Special investigation is required for assessment of the
attribute criterion level of self-confidence of {X4 }, and the attribute criterion
intelligence of {X5 }.

6.3 Variable Weight Grey Clustering Model

The variable weight grey clustering model, the fixed weight grey clustering eval-
uation model, the grey clustering evaluation model using end-point and center-point
triangular possibility functions, and the grey clustering evaluation model based on
mixed possibility functions are all grey clustering evaluation models based on
different possibility functions. Therefore, before the introduction of the variable
weight grey clustering model in this section, the Four kinds of possibility functions
are explained next.
6.3 Variable Weight Grey Clustering Model 111

Fig. 6.1 The possibility

function of typical form

Definition 6.3.1 Assume that there are n objects to be classified according to

m criteria into s different grey classes. Classifying the ith object into the kth grey
class according to the observed value of the ith object judged against the jth
criterion, xij , i ¼ 1; 2;    n; j ¼ 1; 2;    m; is called grey clustering (Deng 1985).
The possibility function of the jth criterion of the kth subclass is denoted by
fjk ðÞ; j ¼ 1; 2;    m; k ¼ 1; 2;    s:
Definition 6.3.2 Assume that the possibility function of the jth criterion of kth
subclass fjk ðÞ takes the typical form shown in Fig. 6.1. Then xkj ð1Þ; xkj ð2Þ; xkj ð3Þ;
and xkj ð4Þ are referred to as turning points of fjk ðÞ:
The possibility function of typical form is denoted by f kj ½xkj ð1Þ; xkj ð2Þ; xkj ð3Þ;
xkj ð4Þ:
Definition 6.3.3 Assume that the possibility function of the jth criterion of kth
subclass fjk ðÞ does not have the first and second turning points xkj ð1Þ and xkj ð2Þ; as
shown in Fig. 6.2. Then fjk ðÞ is referred to as the possibility function of lower
measure.
The possibility function of lower measure is denoted by f kj ½; ; xkj ð3Þ; xkj ð4Þ:
Definition 6.3.4 Assume that the possibility function of the jth criterion of kth
subclass fjk ðÞ does not have the third turning point xkj ð3Þ; or that the second and
third turning points xkj ð2Þ and xkj ð3Þ of fjk ðÞ coincide, as shown in Fig. 6.3. In this

Fig. 6.2 The possibility

function of lower measure
112 6 Grey Clustering Evaluation Models

Fig. 6.3 The possibility

function of moderate measure

Fig. 6.4 The possibility

function of upper measure

case, fjk ðÞ is referred to as a possibility function of moderate measure, or a tri-

angular possibility function. The possibility function of moderate measure, or tri-
angular possibility function, is denoted by f kj ½xkj ð1Þ; xkj ð2Þ; ; xkj ð4Þ:
Definition 6.3.5 Assume that the possibility function of the jth criterion of kth
subclass fjk ðÞ does not have turning points xkj ð3Þ and xkj ð4Þ; as shown in Fig. 6.4.
Function fjk ðÞ is then referred to as a possibility function of upper measure. The
possibility function of upper measure is denoted by f kj ½xkj ð1Þ;xkj ð2Þ; ; :
Proposition 6.3.1 (1) For the possibility function of typical form as shown in
Fig. 6.1, we have:
8
> 0 x 62 ½xkj ð1Þ; xkj ð4Þ
>
>
>
>
>
> x  xkj ð1Þ
>
> x 2 ½xkj ð1Þ; xkj ð2Þ
< xk ð2Þ  xk ð1Þ
j j
fjk ðxÞ ¼ ð6:1Þ
>
> 1 x 2 ½xkj ð2Þ; xkj ð3Þ
>
>
>
>
>
> xkj ð4Þ  x
>
: k x 2 ½xkj ð3Þ; xkj ð4Þ
xj ð4Þ  xkj ð3Þ
6.3 Variable Weight Grey Clustering Model 113

(2) For the possibility function of lower measure as shown in Fig. 6.2, we have:
8
>
> 0 x 62 ½0; xkj ð4Þ
<
fjk ðxÞ ¼ 1 x 2 ½0; xkj ð3Þ ð6:2Þ
>
> x k
ð4Þx
: kj k x 2 ½xj ð3Þ; xj ð4Þ
k k
x ð4Þx ð3Þ
j j

(3) For the possibility function of moderate measure as shown in Fig. 6.3, we
have:
8
>
> 0 x 62 ½xkj ð1Þ; xkj ð4Þ
>
< xxkj ð1Þ
fjk ðxÞ ¼ xkj ð2Þxkj ð1Þ
x 2 ½xkj ð1Þ; xkj ð2Þ ð6:3Þ
>
>
>
: k xj ð4Þx
k
x2 ½xkj ð2Þ; xkj ð4Þ
xj ð4Þxkj ð2Þ

(4) For the possibility function of upper measure as shown in Fig. 6.4, we have:
8
>
> 0; x\xkj ð1Þ
<
xxkj ð1Þ
fjk ðxÞ ¼ ; x 2 ½xkj ð1Þ; xkj ð2Þ ð6:4Þ
>
>
xkj ð2Þxkj ð1Þ
: 1; x  xkj ð2Þ

Definition 6.3.6 (1) For the possibility function of typical form as shown in
Fig. 6.1, let kkj ¼ 12 ðxkj ð2Þ þ xkj ð3ÞÞ:
(2) For the possibility function of lower measure as shown in Fig. 6.2, let
kkj ¼ xkj ð3Þ:
(3) For the possibility function of moderate measure as shown in Fig. 6.3 and the
possibility function of upper measure as shown in Fig. 6.4, let kkj ¼ xkj ð2Þ:
Then kkj is referred to as the basic value of the jth criterion of the kth subclass.

Definition 6.3.7 Assume that kkj is the basic value of the jth criterion of the kth
subclass. Then the following formula is referred to as the weight of the jth criterion
of kth subclass (Deng 1985):
,
X
m
gkj ¼ kkj kkj ð6:5Þ
j¼1

Definition 6.3.8 Assume that xij , i ¼ 1; 2;    n; j ¼ 1; 2;    m is the observed value

of object i with regard to the jth criterion, fjk ðÞ the possibility function and gkj the
114 6 Grey Clustering Evaluation Models

weight of the jth criterion of the kth subclass, with j ¼ 1; 2;    m; k ¼ 1; 2;    s:

Then the following is referred to as the grey clustering coefficient of variable weight
for object i to belong to the kth grey class (Deng 1985):

X
m
r ki ¼ fjk ðxij Þ  gkj ð6:6Þ
j¼1

Definition 6.3.9 (1) The following formula is referred to as the clustering coeffi-
cient vector of object i:

X
m X
m X
m
ri ¼ ðr1i ; r2i ;    ; rsi Þ ¼ ð fj1 ðxij Þg1j ; fj2 ðxij Þg2j ;    ; fjs ðxij Þgsj Þ
j¼1 j¼1 j¼1

(2) The following is referred to as the cluster coefficient matrix:

2 3
r11 r21  rs1
X 6 r12 r22  rs2 7
6 7
¼ ðrki Þ ¼ 6 . .. .. .. 7
4 .. . . . 5
r1n r2n  rsn

Definition 6.3.10 If max1  k  s frki g ¼rki , then object i belongs to grey class k  .
The variable weight clustering method is used to study problems with criteria that
have the same meanings and units. Otherwise, it is not appropriate to employ this
method. Also, if the numbers of observed values of individual criteria are greatly
different from each other, this clustering method should not be applied.
Example 6.3.1 Assume that we are interested in the study of three economic dis-
tricts with the value-added by the primary, secondary and tertiary industries as the
three cluster criteria. The observational values xij, i = 1, 2, 3; j = 1, 2, 3, of the ith
economic district with respect to the jth criterion is given in the following matrix A,
where the unit of the three criteria is same as a hundred million RMB:
2 3 2 3
x11 x12 x13 80 20 100
A ¼ ðxij Þ ¼ 4 x21 x22 x23 5 ¼ 4 40 30 30 5
x31 x32 x33 10 90 60

Please try to perform a synthetic clustering based on high, medium, and low
added values.
Solution: Assume that the possibility functions fjk ðÞ for the jth criterion of the kth
subclass are as follows:
6.3 Variable Weight Grey Clustering Model 115

f11 ½30; 80; ; ; f12 ½10; 40; ; 70; f13 ½; ; 10; 30

f21 ½30; 90; ; ; f22 ½20; 50; ; 90; f23 ½; ; 20; 40

f21 ½40; 100; ; ; f32 ½30; 60; ; 90; f33 ½; ; 30; 50

It follows that:
8 8
< 0;x\30 < 0; x 62 ½10; 70
f11 ðxÞ ¼ 8030
x30
; 30  x\80 ; f12 ðxÞ ¼ x10
; 10  x\40
: : 4010
x [ 80 7040; 40  x\70
70x
1;
8 8
< 0; x 62 ½0; 30 < 0; x\30
f1 ðxÞ ¼
3
1; 0  x\10 ; f21 ðxÞ ¼ x30
; 30  x\90
: 30x : 9030
3010; 10  x\30 1; x [ 90
8 8
< 0; x 62 ½20; 90 < 0; x 62 ½0; 40
f22 ðxÞ ¼ x20
; 20  x\50 ; f2 ðxÞ ¼
3
1; 0  x\20
: 5020 : 40x
9050; 50  x\90 4020; 20  x\40
90x

8 8
< 0;
x\40 < 0; x 62 ½30; 90
f31 ðxÞ ¼ 10040
x40
; 40  x\100 ; f32 ðxÞ ¼ x30
; 30  x\50
: : 5030
x [ 100 9050; 50  x\90
90x
1;
8
< 0; x 62 ½0; 50
f33 ðxÞ ¼ 1; 0  x\30
: 50x
5030; 30  x\50

Therefore:

k11 ¼ 80; k12 ¼ 90; k13 ¼ 100; k21 ¼ 40; k22 ¼ 50; k23 ¼ 60; k31 ¼ 10; k32 ¼ 20; k33
¼ 30

kk
By gkj ¼ P3 j we have:
j¼1
kkj

80 90 100 2 40 50 60 10
g11 ¼ ; g12 ¼ ; g13 ¼ ; g1 ¼ ; g22 ¼ ; g23 ¼ ; g31 ¼ ;
270 270 270 150 150 150 60
20 3 30
g2 ¼ ; g3 ¼
3
60 60
116 6 Grey Clustering Evaluation Models

Pm
Thus, from r ki ¼ j¼1 fj ðxij Þgj ,
k k
when i ¼ 1 for economic district 1, we have:

X
3
80 90 100
r 11 ¼ fj1 ðx1j Þg1j ¼ f11 ð80Þ þ f21 ð20Þ þ f31 ð100Þ ¼ 0:6667
j¼1
270 270 270

Similarly, we obtained the following:

r21 ¼ 0; r31 ¼ 0:3333

Therefore, r1 ¼ ðr11 ; r21 ; r31 Þ ¼ ð0:6667; 0; 0:3333Þ:

Similarly, we can calculate the clustering coefficient vector for economic dis-
tricts 2 and 3 as done foreconomic district 1.
When i ¼ 2; r2 ¼ ðr12 ; r22 ; r32 Þ ¼ ð0:0593; 0:3778; 0:6667Þ:
When i ¼ 3; r3 ¼ ðr13 ; r23 ; r33 Þ ¼ ð0:4667; 0:4; 0:1667Þ:
The clustering coefficient matrix is as follows:
2 3 2 3
X r11 r21 r31 0:6667 0 0:3333
4
¼ ðri Þ ¼ r12
k
r22 r32 5 ¼ 4 0:0593 0:3778 0:6667 5
r13 r23 r33 0:4667 0:4 0:1667

From max1  k  3 frk1 g ¼r11 ¼ 0:6667, max1  k  3 frk2 g ¼r32 ¼ 0:6667,

max1  k  3 fr3 g ¼r3 ¼ 0:4667; it follows that the second economic district belongs
k 1

to the low grey class of added value, and the first and third economic districts
belong to the high grey class of added value. Furthermore, from the cluster coef-
ficients r11 ¼ 0:6667 and r13 ¼ 0:4667; it follows that there still exists some dif-
ferences between the first and third districts, even though both belong to the high
grey class of added value. If the grey classes of added value are refined, that is, if
we use five grey classes such as high, mid-high, medium, mid-low, and low added
value, then different results can be obtained.
Furthermore, to determine the possibility function for the jth criterion of the kth
subclass, it is generally possible to use the background information of the problem
at hand. When resolving practical problems, one can determine the possibility
functions from either the angle of the objects that are to be clustered or by looking
at all the same type objects in the whole system, not just the ones involved in the
clustering. For example, in Example 6.3.1, we could determine the possibility
functions not only from the three economic districts in question, but also from the
same level of economic districts in a city, a province, or from around the nation.
Therefore, the results of grey clustering evaluation can only be applied to a certain
range, which is the same as the one used in the determination of relevant possibility
functions.
6.4 Fixed Weight Grey Clustering Model 117

6.4 Fixed Weight Grey Clustering Model

When the criteria for clustering have different meanings, dimensions (units), and
drastically different numbers of observed data points, the variable weight clustering
method can lead to weak criteria. There are two ways to get around this problem.
The first is to transform the sample of data values of all the criteria into
non-dimensional values by applying either the initiating operator or the averaging
operator, and then clustering the transformed data. When employing this method,
all the criteria are treated equally so that no difference played by the criteria in the
process of clustering is reflected. The second way to get around the weak criteria
problem is to assign each clustering criterion a weight ahead of the clustering
process. In this section, we address this second method.
Definition 6.4.1 Assume that xij is the observed value of object i with regard to
criterion j, i ¼ 1; 2;    n; j ¼ 1; 2;    m; and fjk ðÞ the possibility function of the jth
criterion of the kth subclass, j ¼ 1; 2;    m; k ¼ 1; 2;    ; s: If the weight gkj of the
jth criterion of the kth subclass is not a function of k, j ¼ 1; 2;    m; k ¼ 1; 2;    ; s:
That is, if for any k1 ; k2 2 f1; 2;    ; sg we always have gkj 1 = gkj 2 , then the symbol
gkj can be written as gj , j ¼ 1; 2;    m; with the superscript k removed. In this case,
the following is referred to as the fixed weight clustering coefficient for object i to
belong to the kth grey class (Liu 1993).

X
m
r ki ¼ fjk ðxij Þgj ð6:7Þ
j¼1

Definition 6.4.2 In formula (6.7), if gj ¼ m1 ; for j ¼ 1; 2;    ; m; then the following

is referred to as the equal weight clustering coefficient for object i to belong to the
kth grey class:

X
m
1X m
r ki ¼ fjk ðxij Þgj ¼ f k ðxij Þ
j¼1
m j¼1 j

The method of clustering objects by using grey fixed weight clustering coeffi-
cients is known as grey fixed weight clustering. The method which uses grey equal
weight clustering coefficients is known as grey equal weight clustering.
Grey fixed weight clustering can be carried out according to the following steps:
Step 1: Determine the possibility function fjk ðÞ for the jth criterion of the kth
subclass, j ¼ 1; 2;    m; k ¼ 1; 2;    ; s:
Step 2: Determine a clustering weight gj for each criterion j ¼ 1; 2;    m:
Step 3: Based on the possibility functions fjk ðÞ obtained in step 1, the clustering
weights gj obtained in step 2, and the observed data value xij of object i with
118 6 Grey Clustering Evaluation Models

respect to criterion j, calculate the fixed weight clustering coefficients

P
r ki ¼ m j¼1 fj ðxij Þ, i ¼ 1; 2;    n; j ¼ 1; 2;    m; k ¼ 1; 2;    ; s
k

Step 4: If max1  k  s frki g ¼rki , then object i belongs to grey class k  :
Example 6.4.1 Let us perform a grey clustering for the ecological adaptation of
major strains of trees commercially used in China (Li et al. 1994). China is a huge
country with a very diverse ecological environment, and different strains of trees
obviously require different growing conditions. The area where a certain strain of
trees has been growing reflects the adaptability of the strain to that particular
ecological environment. We now classify ecological environmental conditions into
four main quantification criteria:
(1) Geographical measure;
(2) Temperature measure;
(3) Precipitation measure; and
(4) Arid measure.
Here, geographical measure is an index representing the geographical width of
the region in which the strain of trees grows. The numerical value of this measure is
given by the product of differences of longitudes in the directions of east and west,
and latitudes in the directions of south and north. The temperature measure indi-
cates the adaptability of the strain of trees to various temperatures. Its numerical
value is computed by using the difference of annual average temperatures of the
southern and the northern bounds of the growing region. The precipitation measure
is the adaptability of the trees to precipitation conditions. Its numerical value is
recorded as the difference between the maximum and minimum annual average
precipitation in all areas of the growing region. The arid measure is selected to
describe a strain’s adaptability to arid conditions in the atmosphere. Its value is the
difference between the maximum and minimum annual average aridities in different
areas of the growing region.
Statistics regarding the four measures for the 17 main strains of trees planted in
China are given in Table 6.3.
With such data it is possible to carry out grey clustering based on wide adapt-
ability, medium adaptability, and narrow adaptability.
Solution: Because the meanings of the criteria are different and there exists much
difference among the values observed, we must apply the fixed weight clustering
method.
Step 1: Assume that the possibility functions fjk ðÞðj ¼ 1; 2; 3; 4; k ¼ 1; 2; 3Þ for
the jth criterion of the kth subclass are as follows:

f11 ½100; 300; ; ; f12 ½50; 150; ; 250; f13 ½; ; 50; 100

f21 ½3; 10; ; ; f22 ½2; 6; ; 10; f23 ½; ; 4; 8

6.4 Fixed Weight Grey Clustering Model 119

Table 6.3 The four measures of the 17 main strains of trees in China
Measure trees Geo. eco. Temp. eco Prec. eco Arid eco.
measure measure measure measure
1 Camphor pine 22.50 4 0 0
2 Korean pine 79.37 6 600 0.75
3 Northeast China ash 144.00 7 300 0.75
4 Diversiform-leaved 300.00 6.1 189 12.00
poplar
5 Sacsaoul 456.00 12 250 12.00
6 Chinese pine 189.00 8 700 1.5
7 Oriental arborvitae 369.00 8 1300 2.25
8 White elm 1127.11 16.2 550 3.00
9 Dryland willow 260.00 11 600 1.00
10 Chinese white 200.00 8 600 1.25
poplar
11 Oak 475.00 10 1000 0.75
12 Huashan pine 314.10 8 900 0.75
13 Masson pine 282.80 7.4 1300 0.5
14 China fir 240.00 8 1200 0.5
15 Bamboo 160.00 5 1000 0.25
16 Camphor tree 270.00 8 1200 0.25
17 Southern Asian pine 9.00 1 200 0

f31 ½200; 1000; ; ; f32 ½100; 600; ; 1100; f33 ½; ; 300; 600

f41 ½0:25; 1:25; ; ; f42 ½0; 0:5; ; 1; f43 ½; ; 0:25; 0:5

Step 2: Let the weights for the geographical, temperature, precipitation, and
aridity measures be:

g1 ¼ 0:3; g2 ¼ 0:25; g3 ¼ 0:25; g4 ¼ 0:2

Pm
Step 3: Based on r ki ¼ j¼1 fj ðxij Þ  gj ;
k
i ¼ 1; 2;    ; 17; k ¼ 1; 2; 3 and
Table 6.3, when i = 1,

X
4
r 11 ¼ fj1 ðx1j Þgj ¼ f11 ð22:5Þ 0:3 þ f21 ð4Þ 0:25 þ f31 ð0Þ 0:25 þ f41 ð0Þ 0:2
j¼1
¼ 0:0357

X
m X
m
and r 21 ¼ fj2 ðx1j Þgj ¼ 0:125; r 31 ¼ fj3 ðx1j Þgj ¼ 1
j¼1 j¼1
120 6 Grey Clustering Evaluation Models

Therefore,

r1 ¼ ðr11 ; r21 ; r31 Þ ¼ ð0:0357; 0:125; 1Þ

Similarly, we can calculate and obtain:

r2 ¼ ðr12 ; r22 ; r32 Þ ¼ ð00:3321; 00:6881; 00:2488Þ

r3 ¼ ðr13 ; r23 ; r33 Þ ¼ ð00:3401; 00:6695; 00:3125Þ

r4 ¼ ðr14 ; r24 ; r34 Þ ¼ ð00:6107; 00:2883; 00:3688Þ

r5 ¼ ðr15 ; r25 ; r35 Þ ¼ ð0 0:7656; 00:075; 00:25Þ

r6 ¼ ðr16 ; r26 ; r36 Þ ¼ ð00:6683; 00:508; 0Þ

r7 ¼ ðr17 ; r27 ; r37 Þ ¼ ð00:9286; 00:125; 0Þ

r8 ¼ ðr18 ; r28 ; r38 Þ ¼ ð00:8594; 00:225; 00:0417Þ

r9 ¼ ðr19 ; r29 ; r39 Þ ¼ ð00:765; 00:25; 0Þ

r10 ¼ ðr110 ; r210 ; r310 Þ ¼ ð00:6536; 00:525; 0Þ

r11 ¼ ðr111 ; r211 ; r311 Þ ¼ ð00:9; 00:15; 0Þ

r12 ¼ ðr112 ; r212 ; r312 Þ ¼ ð00:7973; 00:325; 0Þ

r13 ¼ ðr113 ; r213 ; r33 Þ ¼ ð00:7313; 00:3625; 00:0375Þ

r14 ¼ ðr114 ; r214 ; r34 Þ ¼ ð00:6886; 00:355; 0Þ

r15 ¼ ðr115 ; r25 ; r315 Þ ¼ ð00:4114; 00:6075; 00:3875Þ

r16 ¼ ðr116 ; r216 ; r316 Þ ¼ ð00:6836; 00:225; 00:2Þ

Furthermore,

r17 ¼ ðr117 ; r217 ; r37 Þ ¼ ð0; 00:05; 1Þ

Step 4: Based on the following facts, it follows that trees with numberings 4, 5,
6, 7, 8, 9, 10, 11, 12, 13, 14, 16, are strains with wide adaptability:
6.4 Fixed Weight Grey Clustering Model 121

max frk1 g ¼ r31 ¼ 1; max frk2 g ¼ r22 ¼ 00:6881;

1k3 1k3
max frk3 g ¼ r23 ¼ 00:6695
1k3

max frk4 g ¼ r14 ¼ 00:6107; max frk5 g ¼ r15 ¼ 0 0:7656;

1k3 1k3
max frk6 g ¼ r16 ¼ 00:6683
1k3

max frk7 g ¼ r17 ¼ 00:9286; max frk8 g ¼ r18 ¼ 00:8594;

1k3 1k3
max frk9 g ¼ r19 ¼ 00:765
1k3

max frk10 g ¼ r110 ¼ 00:6536; max frk11 g ¼ r111 ¼ 00:9;

1k3 1k3
max frk12 g ¼ r112 ¼ 0:91
1k3

max frk13 g ¼ r113 ¼ 0:82; max frk14 g ¼ r114 ¼ 00:6886;

1k3 1k3
max frk15 g ¼ r215 ¼ 00:6075
1k3

max frk16 g ¼ r116 ¼ 00:6836; max frk17 g ¼ r317 ¼ 1

1k3 1k3

Such strains are diversiform-leaved poplars, sacsaouls, Chinese pines, oriental

arborvitaes, white elms, dryland willows, Chinese white poplars, oaks, Huashan
pines, masson pines, China firs, and camphor trees. These trees have an extremely
strong ability to adapt themselves to natural ecological environments, can grow well
in most parts of China, and should be widely introduced. The trees named Korean
pine, Northeast China Ash, and bamboo with numberings 2, 3, and 15, respectively,
belong to the grey class of medium adaptability, and can be introduced to a rela-
tively large area in China. Finally, trees with the names camphor pine and South
Asian pine, and numberings 1 and 17, respectively, belong to the grey class of
narrow adaptability, where camphor pines are found near the Northern border of
China and South Asian pines are mainly located near the Southern border of China.

6.5 Grey Clustering Evaluation Models Based on Mixed

Possibility Functions

6.5.1 Grey Clustering Evaluation Model Based

on End-Point Mixed Possibility Functions

The Grey clustering evaluation model based on mixed end-point triangular possi-
bility functions is a new model (Liu et al. 2015). Compared with end-point
122 6 Grey Clustering Evaluation Models

triangular possibility functions, the new model has changed the possibility function
for grey class 1 to the possibility function of lower measure, and the possibility
function for grey class s to the possibility function of upper measure. Additionally,
the new model has avoided the problem of extension of the bound of value of each
clustering index. The Grey cluster evaluation model based on mixed end-point
triangular possibility function is suitable for situations where all grey boundaries are
clear, but the most likely points belonging to each grey class are unknown. The
modeling steps are explained below.
Step 1: S grey classes are needed in an evaluation, and the value range of each
index is also divided into s classes. For example, the value range ½a1 ; as þ 1  of index
j can be divided into s small intervals:

½a1 ; a2 ;    ; ½ak1 ; ak ;    ; ½as1 ; as ; ½as ; as þ 1 

The value of ak ðk ¼ 2;    ; sÞ can be determined by the actual assessment

requirements or the qualitative research results.
Step 2: Determine the turning point k1j and ksj of ½a1 ; a2  and ½as ; as þ 1  that
correspond to grey classes 1 and s. At the same time, calculate the geometric
center-point kk ¼ ðak þ ak þ 1 Þ=2 for each small interval ½ak ; ak þ 1 ;
k ¼ 2;    ; s  1:
Step 3: For grey class 1 and grey class s, construct the corresponding possibility
function of lower measure fj1 ½; ; k1j ; k2j  and the possibility function of upper
measure fjs ½kjs1 ; ksj ; ; :
Assume that x is an observation of index j, when x 2 ½a1 ; k2j  or x 2 ½kjs1 ; as þ 1 ;
using formulas (6.8) or (6.9), respectively:
8
>
< 0 x 62 ½a1 ; k2j 
fj1 ðxÞ ¼ 1 x 2 ½a1 ; k1j  ; ð6:8Þ
> kj x
2
: x 2 ½k1j ; k2j 
k2j k1j

or
8
>
< 0 x 62 ½kjs1 ; as þ 1 
xkjs1
fjs ðxÞ ¼ x 2 ½kjs1 ; ksj  : ð6:9Þ
>
:
ksj kjs1
1 x 2 ½ksj ; as þ 1 

By using one of these formulas, the possibility degree of fj1 ðxÞ or fjs ðxÞ regarding
grey class 1 and grey class s can be calculated.
Step 4: For grey class kðk 2 f2; 3;    ; s  1gÞ; connecting point ðkkj ; 1Þ with
center-point ðkk1
j ; 0Þ of grey class k  1 (or turning pointðk1j ; 0Þ of grey class 1),
and connecting ðkkj ; 1Þ with center-point ðkkj þ 1 ; 0Þ of grey class k þ 1 (or turning
point ðksj ; 0Þ of grey class s), we can get the triangular possibility function
6.5 Grey Clustering Evaluation Models Based on Mixed Possibility Functions 123

y = f j1 ( x) y = f j2 ( x ) y = f j k ( x) y = f j s −1 ( x ) y = f js ( x )
1

o a1j λ j a 2j λ j2 a 3j λ 3j a 4j
1
a kj −1 λ jk −1a kj λ k a kj +1λ jk +1a kj + 2 a sj −2 λ js −2a sj −1λ js −1a sj λ js a sj +1 x
j

Fig. 6.5 The end-point mixed possibility function

fjk ½kjk1 ; kkj ; ; kkj þ 1 ; j ¼ 1; 2;    ; m; k ¼ 2; 3;    ; s  1 of index j regarding grey

class k (shown in Fig. 6.5).
For index j; x is an observation of it when k ¼ 2; 3;    ; s  1; according to
formula (6.10):
8
>
> 0 x 62 ½kjk1 ; kkj þ 1 
>
< xkjk1
fjk ðxÞ ¼ kkj kk1
x 2 ½kjk1 ; kkj  ð6:10Þ
>
>
j
>
: kjk þ 1 x
kjk þ 1 kkj
x2 ½kkj ; kkj þ 1 

This formula allows the possibility degree of fjk ðxÞ regarding grey class
kðk 2 f2; 3;    ; s  1gÞ to be calculated.
Step 5: Determine the weight wj ; j ¼ 1; 2;    ; m of each index.
Step 6: Calculate the clustering coefficient rki of object iði ¼ 1; 2;    ; nÞ
regarding grey class kðk ¼ 1; 2;    ; sÞ :

X
m
rki ¼ fjk ðxij Þ  wj : ð6:11Þ
j¼1

fjk ðxij Þ is the possibility function of index j and subclass k,wj is the weight of
index j among comprehensive clustering.

Step 7: By max1  k  s frki g ¼ rki we can determine whether or not object i
belongs to grey class k .
When there are multiple objects belonging to the same grey class k , we can
further determine individual objects’ precedence in grey class k  on the basis of the
size of integrate clustering coefficients.
Example 6.5.1 Five vendors A, B, C, D, E who undertake the development of the
C919 body component for Commercial Aircraft Corporation of China Ltd
(COMAC) are evaluated on their performance and are divided into four classes
including “excellent”, “good”, “medium” and “poor”.
124 6 Grey Clustering Evaluation Models

Step 1: Set the evaluation index system for vendor performance.

The evaluation index system for vendor performance reflects the specific
requirements of the main manufacturers to vendors. It is an important basis for the
main manufacturers to comprehensively evaluate the vendors and make final
management decisions.
Factors that affect vendor performance are very complex. Main manufactures’
foci on vendor performance are also not the same across different stages. After four
rounds of expert investigation, six first-grade evaluation indexes are determined,
including quality, cost, delivery, cooperation, technology and service.
At development stage, the four second-grade indexes of quality are pass rate of
product, quality control system, airworthiness certification ability and control of
sub-vendor. The three second-grade indexes of cost are price, logistics costs and
price stability. The two second-grade indexes of delivery are punctuality and
flexibility. The three second-grade indexes of cooperation are credit, information
communication and cooperation intention. The five second-grade indexes of tech-
nology are professional R&D staff, R&D investment, number of invention patents,
market share and technology level. The four second-grade indexes of service are
quick response, spare part support, training and technology support.
Among those indexes, pass rate and market share are shown in percentage. Price
and logistics costs are quantitative indexes and the unit is ten thousand yuan. The
smaller the indexes, the better. The unit of professional R&D staff is person, R&D
investment is ten thousand yuan, and the unit of patent number is an item. The
bigger the indexes, the better. Other indexes like quality control system, price
stability, delivery punctuality, flexibility, credit, information communication,
technology level, quick response, spare part support, training and technology
support are all qualitative indexes. They are usually quantified by expert grade.
Here, the grade is a 10-point scale score and decimal points are allowed.
If vendors who have different tasks are evaluated together, most quantitative
indexes such as price and logistics costs cannot be compared. Therefore, at this
point we need to invite experienced experts to make qualitative assessments of
quantitative indexes by grading them as a 10-point scale score. The evaluation
index system for vendor performance and its weight at development stage are
shown in Table 6.4.
For the evaluation of vendor performance at development stage, we use the
index system shown in Table 6.4.
Step 2: According to the evaluation results, the value range of each index is
divided into four grey classes. The value of second-grade indexes are usually
divided into four small sections based on the sample value. Considering the opinion
of COMAC, the effect sample matrix of second-grade index is omitted. Here are the
actual values of six first-grade indexes that are obtained by weighted integration of
the second-grade indexes as 10-point scale scores. The values are
yij ; ði ¼ 1; 2    ; 5; j ¼ 1; 2;    6Þ as shown in Table 6.5.
6.5 Grey Clustering Evaluation Models Based on Mixed Possibility Functions 125

Table 6.4 The evaluation index system for vendor performance and its weight at development
stage
First-grade index and its Second-grade index Code Unit Weight
weight
Quality (22 %) Pass rate x1 % 6
Quality control system x2 Qualitative 6
Airworthiness certification x3 Qualitative 5
ability
Control of sub-vendor x4 Qualitative 5
Cost (18 %) Price x5 Ten thousand 8
yuan
Logistic cost x6 Ten thousand 4
yuan
Price stability x7 Qualitative 6
Delivery (17 %) Punctuality x8 Qualitative 12
Flexibility x9 Qualitative 5
Cooperation (13 %) Credit x10 Qualitative 6
Information communication x11 Qualitative 4
Cooperation intention x12 Qualitative 3
Technology (16 %) Professional R&D staff x13 Person 3
R&D investment x14 Ten thousand 3
yuan
Number of invention patent x15 Item 3
Market share x16 % 3
Technology level x17 Qualitative 4
Service (14 %) Quick response x18 Qualitative 4
Spare part support x19 Qualitative 4
Training x20 Qualitative 3
Technology support x21 Qualitative 3

Table 6.5 The actual values of first-grade index of five vendors

Vendor Actual value yij
Quality Cost Delivery Technology Cooperation Service
A 9.1 7.8 8.4 9 9.5 9.3
B 9.3 7.5 9 9.2 9 9
C 9 8.6 8.7 9 9.1 9.1
D 8.9 8.5 9 9.1 9.6 9.2
E 8.6 9 8.6 9 9.7 9.5

The six first-grade indexes are all in 10-point scores, and the value range is
[0, 10]. Interval [0, 10] is sub-divided into 4 small intervals as [0, 6), [6, 7.5), [7.5,
9), [9,10], which correspond to “poor”, “medium”, “good” and “excellent”.
126 6 Grey Clustering Evaluation Models

Table 6.6 The clustering The clustering objects The clustering coefficient
coefficient regarding to each
grey class of five vendors r1i r2i r3i r4i
A 0 5.4 42.04 52.56
B 0 9 34.44 56.56
C 0 0 47.44 52.56
D 0 0 39.28 60.72
E 0 0 40.48 59.52

Step 3: Determine the turning point k1j ¼ 5; k4j ¼ 9:5 of [0, 6) and [9, 10] that
correspond to grey class 1 and grey class 4. At the same time, calculate the
center-point of [6, 7.5) and [7.5, 9), k2j ¼ 6:75; k3j ¼ 8:25:
Step 4: By using formulas (6.8), (6.9), and (6.10), the possibility functions of
index j regarding grey class k(k = 1, 2, 3, 4) can be obtained as follows:
8 8
< 0 x 62 ½0; 6:75 < 0 x 62 ½5; 8:25
fj1 ðxÞ ¼ 1 x 2 ½0; 5Þ fj2 ðxÞ ¼ x5
x 2 ½5; 6:75Þ
: : 1:75
6:75x
1:75 x 2 ½5; 6:75 8:25x
1:5 x 2 ½6:75; 8:25
8
> x 62 ½6:75; 9:5 8
>
< 0 < 0 x 62 ½8:25; 10
x 2 ½6:75; 8:25Þ
fj3 ðxÞ ¼ x6:75
fj4 ðxÞ ¼ x8:25
x 2 ½8:25; 9:5Þ
>
>
1:5
: 1:25
: 9:5x
1 x 2 ½9:5; 10
1:25 x 2 ½8:25; 9:5

where j ¼ 1; 2;    6:
Step 5: The weight of each index wj ; j ¼ 1; 2; 3; 4; 5; 6 is shown in Table 6.4.
Step 6: According to formula (6.11), the clustering coefficient regarding the grey
class of five vendors can be calculated ði ¼ 1; 2; 3; 4; 5; k ¼ 1; 2; 3; 4Þ; as shown in
Table 6.6.  
Step 7: As can be seen from the results of max1  k  4 rkA ¼ 52:56 ¼ r4A ,
     
max1  k  4 rkB ¼ 56:56 ¼ r4B , max1  k  4 rkC ¼ 52:56 ¼ r4C , max1  k  4 rkD
 
¼ 60:72 ¼ r4D max1  k  4 rkE ¼ 59:52 ¼ r4E , the performance of five vendors A,
B, C, D, E at development stage all reach the level of “excellent”. Among those
vendors, the clustering coefficient of vendor D regarding grey class “excellent” is
the highest and vendor E takes the second place. However, the difference between
D and E is very small, so the two vendors belong to the same level. Then comes
vendor B. The coefficient of vendor A and C regarding grey class “excellent” is the
smallest.
Further investigation reveals that the indexes belonging to class “excellent” of
vendors D and E are technology, cooperation and service. There is much room for
improvement in terms of quality and cost for D, and in terms of quality and delivery
for E. The main problem for vendor B is its high cost. Although the evaluation on
6.5 Grey Clustering Evaluation Models Based on Mixed Possibility Functions 127

cooperation and service is quite good, the value is still on the low side compared
with other vendors. For vendor A and C, the main problems are cost and delivery.
The management department of COMAC can focus on each vendor according to
their own problems and improve their whole performance level promptly.
In this example, the value range of each index as well as its turning point and
center-point k1j ; k2j ; k3j ; k4j ; j ¼ 1; 2;    ; 6 regarding different grey classes are
determined according to the expert evaluation results of vendors A, B, C, D, and E.
Also, the conclusion only applies to the current situation of those vendors. The
results of grey clustering evaluation can be used with a certain scope: the scope
used when determining the possibility function is the one that can be used in the
evaluation results. The so called “excellent”, “good”, “medium” and “poor” classes
are also relative. Vendors A, B, C, D, and E are all prominent enterprises in China.
Although they are very strong, there is a big gap between their performance and that
of similar manufacturers around the world.

6.5.2 Grey Clustering Evaluation Model Based

on Center-Point Mixed Possibility Functions

This section addresses an improvement in the triangular possibility function. Such

an improvement entails changing the center-point triangular possibility function
which corresponds to class 1 to a possibility function of lower measure, and
changing the triangular possibility function which corresponds to class s to a
possibility function of upper measure. This improvement allows us to avoid having
to extend the bound of value of each clustering index.
Definition 6.5.1 For grey class kðk 2 f2; 3;    ; s  1gÞ; the point which most
likely belongs to grey class k is called the center-point of grey class k.
The center-point may or may not be the midpoint. This is determined by the
maximum likelihood of such a point to belong to the grey class.
The modeling steps of the grey cluster evaluation model using center-point
triangular possibility functions are as follows (Liu et al. 2015).
Step 1: Assume that ½aj ; bj  is the range of index j: According to the evaluation
requirements, we divide ½aj ; bj  into s small intervals. Then we determine the turning
point k1j ; ksj of grey classes 1 and s, and the center-point k2j ; k3j ;    ; kjs1 of grey
class kðk 2 f2; 3;    ; s  1gÞ; respectively.
Step 2: Construct the corresponding lower measure possibility function
fj ½; ; k1j ; k2j ; and the upper measure possibility function fjs ½kjs1 ; ksj ; ;  for
1

grey classes 1 and s (see Fig. 6.1).

Assume x is an observation value of index j: When x 2 ½aj ; k2j  or x 2 ½kjs1 ; bj ;
the possibility degree of fj1 ðxÞ or fjs ðxÞ regarding grey classes 1 and s can be
calculated by using formulas (1) and (2), respectively.
128 6 Grey Clustering Evaluation Models

Fig. 6.6 Center-point mixed possibility function

8
>
< 0 x 62 ½aj ; k2j 
fj1 ðxÞ ¼ 1 x 2 ½aj ; k1j  ð6:12Þ
> kj x
2
: x 2 ½k1j ; k2j 
k2j k1j

8
>
< 0 x 62 ½kjs1 ; bj 
xkjs1
fjs ðxÞ ¼ x 2 ½kjs1 ; ksj  ð6:13Þ
>
:
ksj kjs1
1 x 2 ½ksj ; bj 

Step 3: For grey class kðk 2 f2; 3;    ; s  1gÞ; by connecting point ðkkj ; 1Þ with
center-point ðkjk1 ; 0Þ of grey class k  1 (or turning point ðk1j ; 0Þ of grey class 1),
and by connecting ðkkj ; 1Þ with center-point ðkjk þ 1 ; 0Þ of grey class k þ 1 (or turning
point ðksj ; 0Þ of grey class s), we get triangular possibility function
fjk ½kjk1 ; kkj ; ; kkj þ 1 ; j ¼ 1; 2;    ; m; k ¼ 2; 3;    ; s  1 of index j regarding grey
class k (see Fig. 6.6).
Assume that x is an observation value of index j: The degree of membership
fjk ðxÞ regarding grey class kðk 2 f2; 3;    ; s  1gÞ can be calculated by using
formula (6.14).
8
>
> 0 x 62 ½kjk1 ; kkj þ 1 
>
< xkjk1
fjk ðxÞ ¼ kkj kk1
x 2 ½kjk1 ; kkj  ð6:14Þ
>
>
j
>
: kjk þ 1 x
kjk þ 1 kkj
x2 ½kkj ; kkj þ 1 

Step 4: Determine the weight wj ; j ¼ 1; 2;    ; m of each index.

Step 5: Compute clustering coefficient rki of object iði ¼ 1; 2;    ; nÞ regarding
grey class kðk ¼ 1; 2;    ; sÞ; as seen in Fig. (6.15).
6.5 Grey Clustering Evaluation Models Based on Mixed Possibility Functions 129

Fig. 6.7 Evaluation indicator

system of project for Discipline construcƟon evaluaƟon
discipline construction

Academic communicaƟon
ConstrucƟon condiƟons
Student culƟvaƟon

Discipline plaƞorm
ScienƟfic researc h
Faculty
x1

x2 x3 x4
x5 x6

X
m
rki ¼ fjk ðxij Þ  wj ð6:15Þ
j¼1

fjk ðxij Þ is the possibility function of index j for subclass k; while wj is the weight
of comprehensive clustering of index j:

Step 6: By max1  k  s frki g ¼ rki , determine that object i belongs to grey
class k .
When there are multiple objects that belong to the same grey class k , we can
further determine the precedence of individual objects in grey class k on the basis
of the size of integrate clustering coefficients.
Example 6.5.2 The evaluation of a project for discipline construction at a university
will be used to illustrate the application of the grey cluster evaluation models, which
are based on mixed center-point triangular possibility functions.
Based on extensive surveys, there are 6 primary indicators to reflect the per-
formance of a discipline construction project, including faculty, scientific research,
student cultivation, discipline platform construction, conditions for construction and
academic communication. The corresponding weights are 0.21, 0.24, 0.23, 0.14,
0.1, and 0.08, respectively (see Fig. 6.7).
We convert the evaluation scores of each indicator to centesimal system for
convenience. The evaluation results are divided into four grey classes including
“excellent”, “good”, “medium” and “poor”, according to requirements of the uni-
versity authorities. 41 projects for discipline construction have been conducted from
2010 to 2014. All the evaluation scores of the 6 indexes of these 41 projects for
discipline construction are laid in the interval of [40, 100]. We set up the turning
point k4j ¼ 90 for grey class “excellent” and the turning point k1j ¼ 60 for grey class
“poor”, as well as the most likely points k3j ¼ 80; k2j ¼ 70; which belong to grey
classes “good” and “medium”.
130 6 Grey Clustering Evaluation Models

Since the evaluation scores of each indicator are converted to centesimal system,
the possibility function of all 6 indicators on four grey classes of “poor”, “medium”,
“good”, and “excellent” are the same:
8 8
< 0 x 62 ½40; 70 < 0 x 62 ½60; 80
fj1 ðxÞ ¼ 1 x 2 ½40; 60 ; fj2 ðxÞ ¼ 7060
x60
x 2 ½60; 70 ;
: 70x : 80x
7060 x 2 ½60; 70 8070 x 2 ½70; 80
8 8
< 0 x 62 ½70; 90 < 0 x 62 ½80; 100
fj3 ðxÞ ¼ x70
x 2 ½70; 80 ; fj4 ðxÞ ¼ 9080
x80
x 2 ½80; 90
: 8070 :
90x
9080 x 2 ½80; 90 1 x 2 ½90; 100

where the possibility function of each indicator for grey class “poor” is a possibility
function of lower measure, each indicator for grey class “excellent” is a possibility
function of upper measure, and each indicator for grey classes “medium” and
“good” are triangular possibility functions. The values of the 6 indicators for a
university’s discipline construction project are shown in Table 6.7.
The values of possibility functions for the different grey classes of each indicator
can be calculated by using fj1 ðxÞ fj2 ðxÞfj3 ðxÞfj4 ðxÞ; j ¼ 1; 2;    ; 6: The grey clus-
tering coefficient di can be calculated by using formula (6.4). The outcomes are
shown in Table 6.8.
Based on the results in Table 6.8, we can confirm that the project belongs to grey
 
class “excellent” according to max1  k  4 dki ¼ d4i ¼ 0:419: Therefore, the effect
of the project for discipline construction is remarkable. But the grey clustering
coefficient which suggests that the project belongs to grey class “good” is
d3i ¼ 0:413: This result is very close to d4i . It also shows that the execution effect of
the project for discipline construction is situated between grey classes “excellent”
and “good”. As for the sub-indicators, the indicator on student cultivation belongs
to grey class “excellent”, and reached a high level. The indicator on scientific
research is situated between grey classes “good” and “excellent”, but close to grey
class “excellent”. The indicators on faculty and discipline platform construction

Table 6.7 The values of 6 indicators of a project for discipline construction

Indicator Faculty Scientific Student Discipline Construction Academic
research cultivation platform conditions communication
value 81 87 92 78 74 53

Table 6.8 Grey clustering Grey class x1 x2 x3 x4 x5 x6 di

coefficients of each indicator
for different grey classes Excellent 0.1 0.7 1.0 0 0 0 0.419
Good 0.9 0.3 0 0.8 0.4 0 0.413
Medium 0 0 0 0.2 0.6 0 0.088
Poor 0 0 0 0 0 1.0 0.080
6.5 Grey Clustering Evaluation Models Based on Mixed Possibility Functions 131

basically belong to grey class “good”, which indicates that the implementation of
these two indicators are better. The indicator on construction conditions is situated
between grey classes “good” and “medium”, but closer to grey class “medium”.
The indicator on academic communication belongs to grey class “poor “, which
suggests that there are still significant shortcomings in construction conditions and
academic communication that require further strengthening.
The grey cluster evaluation model based on mixed possibility function is more
suitable to solve problems of poor information clustering evaluation. On the other
hand, grey cluster evaluation model using center-point triangular possibility func-
tions is suitable for problems where it is relatively easy to judge the most likely
points belonging to each grey class, but the grey boundaries are not clear. Finally,
grey cluster evaluation model using end-point triangular possibility functions is
suitable for situations where all grey boundaries are clear, but the most likely points
belonging to each grey class are unknown.
These new models have helped to extend the bound of value of each clustering
index for people using the grey cluster evaluation model based on pure triangular
possibility function for more than 20 years.

6.6 Practical Applications

Example 6.6.1 In this case study, we will look at the system of evaluation criteria
for dominant regional industries and an actual selection of dominant industries for
Jiangsu province, PR China (Dang et al. 2003).
By a dominant industry we mean that, during a special time period of economic
development of either a country or a geographic region, the particular industry plays
a leading role in the regional structure of industries and dominates the overall
development of the regional economy. Such an industry possesses a strong market
position and the capability to advance its own technological development.
Historically, the process for a regional structure of industries to rise to a high level
of sophistication has always been signaled by the development of new dominant
industries and changes of dominance from one group of industrial leaders to
another. Japanese scholar Shinohara Miyohei proposed two theoretical standards
for planning for Japan’s industrial structure: (1) income elasticity of demand; and
(2) rising productivity. Economist Albert Otto Hirschman studied dominant
industries from the angle of mutual connection of different industries. He believed
that some complementary relationships between certain industries are stronger than
those between other industries.
The government locates “strategic” or dominant industries in the economic system
in places where the chain connections of input-output are much stronger and much
more intimate than those in other places. Therefore, in the development of a
132 6 Grey Clustering Evaluation Models

regional economy, individual industries play different roles and functions in the
regional industrial system, in which one or several industries control the leading
position. These industries constitute the leading industries and form the group of
dominant industries of the region. Each development of modern regional economies
is in fact represented by the development of dominant industries. The impacts of
dominant industries on regional economic growth can be seen in three aspects. One
is their forward looking influence, that is, the impact of dominant industries on
industries that use the products of dominant industries. The second is their back-
ward looking influence, that is, the impact the dominant industries exert on
industries from which they receive their production materials. The third is the side
influence, that is, the impact of dominant industries on the development and
prosperity of the regional economy. Through these three areas of impact, dominant
industries project their industrial superiority onto other regional industries so that
the overall regional economy is positively affected. Therefore, there is a need to first
locate the dominant industries of a specific time period and region. Then, by further
developing these dominant industries, one can expect coordinated development of
various economic sectors of the specific region.
Based on the theory of dominant industries, each regional industrial system can
be deconstructed into dominant industries, the serving industries that are established
to provide support for the smooth development of the dominant industries, and
other potential leaders existing among other general industries. In short, the basic
idea for adjusting and optimizing the regional structure of industries is to correctly
select and optimize the development of the dominant industries so that they can
effectively carry out the task of dividing and forming specialized economic zones
across the region of their coverage. It is possible to improve dominant industries’
capability to bring about the healthy development of other regional industries, and
to help develop relevant industries, especially those forwardly impacted by domi-
nant industries. This can be done by purposefully prolonging the product chain as
much as possible, and by maintaining and strengthening existing supporting eco-
nomic sectors while increasing these sectors’ quality. Other actions include actively
developing basic economic infrastructures while removing the “bottlenecks” that
place constraints on the development of the regional economy, and supporting other
general business activities so that, when the current dominant industries are
weakened due to the changing environment, new leading industries can emerge in a
timely fashion in order to keep the regional economy in its normal cycle of
development without interruption.
Because dominant regional industries determine the formation and evolution of
the region’s economic system, in order to promote the healthy growth of the
regional economy it is important to clearly identify these industries, and to gain a
good understanding of their relationships with other relevant industries, supporting
economic sectors, and more general economic areas. To a degree, the formation and
evolution of dominant regional industries are influenced by the region’s economic
structure, market preferences, and the availability of resources. Therefore, the
6.6 Practical Applications 133

encouragement and promotion of the development of dominant industries are both

the start and the ultimate goal of introducing regional economic policies. When
selecting dominant industries, we should comply with the following: principle of
resource superiority, principle of market demand-supply, principle of economic
benefits, principle of related benefits, principle of industrial superiority, principle of
technological advancement, principle of sustainable development, principle of
sufficient employment, principle of giving priority to the development of “bottle-
neck” areas, among others.
The characteristics of dominant regional industries and the principles of how to
select dominant regional industries provide us with a clear direction on how to
pinpoint particular regional industries as the leaders. Considering the specific cir-
cumstances of Jiangsu Province, People’s Republic of China, we choose the fol-
lowing factors as the system of evaluation criteria for determining the dominant
regional industries of the province.
(1) Degrees of industrial incidence
The degrees of industrial incidence contain two parts: degree of forward incidence
and degree of backward incidence. Both of these concepts represent the degrees of
influence, also known as the spreading effect, on the changes of the input-output of
other industries, as directly or indirectly affected by changes in the demand and
supply of one industry. The most commonly employed degrees of influence are the
sensitivity coefficient and the influence coefficient. Assume that bij is the (i, j)-entry
in the complete consumption coefficient matrix B studied in the input-output
analysis, and n the number of defined industries. Then, the sensitivity coefficient li
of the ith industry is defined by:
Pn
j¼1 bij
li ¼ 1 Pn Pn ð6:16Þ
n i¼1 j¼1 bij

The magnitude of li reflects the degree of influence of an increase in the output

of each industry on the output of the ith industry. It is a degree of forward incidence.
The greater the sensitivity coefficient li , the better the forward incidence capability
of the ith industry.
The influence coefficient cj of the jth industry is defined by:
Pn
bij
P Pn
cj ¼ 1 n i¼1 ð6:17Þ
n i¼1 j¼1 bij

The magnitude of cj reflects the influence of an increase in the demand of the jth
industry on the output of other industries. It is a degree of backward incidence. The
greater the influence coefficient cj , the stronger the capability of backward incidence
of the jth industry. If cj > 1, it means that the effect of this industry is above the
average among all industries. If cj = 1, it means that the effect of this industry is
134 6 Grey Clustering Evaluation Models

right at the mean level of all industries. If cj < 1, it means that the effect of this
industry is below the mean level. Similar explanations can be given to li .
(2) Income Elasticity of Demand
Assume Y is the domestic GDP and xi the market demand for the products of the ith
industry. Then the income elasticity ei of demand of the ith industry is given by:

y @xi
ei ¼ ð6:18Þ
xi @y

This formula represents the effect of changes in the domestic GDP on the
demand of the products of the ith industry. Generally, industries with large income
elasticity of demand have strong capability of market expansion. When ei < 1, it
means that the expanding rate of demand of the ith industry is smaller than the rate
of increase of the national GDP. When ei > 1, it means that the expanding rate of
demand of the ith industry is greater than the rate of increase of the national
GDP. When ei is equal to or close to 1, it implies that the expanding rate of demand
of the ith industry is about the national average of increase. A regional dominant
industry should have a relatively large income elasticity of demand.
(3) The Growth
Assume that x0i is the initial demand for input of the ith industry, and ri the average
rate of increase. Then the demand for input of the ith industry is given by:

xti ¼ x0i ð1 þ ri Þt

The greater the ri value, the faster the ith industry grows, and the more important
its position and role in the regional economic system.
(4) Employment Rate
Given that China is currently experiencing a huge amount of surplus labor,
employment has become the most important problem at all levels of the govern-
ment. The creation of a scenario of sufficient employment has been one of the main
foci of the governmental effort to adjust the industrial structure. To this end, let us
employ the integrated employment coefficient to evaluate the absorption capability
of each industry. Assume that vi stands for the labor wage of the ith industry, qi the
total output of the ith industry, and cij the Wassily Leontief inverse matrix coeffi-
cient as listed in the input-output table. Then the integrated employment coefficient
bi of the ith industry is given as follows:

X
n
vi
bi ¼ cij ð6:19Þ
j¼1
qi

The magnitude of bi reflects the amount of labor needed to directly and indi-
rectly produce 1 unit of the final product by the ith industry and its relevant
industries. In the remainder of this case study, we look at the actual selection of
6.6 Practical Applications 135

dominant industries in Jiangsu Province in China. Based on the 1997 input-output

table of Jiangsu Province and the yearly statistical almanacs, we classified all
industries in Jiangshu Province into three classes: dominant, supporting, and gen-
eral industries, respectively, using the grey evaluation method of fixed weights:
(1) Based on the 1997 input-output table, we calculated the sensitivity coefficient,
influence coefficient, and integrated employment coefficient for each industry;
and
(2) By using the annual statistical almanacs, we computed the income elasticity of
demand and rates of growth.
Because the industries listed in the annual almanacs are different to those used in
the input-output table, we combined the 40 industries as shown in the input-output
table into 34 industries. The specific data values of the criteria are provided in
Table 6.9.

Table 6.9 Computational results of all the criteria for 34 industries

Name Sensitivity Influence Income Growth Overall
coefficient coefficient elasticity of rate ri employment
li ci demand ei rate bi
Agriculture 0.9633 0.6756 0.4488 1.8615 0.7286
Coal mining 0.8889 0.6876 −0.10524 −1.13421 0.5936
Extraction of 0.8199 0.3589 2.173455 1 0.3528
petroleum and −7.73033
r natural gas
Metal ore mining 0.2161 0.7366 −0.05618 −0.60063 0.6717
Non-metal ore 0.6359 0.7649 −0.30407 −3.39012 0.4226
mining
Food and tobacco 0.8249 1.0321 0.178727 1.842096 0.5096
processing
Textile 1.2055 1.2583 0.385003 3.850923 0.3982
Clothing, leather, 0.2598 1.1266 0.474869 4.690773 0.6244
down, and fabrics
Lumber processing, 0.2896 1.0386 0.961863 8.923825 0.4542
furniture making
Paper making, 0.9162 1.1368 1.099007 10.0312 0.4746
printing, office
supplies
Petroleum refinery, 1.2970 0.9357 1.239179 11.1295 0.3630
coking
Chemicals 3.3553 1.0795 1.724493 14.70029 0.3834
Manufactures using 1.1903 1.0556 1.220044 10.98148 0.4596
non-metal minerals
(continued)
136 6 Grey Clustering Evaluation Models

Table 6.9 (continued)

Name Sensitivity Influence Income Growth Overall
coefficient coefficient elasticity of rate ri employment
li ci demand ei rate bi
Metal smelting, 4.7383 1.4733 0.142253 1.474251 0.4607
rolling processing
Metal products 0.8404 1.3129 1.875618 15.74695 0.4002
Machinery 1.2090 1.2996 0.867199 8.139289 0.4296
Transportation 0.6622 1.5949 0.70133 6.721973 0.4370
equipment
Electric machinery, 0.8473 1.4695 0.822944 7.766583 0.4154
equipment
Electronic, 1.1501 1.2383 1.078045 9.864103 0.3957
communication
equipment
Instruments, meters, 0.1256 1.2873 3.107475 23.37564 0.4362
and office appliances
Other manufacturing 1.1311 1.0108 1.921372 16.0583 0.5896
Supply of 1.6183 0.7644 1.323333 11.77354 0.4471
Electricity, steam
hot water
Production, supply 0.0946 1.4203 2.832106 21.79184 0.6892
of coal and gas
Production, supply 0.1208 0.7659 −0.0766 −0.82168 0.5050
of tap water
Construction 0.3474 1.1201 1.621324 13.96896 0.5311
Transportation, 1.8850 0.6446 1.75804 14.9351 0.4722
postal service
Food services 3.1201 0.7710 0.935783 8.709375 0.5482
Financial, insurance 1.1328 0.5166 0.468103 4.628241 0.2747
Real estate 0.2238 0.9941 1.606129 13.86004 0.4061
Societal Services 1.1264 0.7829 2.027072 16.76815 0.5008
Health, sports, 0.2152 1.2420 1.624586 13.99229 0.5205
welfare
Education, arts, 0.1110 0.7258 2.370946 18.99219 0.7162
radio, TV, movies
Service to scientific 0.1969 0.8553 1.451781 12.73562 0.5796
research, technology
Government and 0.2412 0.9139 1.72827 14.7268 0.5478
others

According to the grey fixed weight clustering model and the observed values of
the criteria, let the possibility function of the kth subclass of the jth criterion be
fjk ðÞ: Through using a Delphi postal mail survey, we determined the types of
possibility functions and weights of the individual criteria used in this
6.6 Practical Applications 137

comprehensive evaluation. In particular, the possibility functions of the sensitivity

coefficient, influence coefficient, income elasticity coefficient, growth rate, and
integrated employment coefficient are respectively given below:

f11 ½0:9; 1:3; ; f12 ½0:6; 0:9; ; 1:2; f13 ½; ; 0:5; 1

f21 ½0:9; 1:3; ; ; f22 ½0:5; 0:9; ; 1:3; f23 ½; ; 0:4; 0:9

f31 ½1; 1:6; ; ; f32 ½0:5; 1; 1:5; f33 ½; ; 0:4; 1

f41 ½6; 12; ; ; f42 ½3; 7; ; 11; f43 ½; ; 4; 10

f51 ½0:3; 0:5; ; ; f52 ½0:25; 0:45; ; 0:7; f53 ½; 0:25; 0:4

The corresponding weights for these criteria are, respectively: 0.14, 0.14, 0.24,
0.24, and 0.24. The computed comprehensive evaluation results of all industries are
given in Table 6.10.

Table 6.10 Comprehensive evaluation results of all 34 industries

Code Name Dominant Supporting General MAX Clustering
results
1 Agriculture 0.2621 0.1719 0.5335 0.5335 General
2 Coal mining 0.24 0.3026 0.5705 0.5705 General
3 Extraction of petroleum 0.5433 0.2259 0.2659 0.5433 Dominant
and natural gas
4 Metal ore mining 0.24 0.1099 0.6657 0.6657 General
5 Non-metal ore mining 0.1471 0.3165 0.6197 0.6197 General
6 Food and tobacco 0.2862 0.3815 0.529 0.529 General
processing
7 Textile 0.3501 0.2434 0.4828 0.4828 General
8 Clothing, leather, down, 0.3193 0.2347 0.5624 0.5624 General
and fabrics
9 Lumber processing, 0.3505 0.6737 0.1983 0.6737 Supporting
furniture making
10 Paper making, printing, 0.4989 0.6565 0.0234 0.6565 Supporting
office supplies
11 Petroleum refinery, coking 0.5278 0.3882 0.0592 0.5278 Dominant
12 Chemicals 0.7829 0.2372 0.0265 0.7829 Dominant
13 Manufactures using 0.6348 0.4563 0 0.6348 Dominant
non-metal minerals
14 Metal smelting, rolling 0.4728 0.2297 0.48 0.4728 Dominant
processing
(continued)
138 6 Grey Clustering Evaluation Models

Table 6.10 (continued)

Code Name Dominant Supporting General MAX Clustering
results
15 Metal products 0.7402 0.2924 0.0446 0.7402 Dominant
16 Machinery 0.4891 0.5635 0.1275 0.5635 Supporting
7 Transportation equipment 0.3332 0.5733 0.3451 0.5733 Supporting
18 Electric machinery, 0.34914 0.6629 0.2029 0.6629 Supporting
equipment
19 Electronic, communication 0.5065 0.4904 0.012 0.5065 Dominant
equipment
20 Instruments, meters, and 0.7789 0.2278 0.14 0.7789 Dominant
office appliances
21 Other manufacturing 0.8396 0.2393 0 0.8396 Dominant
22 Supply of Electricity, 0.6767 0.4138 0.0379 0.6767 Dominant
steam hot water
23 Production, supply of coal 0.86 0.001 0.14 0.86 Dominant
and gas
24 Production, supply of tap 0.24 0.2802 0.6575 0.6575 General
water
25 Construction 0.797 0.2251 0.14 0.797 Dominant
26 Transportation, postal 0.8266 0.2692 0.0715 0.8266 Dominant
service
27 Food services 0.4883 0.5871 0.1134 0.5871 Supporting
28 Financial, insurance 0.0814 0.1645 0.7354 0.7354 General
29 Real estate 0.6402 0.2943 0.14 0.6402 Dominant
30 Societal Services 0.7992 0.3245 0.0327 0.7992 Dominant
31 Health, sports, welfare 0.8397 0.1926 0.14 0.8397 Dominant
32 Education, arts, radio, TV, 0.72 0.079 0.1887 0.72 Dominant
movies
33 Service to scientific 0.6607 0.263 0.1525 0.6607 Dominant
research, technology
34 Government and others 0.7248 0.2812 0.14 0.7248 Dominant

By analyzing these comprehensive evaluation results, we can obtain the fol-

lowing conclusions. Firstly, the industries that can possibly be considered as
dominant include: Extraction of petroleum and natural gas, Petroleum refinery,
coking, Chemicals, Manufactures using non-metal minerals, Metal smelting, rolling
processing, Metal products, Electronic and communication equipment, Instruments,
meters, and office appliances, Other manufacturing, Supply of Electricity and steam
hot water, Production and supply of coal and gas, Construction, Transportation and
postal service, Real estate, Societal Services, Health, sports, and welfare, Education,
arts, radio, TV, and movies, Service to scientific research and technology, and
Government and others.
6.6 Practical Applications 139

Secondly, the industries that can potentially be considered as supporting include:

Lumber processing and furniture making, Paper making, printing and office sup-
plies, Machinery, Transportation equipment, Electric machinery and equipment,
and Food services.
Thirdly, the class of general industries includes: Agriculture, Coal mining, Metal
ore mining, Non-metal ore mining, Food and tobacco processing, Textile, Clothing,
leather, down, and fabrics, and Financial and insurance.
One of the main tasks facing the economic development of Jiangsu Province is
to foster the desired economic growth through a transformation from the current
extensive industry type to a more concentrated and constrained industry type, and
through an elevation of the current industrial structure. One key success factor is to
correctly select and actively promote the development of dominant industries. By
taking advantage of these industries’ relatively strong capability to attract and attain
advanced technologies, their high degrees of inter-connections, and their superior
ability of penetration, the overall consumption of material and energy resources can
be reduced while product quality and attached commercial value are increased.
Therefore, elevations of the entire industrial structure can develop in a sustainable
fashion and create employment opportunities in large numbers. Each elevation of
the industrial structure also represents a process of replacing obsolete leading
industries with new ones. The existence of dominant industries is variable; it
changes and evolves along with development of the regional economy. When old
branded leaders are replaced by newcomers, the existing industrial structure also
changes accordingly. Therefore, when dominant industries are fostered and pro-
moted, government officers have to have an eye on the overall planning, correctly
manage the development relationships between various industries, and maintain the
coordinated development of the regional economy.
Chapter 7
Series of GM Models

7.1 Introduction

Model GM (1, 1) is the basic model of grey prediction theory and has been used
widely since its development in the early 1980s. Grey system theory is a new
methodology that focuses on problems involving small data and poor information. It
addresses uncertain systems with partially known information through generating,
excavating, and extracting useful information from what is available. The theory
enables a correct description of a system’s running behavior and its evolution law,
and thus generates quantitative predictions of future system changes. Incomplete and
inaccurate information is the basic characteristic of uncertainty systems. In a 2012
literature review by Liu and Forrest, the characteristics of uncertainty systems and
the role of the uncertainty models in uncertainty systems research was discussed. In
that review, the author clearly pointed out that pursuing a meticulous model in the
case of incomplete information and inaccurate data was impossible (Liu and Forrest
2012). Professor Zadeh’s incompatibility principle also clearly states that, when the
complexity of the system grows, our ability to make an accurate and significant
description of a system’s characteristics decreases until it reaches a threshold value
that, if it exceeded, accuracy and significance will become mutually exclusive
characteristics (Zadeh 1994). The incompatibility principle tells us that pursuing fine
one-sidedly will reduce the feasibility and significance of the results. A refined
model is not an effective means address complex systems. Many practical systems
examined by prominent scholars have incomplete information as their main char-
acteristics. This is prevalent in the real world, and uncertainty systems involving
small data and poor information prove to be rich resources for grey system theory.
In the last 30 years, much research has been carried out on the practical appli-
cations of Model GM (1, 1), and new research results emerge continuously. Recent
studies on model GM (1, 1) are focusing on how to further optimize the model and
improve its simulated and predictive results. Such research can be roughly divided
into the following areas: (1) research on the nature and characteristics of model GM

© Springer Science+Business Media Singapore 2017 141

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_7
142 7 Series of GM Models

(1, 1) (Ji et al. 2001); (2) studies about initial value selection (Dang et al. 2005);
(3) research on the optimization of model parameters (Xiao et al. 2000); (4)
attempts to improve the simulation accuracy of the model by recreating the back-
ground value (Tan 2005; Li and Dai 2004); (5) attempts to optimize the model
through different modeling methods (Song et al. 2002; Wang 2003); (6) research on
discrete model GM (1, 1) (Xie and Liu 2005); (7) modeling for non-equidistant
sequence and model optimization (Wang et al. 2008; Luo 2010); (8) research on the
application bound of different models (Liu and Deng 2010); (9) and research
combining a grey system model with other soft computing methods to improve the
accuracy of the model (Salmeron 2010; Zhang 2007).
The buffer operator proposed in 1991 (Liu 1991) has attracted much research
attention in recent years. Buffer operator is essentially a method for processing raw
data, rather than a technique to improve the degree of accuracy of the GM (1, 1)
model simulation and prediction. During the period of raw data collection, a system
is likely to suffer interference from external shocks, which means that such data will
be distorted and unlikely to reflect the operation of the system’s behavior. In such
case, researchers can choose or construct a suitable buffer operator by following a
qualitative analysis of the data to eliminate the impact of the distorted data sequence
and keep the true nature of the data.
The above-mentioned studies have played a positive role in improving the
degree of accuracy of the simulation and prediction of model GM (1, 1), and in
helping scholars engaged in applied research make appropriate selection and use of
grey prediction models.

7.2 The Four Basic Forms of GM (1, 1)

In this section we present definitions of four basic forms of model GM (1, 1),
including Even Grey Model (EGM), Original Difference Grey Model (ODGM),
Even Difference Grey Model (EDGM) and Discrete Grey Model (DGM). The
properties and characteristics of different models are discussed in-depth.

7.2.1 The Basic Forms of Model GM (1, 1)

Definition 7.2.1 Let X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ; xð0Þ ðkÞ  0; X ð1Þ be the
1-AGO sequence of X ð0Þ ; that is

X ð1Þ ¼ ½xð1Þ ð1Þ; xð1Þ ð2Þ;    ; xð1Þ ðnÞ

7.2 The Four Basic Forms of GM (1, 1) 143

Pk
where xð1Þ ðkÞ ¼ i¼1 xð0Þ ðiÞ, k ¼ 1; 2;    n Then

xð0Þ ðkÞ þ axð1Þ ðkÞ ¼ b ð7:1Þ

is referred to as the original form of model GM (1, 1), which is a difference

equation.
The parameter vector ^a ¼ ½a; bT of formula (7.1) can be estimated using the
least square method, which satisfies

^a ¼ ðBT BÞ1 BT Y ð7:2Þ

where
2 3 2 ð0Þ 3
 xð1Þ ð2Þ 1 x ð2Þ
6  xð1Þ ð3Þ 17 6 xð0Þ ð3Þ 7
6 7 6 7
B¼6 .. .. 7 ;Y ¼6 . 7 ð7:3Þ
4 . . 5 4 .
. 5
 xð1Þ ðnÞ 1 xð0Þ ðnÞ

Definition 7.2.2 Based on the original form of model GM (1, 1) and formula (7.2),
which is used to estimate the model’s parameters, then the model that takes the
solution of the original difference Eq. (7.1) as the time response formula is called
the original difference form of model GM (1, 1), and is referred to as Original
Difference Grey Model (ODGM) for short (Liu et al. 2015).
Definition 7.2.3 Let X ð0Þ ; X ð1Þ and, just like Definition 7.2.1, let

Z ð1Þ ¼ ðzð1Þ ð2Þ; zð1Þ ð3Þ;    ; zð1Þ ðnÞÞ;

where zð1Þ ðkÞ ¼ 12 ðxð1Þ ðkÞ þ xð1Þ ðk  1ÞÞ, then

xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b ð7:4Þ

is referred to as the even form of model GM (1, 1).

The even form of model GM (1, 1) is also essentially a difference equation. The
parameter vector of formula (7.4) can also be estimated with formula (7.2), but it
should be noted that the elements of matrix B are different from those in formula
(7.3), which is
2 3
 zð1Þ ð2Þ 1
6  zð1Þ ð3Þ 17
6 7
B¼6 .. .. 7 ð7:5Þ
4 . .5
 zð1Þ ðnÞ 1
144 7 Series of GM Models

Definition 7.2.4 The following differential equation

d xð1Þ
þ a xð1Þ ¼ b ð7:6Þ
dt

is called a whitenization(or image) equation of the even form xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b
of model GM (1, 1).
Definition 7.2.5 Replace matrix B of formula (7.2) with (7.5), according to
parameter vector ^a ¼ ½a; bT of the least squares estimator of (7.6) and the solution
of whitenization equation Eq. (7.6), and model the difference, differential hybrid
model of the time response formula of GM (1, 1). This is called the even hybrid
form of model GM (1, 1), and is referred to as Even Grey Model (EGM) for short
(Deng 1982, 1985).
Definition 7.2.6 The parameter a of Even GM (1, 1) is called development index
and b is called grey actuating quantity. The development index reflects the trend of
^xð1Þ and ^xð0Þ .
Even Model GM (1, 1) is the grey prediction model proposed firstly by Professor
Deng Julong, and is currently the most influential, widely used form. When
researchers mention model GM (1, 1) they are often referring to EGM.
Definition 7.2.7 Based on the even form of model GM (1, 1) and the estimated
model parameters, then the model that takes the solution of the even difference
Eq. (7.4) as the time response formula is called the even difference form of model
GM (1, 1), and is referred to as Even Difference Grey Model(EDGM) for short (Liu
et al. 2015).
Definition 7.2.8 The difference equation as follows

xð1Þ ðk þ 1Þ ¼ b1 xð1Þ ðkÞ þ b2 ð7:7Þ

is called a discrete form of model GM (1, 1), and is referred to as Discrete Grey
Model(DGM) for short (Xie and Liu 2005).
The parameter vector b^ ¼ ½b ; b T in Eq. (7.7) is similar to formula (7.2), where
1 2
2 3 2 3
xð1Þ ð1Þ 1 xð1Þ ð2Þ
6 xð1Þ ð2Þ 1 7 6 xð1Þ ð3Þ 7
6 7 6 7
B¼6 .. .. 7; Y ¼ 6 . 7
4 . . 5 4 .. 5
xð1Þ ðn  1Þ 1 xð1Þ ðnÞ

The four different models of GM (1, 1) use only the system’s behavior data
sequence to model the predictive models and belong to the simple and practical
modeling method with a single sequence. Time series data involve only the regular
time variable, whereas horizontal sequence data involves only a regular variable
with a target number. Without involving the other explanatory variables, the
7.2 The Four Basic Forms of GM (1, 1) 145

application of this modeling method is relatively simple and can be found out the
practical changing information. This method is widely used.

7.2.2 Properties and Characteristics of the Basic Model

Theorem 7.2.1 The time response sequence of the Even Model GM (1, 1) is as
follow:

b b
^xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þ eaðk1Þ þ ; k ¼ 1; 2;    n ð7:8Þ
a a
d xð1Þ
Proof The solution of whitenization equation dt þ a xð1Þ ¼ b is

b
xð1Þ ðtÞ ¼ C eat þ : ð7:9Þ
a

When t = 1, we let xð1Þ ð1Þ ¼ xð0Þ ð1Þ, and feed into Eq. (7.9); we can obtain
C ¼ ½xð0Þ ð1Þ  ba ea . After that we take C into Eq. (7.9) and can get

b b
^xð1Þ ðtÞ ¼ ðxð0Þ ð1Þ  Þ eaðt1Þ þ : ð7:10Þ
a a

Equation (7.8) is the discrete form of Eq. (7.10). From Eq. (7.8)’s regressive
reduction formula

^xð0Þ ðkÞ ¼ að1Þ ^xð1Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ; k ¼ 1; 2;    n;

we can obtain the time response formula of X(0), that is

b
^xð0Þ ðkÞ ¼ ð1  ea Þðxð0Þ ð1Þ  Þ eaðk1Þ ; k ¼ 1; 2;    n ð7:11Þ
a
Theorem 7.2.2 The time response formula of formula (7.7) of the Discrete
Model GM (1, 1) is

b2 b2
^xð1Þ ðkÞ ¼ ½xð0Þ ð1Þ  bk1 þ ð7:12Þ
1  b1 1  b1

Proof The general solution of the difference Eq. (7.13)

xð1Þ ðk þ 1Þ ¼ Axð1Þ ðkÞ þ B ð7:13Þ

146 7 Series of GM Models

is

B
xð1Þ ðkÞ ¼ CAk þ ; ð7:14Þ
1A

where C is an arbitrary constant and can be defined by the initial conditions.

Formula (7.7) and (7.14) are exactly the same difference equation. Let
A ¼ b1 ; B ¼ b2 , then

b2
xð1Þ ðkÞ ¼ Cbk1 þ : ð7:15Þ
1  b1

When k = 0, let xð1Þ ð0Þ ¼ xð0Þ ð1Þ, and feed into formula (7.15), we can get
b2
C ¼ ½xð0Þ ð1Þ  1b . Then take C into formula (7.15) and we can obtain formula
1
(7.12).
From formula (7.12)’s regressive reduction formula

^xð0Þ ðkÞ ¼ að1Þ ^xð1Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ; k ¼ 1; 2;    n;

we can obtain the time response formula of X(0), that is

b2
^xð0Þ ðkÞ ¼ ðb1  1Þ½xð0Þ ð1Þ  bk1 : ð7:16Þ
1  b1 1

Theorem 7.2.3 The time response formula of Original Difference Model GM (1, 1)
is

b 1 k b
^xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þð Þ þ ð7:17Þ
a 1þa a
Proof From the original form (7.1) of model GM (1, 1) we can get

xð1Þ ðk þ 1Þ  xð1Þ ðkÞ þ axð1Þ ðk þ 1Þ ¼ b: ð7:18Þ

After transposition, we obtain

1 b
xð1Þ ðk þ 1Þ ¼ ð Þxð1Þ ðkÞ þ :
1þa 1þa

Contrast with the difference Eq. (7.13), when we feed A ¼ 1 þ1 a ; B ¼ 1 þb a into

Eq. (7.14), we can obtain

1 k b
xð1Þ ðkÞ ¼ Cð Þ þ ð7:19Þ
1þa a
7.2 The Four Basic Forms of GM (1, 1) 147

When k = 0, let xð1Þ ð0Þ ¼ xð0Þ ð1Þ, feed into formula (7.18) and get
C ¼ ½xð0Þ ð1Þ  ba. Then we feed C into formula (7.19) and can obtain formula
(7.17).
From formula (7.17)’s regressive reduction formula

^xð0Þ ðkÞ ¼ að1Þ ^xð1Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ; k ¼ 1; 2;    n;

we can obtain the time response formula of X(0), which is

b 1 k b b 1 k1 b
^xð0Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þð Þ þ  ½xð0Þ ð1Þ  ð Þ þ :
a 1þa a a 1þa a

That is

b 1 k
^xð0Þ ðkÞ ¼ ðaÞðxð0Þ ð1Þ  Þð Þ ð7:20Þ
a 1þa
Theorem 7.2.4 The time response formula of Even Difference Model GM (1, 1) is

b 1  0:5a k b
xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þð Þ þ ð7:21Þ
a 1 þ 0:5a a
Proof From the even form (7.4) of model GM (1, 1) we can get

xð1Þ ðk þ 1Þ þ xð1Þ ðkÞ

xð1Þ ðk þ 1Þ  xð1Þ ðkÞ þ a ¼ b:
2

After transposition, we obtain

1  0:5a ð1Þ b
xð1Þ ðk þ 1Þ ¼ x ðkÞ þ :
1 þ 0:5a 1 þ 0:5a

Contrast with the difference Eq. (7.13), and feed A ¼ 110:5a

þ 0:5a ; B ¼ 1 þ 0:5a into
b

formula (7.14). We can obtain

2a k b
xð1Þ ðkÞ ¼ Cð Þ þ ð7:22Þ
2þa a

When k = 0, let xð1Þ ð0Þ ¼ xð0Þ ð1Þ, feed it into formula (7.22) and get
C ¼ ½xð0Þ ð1Þ  ba. Then feed C into formula (7.22) and we can obtain formula
(7.21).
From formula (7.21)’s regressive reduction formula

^xð0Þ ðkÞ ¼ að1Þ ^xð1Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ; k ¼ 1; 2;    n;

148 7 Series of GM Models

we can obtain the time response formula of X(0), which is

b 1  0:5a k b b 1  0:5a k1 b

^xð0Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þð Þ þ  ½ðxð0Þ ð1Þ  Þð Þ þ :
a 1 þ 0:5a a a 1 þ 0:5a a

That is

a b 1  0:5a k
^xð0Þ ðkÞ ¼ ð Þðxð0Þ ð1Þ  Þð Þ ð7:23Þ
1  0:5a a 1 þ 0:5a
Lemma 7.2.1 When a ! 0 þ ; 1  0:5a
1 þ 0:5a  ea .

Proof The Maclaurin expansions of ea and 1  0:5a

1 þ 0:5a are as follows:

a2 a3 an
ea ¼ 1  a þ  þ    þ ð1Þn þ oðan Þ
2! 3! n!
1  0:5a a2 a3 an þ 1
¼ 1  a þ  2 þ    þ ð1Þn þ 1 n þ oðan þ 1 Þ
1 þ 0:5a 2 2 2

As n = 3, then there is D ¼ ea  11 þ

 0:5a 3
a3 3
0:5a ¼  6 þ ¼ 12
a a
4 , therefore, when
þ 1  0:5a a
a ! 0 , 1 þ 0:5a  e .
Theorem 7.2.5 When a ! 0 þ , Even Model GM (1, 1) and Discrete Model GM
(1, 1) are equivalent.
Proof From the even form (7.4) of Model GM (1, 1)

1  0:5a ð1Þ b
xð1Þ ðk þ 1Þ ¼ ð Þx ðkÞ þ ;
1 þ 0:5a 1 þ 0:5a
 0:5a
and contrast with the discrete form (7.7), we can obtain b1 ¼ 11 þ 0:5a ; b2 ¼ 1 þ 0:5a
b

and

2ð1  b1 Þ 2b2 b b2
a¼ ;b ¼ ; ¼ : ð7:24Þ
1 þ b1 1 þ b 1 a 1  b1

b2
Take ba ¼ 1b into formula (7.8), we can get
1

b2 b2
^xð1Þ ðkÞ ¼ ½xð0Þ ð1Þ  eaðk1Þ þ ; k ¼ 1; 2;    n ð7:25Þ
1  b1 1  b1

It is known from Lemma 7.2.1 that when a ! 0 þ , therefore, Even Model GM

(1, 1) and Discrete Model GM (1, 1) are equivalent.
Analogously, we can prove that when a ! 0 þ , the four basic forms of model
GM (1, 1), namely Even Model GM (1, 1) (EGM), Original Difference Model GM
(1, 1) (ODGM), Even Difference Model GM (1, 1)(EDGM) and Discrete
7.2 The Four Basic Forms of GM (1, 1) 149

Model GM (1, 1) (DGM) are pairwise equivalent. However, the degree of

approximation between different forms is a difference. This difference leads to
different forms of Model GM (1, 1) being suitable for different situations, and it also
offers a variety of possible options for the actual modeling process.
Theorem 7.2.6 Original Difference Model GM (1, 1) (ODGM), Even Difference
Model GM (1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM) can all accurately
simulate homogeneous exponential sequences.
Since the time response formulas of Original Difference Model GM (1, 1)
(ODGM), Even Difference GM (1, 1) model (EDGM) and Discrete GM (1, 1)
model (DGM) are all geometric sequences, they can accurately simulate homoge-
neous exponential sequences.
In the basic forms of GM (1, 1), the development coefficient (a) reflects the
development states of ^xð1Þ and ^xð0Þ . In general, the variables that act upon the system
of interest should be external or pre-defined. Because GM (1, 1) is a kind of model
constructed on a single sequence, it uses only the behavioral sequence (also referred
to as output sequence or background values) of the system without considering any
externally acting sequences (also referred to as input sequences, or driving quan-
tities). The grey action quantity b in the basic forms of GM (1, 1) is a value derived
from the background values. It reflects changes contained in the data and its exact
intension is grey. This quantity realizes the extension of the relevant intension. Its
existence distinguishes grey systems modeling from the general input-output (or
black-box) modeling. It is also an important test stone of separating the thoughts of
grey systems and those of grey boxes.

7.3 Suitable Ranges of Different GM (1, 1)

The suitable sequences of different basic models of GM (1, 1) (Liu 2015) and the
applicable ranges of EGM (Liu and Deng 2000) are studied by simulation and
analysis with homogeneous exponential sequences, non-exponential increasing
sequences, and vibration sequences. It can provide reference and a basis for people
to choose the correct model in the actual modeling process.

7.3.1 Suitable Sequences of Different GM (1, 1)

For further study of the suitable sequences of four basic forms of model GM (1, 1),
we let a = 0.01, 0.02, 0.03, 0.04, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45,
0.5, 0.55, 0.6, 0.65, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.5, 1.8 and conduct simulation
analysis, respectively. Let k = 1, 2, 3, 4, 5, with the homogeneous exponential
150 7 Series of GM Models

ð0Þ
function xi ðkÞ ¼ eak , and accurate to six decimal places. Then we can get the
corresponding sequences as follows:

ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ

 a ¼ 0:01; X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5ÞÞ
¼ ð1:010050; 1:020201; 1:030455; 1:040811; 1:051271Þ
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
 a ¼ 0:02; X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð4Þ; x2 ð5ÞÞ
¼ ð1:020201; 1:040811; 1:061837; 1:083287; 1:105171Þ
...................................................
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
 a ¼ 1:8; X25 ¼ ðx25 ð1Þ; x25 ð2Þ; x25 ð3Þ; x25 ð4Þ; x25 ð5ÞÞ
¼ ð6:049647; 36:59823; 221:4064; 1339:431; 8103:084Þ:

ð0Þ ð0Þ ð0Þ

We use X1 , X2 , ⋯, X25 as the original data to establish Even Model GM (1, 1)
(EGM), Original Difference Model GM (1, 1) (ODGM), Even Difference Model
GM (1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM), respectively. We can
find that Original Difference Model GM (1, 1) (ODGM), Even Difference
Model GM (1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM) can accurately
simulate homogeneous exponential sequence, which confirms the conclusions of
ð0Þ
Theorem 7.2.6 once again. Using Even Model GM (1,1) (EGM) to simulate X1 ,
ð0Þ ð0Þ
X2 , ⋯, X25 , it is fond that with the increasing of a, the error will also increase.
Table 7.1 shows the average relative error using four kinds of model GM (1, 1) to
ð0Þ ð0Þ ð0Þ
simulate the homogeneous exponential sequence X1 , X2 , ⋯, X25 .
In Table 7.1, we can see that the small errors of Original Difference Model GM
(1, 1) (ODGM), Even Difference Model GM (1, 1) (EDGM) and Discrete
Model GM (1, 1) (DGM) which simulate the homogeneous exponential sequence
are all caused by round-off errors. In fact, the three models can all accurately
simulate the homogeneous exponential sequence.
Then, we limited the range of random numbers at first, and got the non-exponential
ð0Þ ð0Þ ð0Þ
increasing sequence Y1 , Y2 , ⋯, Y25 randomly generated by the homogeneous
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
exponential sequence X1 , X2 , ⋯, X25 , along with the vibration sequence Z1 , Z2 ,
ð0Þ ð0Þ ð0Þ
⋯, With that, when k ¼ 2; 3;    ; 5,
Z25 . zi ðkÞ\zi ðk
 1Þ; i ¼ 1; 2;    ; 25 will
arise in the sequence data but there is a growth trend as a whole, both are equally
accurate to six decimal places. Then we build Even Model GM (1, 1) (EGM), Original
Difference Model GM (1, 1) (ODGM), Even Difference Model GM (1, 1) (EDGM)
ð0Þ ð0Þ ð0Þ ð0Þ
and Discrete Model GM (1, 1) (DGM) using sequences Y1 , Y2 , ⋯, Y25 and Z1 ,
ð0Þ ð0Þ
Z2 , ⋯, Z25 respectively. The errors we can see are in Tables 7.2 and 7.4. Due to
limited space, the generating data are not shown here.
7.3 Suitable Ranges of Different GM (1, 1) 151

Table 7.1 The simulation errors of the homogeneous exponential sequence of four kinds of GM
(1, 1) (%)
Code a EGM DGM ODGM EDGM
ð0Þ 0.01 0.000849 0.000027 0.000027 0.000027
X1
ð0Þ 0.02 0.003468 0.000013 0.000013 0.000013
X2
ð0Þ 0.03 0.007951 0.000018 0.000018 0.000018
X3
ð0Þ 0.04 0.014403 0.000004 0.000004 0.000003
X4
ð0Þ 0.05 0.022922 0.000016 0.000016 0.000016
X5
ð0Þ 0.10 0.100058 0.000008 0.000008 0.000008
X6
ð0Þ 0.15 0.244034 0.000009 0.000009 0.000009
X7
ð0Þ 0.20 0.467588 0.000003 0.000003 0.000007
X8
ð0Þ 0.25 0.783590 0.000005 0.000005 0.000006
X9
ð0Þ 0.30 1.205144 0.000004 0.000004 0.000010
X10
ð0Þ 0.35 1.745610 0.000006 0.000006 0.000010
X11
ð0Þ 0.40 2.418758 0.000004 0.000004 0.000010
X12
ð0Þ 0.45 3.238864 0.000007 0.000007 0.000008
X13
ð0Þ 0.50 4.220851 0.000011 0.000011 0.000008
X14
ð0Þ 0.55 5.380507 0.000003 0.000003 0.000003
X15
ð0Þ 0.60 6.734574 0.000016 0.000016 0.000011
X16
ð0Þ 0.65 8.301040 0.000009 0.000009 0.000006
X17
ð0Þ 0.70 10.099355 0.000021 0.000021 0.000021
X18
ð0Þ 0.80 14.478513 0.000015 0.000015 0.000015
X19
ð0Þ 0.90 20.068449 0.000016 0.000016 0.000022
X20
ð0Þ 1.00 27.110835 0.000047 0.000047 0.000047
X21
ð0Þ 1.10 35.908115 0.000040 0.000040 0.000035
X22
ð0Þ 1.20 46.844843 0.000105 0.000105 0.000105
X23
ð0Þ 1.50 98.188500 0.000129 0.000129 0.000129
X24
ð0Þ
X25 1.80 – 0.000433 0.000433 0.000433

From Table 7.2 we can see that four kinds of model GM (1, 1) can all simulate
the non-exponential increasing sequence to a certain degree. Generally speaking,
the simulation error will increase with the increasing of the development index. In
most cases, the simulation error of the difference, differential hybrid form of Even
Model GM (1, 1) (EGM), is smaller than that of the three discrete forms of Original
Difference Model GM (1, 1) (ODGM), Even Difference Model GM (1, 1) (EDGM)
and Discrete Model GM (1, 1) (DGM). As the non-exponential increasing sequence
152 7 Series of GM Models

Table 7.2 The simulation errors of the non-exponential increasing sequence of four kinds of GM
(1, 1) (%)
Code ; a; þ EGM DGM ODGM EDGM
ð0Þ 0.01 0.030994 0.030429 0.030432 0.030430
Y1
ð0Þ 0.02 0.658978 0.659039 0.660095 0.659572
Y2
ð0Þ 0.03 0.495833 0.495773 0.495768 0.495770
Y3
ð0Þ 0.04 1.010474 1.010308 1.010329 1.010319
Y4
ð0Þ 0.05 1.550886 1.550331 1.550468 1.550401
Y5
ð0Þ 0.10 1.626294 1.704980 1.690324 1.697211
Y6
ð0Þ 0.15 1.343565 1.457800 1.458993 1.458442
Y7
ð0Þ 0.20 5.155856 5.100486 5.229480 5.171925
Y8
ð0Þ 0.25 4.353253 4.893857 4.743792 4.808361
Y9
ð0Þ 0.30 4.736323 5.345755 5.168529 5.244168
Y10
ð0Þ 0.35 5.236438 5.377225 5.192273 5.269577
Y11
ð0Þ 0.40 3.603875 4.166958 4.044567 4.096904
Y12
ð0Þ 0.45 12.834336 15.364230 13.520184 14.246584
Y13
ð0Þ 0.50 7.396770 8.276073 7.878017 8.044898
Y14
ð0Þ 0.55 10.218727 10.084749 10.188912 10.143461
Y15
ð0Þ 0.60 21.073070 23.709858 21.610863 22.440905
Y16
ð0Þ 0.65 6.637022 7.906483 7.629068 7.731359
Y17
ð0Þ 0.70 9.088900 11.000565 10.479505 10.677398
Y18
ð0Þ 0.80 21.156265 30.606589 28.554915 29.245194
Y19
ð0Þ 0.90 14.441947 20.378328 17.104000 18.188008
Y20
ð0Þ 1.00 11.685913 18.463203 17.357496 17.734931
Y21
ð0Þ 1.10 13.011857 20.620317 19.396248 19.782271
Y22
ð0Þ 1.20 17.176472 27.929743 26.163490 26.624283
Y23
ð0Þ 1.50 26.327218 51.915584 50.006882 50.471089
Y24
ð0Þ 1.80 62.460946 75.503705 73.434001 74.070128
Y25

is closer to the homogeneous exponential sequence, the simulation accuracy of the

three discrete models is higher than. When the non-exponential increasing sequence
is close to the homogeneous exponential sequence to a certain extent, the simulation
accuracy of the discrete models will be smaller than that of Even Model GM (1, 1)
(EGM). From the simulation results of the three discrete models GM (1, 1), we can
see that with the increasing of the development coefficient, the simulation accuracy
of Original Difference Model GM (1, 1) (ODGM), and Even Difference Model GM
7.3 Suitable Ranges of Different GM (1, 1) 153

Table 7.3 Statistics for Error sorting EGM DGM ODGM EDGM
sorting the simulation error of
the non-exponential 1 18 5 2 0
increasing sequence of four 2 2 2 15 6
kinds of model GM (1, 1) 3 0 1 5 19
4 5 17 3 0

(1, 1) (EDGM) is higher than that of Discrete Model GM (1, 1) (DGM) in most
cases. The statistics for sorting the simulation error of different models with the
ð0Þ ð0Þ ð0Þ
sequence Y1 , Y2 , …, Y25 in ascending order are presented in Table 7.2.
Table 7.3 shows the statistical results.
As can be seen from Table 7.3, among the four kinds of models, Even
Model GM (1, 1) (EGM) is the most suitable for modeling with a non-exponential
increasing sequence, followed by the Original Differential Model GM (1, 1)
(ODGM) and Even Difference Model GM (1, 1) (EDGM). The error is slightly
larger when using the Discrete Model GM (1, 1) (DGM) to simulate the
non-exponential increasing sequence.
In theory, any simple model which describes a monotonous trend struggles to
describe a change in the vibration sequence. Therefore, we add the limiting con-
ð0Þ
dition of the random number, then the research range is the vibration sequence Z1 ,
ð0Þ ð0Þ ð0Þ ð0Þ
Z2 , …, Z25 . With that, there are some of k ¼ 2; 3;    ; 5, zi ðkÞ\zi ðk  1Þ;
i ¼ 1; 2;    ; 25 will arise in the sequence data, but there is a growth trend as a
whole. We can see from Table 7.4 that, for this specific vibration sequence, the
simulation error of the four kinds of models is significantly higher than the
non-exponential increasing sequence. Similar to the situation of the
non-exponential increasing sequence, in most cases the simulation error of Even
Model GM (1, 1) (EGM) to the vibration sequence is smaller than that of the three
discrete forms of Original Difference Model GM (1, 1) (ODGM), Even Difference
Model GM (1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM). For the vibration
sequence being close to the homogeneous exponential sequence, the simulation
error of the discrete model is smaller than one of the difference, differential hybrid
form of Even Model GM (1, 1) (EGM).
The statistics for sorting the simulation error of different models with the
ð0Þ ð0Þ ð0Þ
vibration sequence Z1 , Z2 , ⋯, Z25 in ascending order are presented in Table 7.4.
Table 7.5 shows the statistical results.
As can be seen in Table 7.5, of the four kinds of models the Even Model GM (1, 1)
(EGM) is more suitable for modeling with vibration sequence than the other three
discrete form models. The error using Discrete Model GM (1, 1) (DGM) to simulate
the vibration sequence is slightly larger than other two discrete form models.
The authors once tried to use the original form (7.1) of Model GM (1, 1) to
estimate the parameter vector ^a ¼ ½a; bT and, in accordance with the solution of
whitenization Eq. (7.6) along with the time response formula of Even Model GM
154 7 Series of GM Models

Table 7.4 Simulation errors of the vibration sequence of four kinds of model GM (1, 1)
Code ; a; þ EGM DGM ODGM EDGM
ð0Þ 0.01 0.298392 0.299400 0.299118 0.299258
Z1
ð0Þ 0.02 0.501223 0.505800 0.504877 0.505331
Z2
ð0Þ 0.03 0.369630 0.378773 0.379089 0.378935
Z3
ð0Þ 0.04 2.583662 2.586760 2.572109 2.579300
Z4
ð0Þ 0.05 2.928035 2.953655 2.899369 2.925619
Z5
ð0Þ 0.10 4.759929 4.791858 4.825226 4.807851
Z6
ð0Þ 0.15 3.802630 3.770562 3.776330 3.773545
Z7
ð0Þ 0.20 11.723459 11.946525 11.393483 11.642630
Z8
ð0Þ 0.25 14.895391 14.979357 15.229595 15.130729
Z9
ð0Þ 0.30 17.953543 17.992976 18.397577 18.241183
Z10
ð0Þ 0.35 7.299184 8.980062 8.537865 8.708603
Z11
ð0Þ 0.40 11.474779 11.519781 11.693309 11.619287
Z12
ð0Þ 0.45 11.988111 12.321804 12.261075 12.286039
Z13
ð0Þ 0.50 12.728220 11.753460 12.270432 12.038094
Z14
ð0Þ 0.55 10.636507 10.285910 10.897796 10.623904
Z15
ð0Þ 0.60 13.393234 13.515007 13.006751 13.227910
Z16
ð0Þ 0.65 15.420377 15.457643 14.690315 15.004381
Z17
ð0Þ 0.70 16.304197 16.365096 15.735103 15.998031
Z18
ð0Þ 0.80 14.542100 14.579829 14.110548 14.310293
Z19
ð0Þ 0.90 33.798587 33.160101 34.928437 34.293058
Z20
ð0Þ 1.00 22.586380 22.384127 22.016157 22.145609
Z21
ð0Þ 1.10 34.305920 34.481612 36.023522 35.484180
Z22
ð0Þ 1.20 23.591927 24.133298 23.323921 21.511839
Z23
ð0Þ 1.50 40.373380 40.475348 42.698005 41.917026
Z24
ð0Þ 1.80 30.380522 54.851229 45.724311 48.579850
Z25

Table 7.5 Statistics for Error sorting EGM DGM ODGM EDGM
sorting the simulation error of
the vibration sequence of four 1 12 4 8 1
kinds of model GM (1, 1) 2 1 7 6 11
3 9 1 2 13
4 3 13 9 0
7.3 Suitable Ranges of Different GM (1, 1) 155

(1, 1) (EGM), modeled the original Model GM (1, 1). After simulating the above
data we found that, even in cases where the development index is very small, the
simulation error was still comparatively large. Also, as the development index
increases, the simulation error increases rapidly. Based on even transformation of
the accumulation data to build the Even Model GM (1, 1), the simulation accuracy
improves greatly. Then a new method which can accurately simulate and predict the
uncertain system involving small data and poor information comes into being.
Among the four basic forms of model GM (1, 1) discussed in Sects. 7.3 and 7.4,
three discrete models can all accurately simulate the homogeneous exponential
sequence. In the real world, a mass of practical data are not the simple homoge-
neous exponential sequence or close to it. This is the fundamental reason that
people prefer to choose Even Model GM (1, 1) (EGM) in the modeling process of
the uncertain system involving small data and poor information, and it can reflect a
satisfactory result in most cases.
In Sects. 7.3 and 7.4, the definitions of four basic forms of model GM (1, 1) are
put forward, and the properties and characteristics of different models are studied
in-depth. The suitable sequences of different models are studied by simulation and
analysis with homogeneous exponential sequences, non-exponential increasing
sequences, and vibration sequences. The main conclusions of the research are as
follows:
(1) The four basic forms of model GM (1, 1), namely Even Model GM (1, 1)
(EGM), Original Difference Model GM (1, 1) (ODGM), Even Difference
Model GM (1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM) are pairwise
equivalent.
(2) Original Difference Model GM (1, 1) (ODGM), Even Difference Model GM
(1, 1) (EDGM) and Discrete Model GM (1, 1) (DGM) can all simulate the
homogeneous exponential sequence accurately.
(3) For the non-exponential increasing sequences and vibration sequences, we
should first choose the difference, differential hybrid form of Even Model GM
(1, 1) (EGM).
(4) For the non-exponential increasing sequences and vibration sequences which
are close to the homogeneous exponential sequences, we should first choose
the discrete form of Original Difference Model GM (1, 1) (ODGM), Even
Difference Model GM (1, 1) (EDGM) or Discrete Model GM (1, 1) (DGM).
The conclusions above can be the reference and basis for choosing an appro-
priate model in the actual modeling process. There is a modeling software corre-
sponding to the models. Interested readers can download it for free from the website
of the Institute for Grey System Studies of Nanjing University of Aeronautics and
Astronautics (http://igss.nuaa.edu.cn) or from the website of the Marie Curie
International Incoming Fellowship project (FP7.People-IIF-GA-2013-629051)
(http://preview.dmu.ac.uk/research/research-faculties-and-institutes/technology/cci/
projects/).
156 7 Series of GM Models

ð0Þ ð0Þ ð0Þ

Example 6.2.1 Let sequences of X1 ; X2 and X3 be as follows,

ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ

X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5ÞÞ
¼ ð1:5; 2:1; 3:0; 4:5; 5:48Þ
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
X2 ¼ ðx2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð4Þ; x2 ð5Þ; x2 ð6ÞÞ
¼ ð1:5; 1:3; 3:0; 3:9; 7:2; 9:5Þ
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
X3 ¼ ðx3 ð1Þ; x3 ð2Þ; x3 ð3Þ; x3 ð4Þ; x3 ð5ÞÞ
¼ ð2; 9; 32; 27; 55Þ

Try to build the Even Model GM (1,1) (EGM), Discrete Model GM (1, 1)
(DGM), Original Difference Model GM (1, 1) (ODGM), and Even Difference
ð0Þ ð0Þ ð0Þ
Model GM (1, 1) (EDGM) using sequences X1 ; X2 and X3 . Compare the
simulation errors.
ð0Þ
Solution: (1) For X1 , we build Even Model GM (1, 1) (EGM), Discrete
Model GM (1, 1) (DGM), Original Difference Model GM (1, 1) (ODGM), and
Even Difference Model GM (1, 1) (EDGM) using 1.5, 2.1, 3.0, 4.5, 5.48. We then
obtained the simulation results as follows.
^1ð0Þ = (1.5000, 2.2459, 3.0428, 4.1225, 5.5853)
Simulation results by EGM: X
^1ð0Þ = (1.5000, 2.2746, 3.0844, 4.1827, 5.6719)
Simulation results by DGM: X
Simulation results by ODGM: X ^ ð0Þ = (1.5000, 2.2600, 3.0726, 4.1772, 5.6789)
1
Simulation results by EDGM: X ^1ð0Þ = (1.5000, 2.2662, 3.0776, 4.1795, 5.6760)
The simulation errors of four different models are presented in Table 7.6.
ð0Þ
(2) For X2 , we build EGM, DGM, ODGM, and EDGM using 1.5, 1.3, 3.0, 3.9,
7.2, 9.5. Then we obtained the simulation results as follows.

Simulation ^2ð0Þ = (1.5000, 1.8632, 2.8290, 4.2955, 6.5220,

results by EGM: X
9.9028)
Simulation ^2ð0Þ = (1.5000, 1.9247, 2.9317, 4.4654, 6.8016,
results by DGM: X
10.3599)
Simulation ^2ð0Þ = (1.5000, 1.8793, 2.8771, 4.4047, 6.7433,
results by ODGM: X
10.3236)
Simulation ^2ð0Þ = (1.5000, 1.8973, 2.8988, 4.4290, 6.7669,
results by EDGM: X
10.3388)

ð0Þ
Table 7.6 Simulation errors of four different models with X1
Models EGM DGM ODGM EDGM
Mean relative errors (%) 4.7363 5.3458 5.1685 5.2442
7.3 Suitable Ranges of Different GM (1, 1) 157

ð0Þ
Table 7.7 Simulation errors of four different models with X2
Models EGM DGM ODGM EDGM
Mean relative errors (%) 11.9881 12.3218 12.2611 12.2860

ð0Þ
Table 7.8 Simulation errors of four different models with X3
Models EGM DGM ODGM EDGM
Mean relative errors (%) 27.2510 25.9994 26.4180 26.1794

The simulation errors of four different models are presented in Table 7.7.
ð0Þ
(3) For X3 ,we build EGM, DGM, ODGM, and EDGM using 2, 9, 32, 27, 55.
Then we obtained the simulation results as follows.
^ ð0Þ = ( 2.0000, 13.9767, 21.6340, 33.4864, 51.8323)
Simulation results by EGM: X3
^3ð0Þ = (2.0000, 15.4516, 23.4647, 35.6332, 54.1122)
Simulation results by DGM: X
ð0Þ
^ = (2.0000, 13.4756, 21.3666, 33.8782,
Simulation results by ODGM: X 3
53.7164)
Simulation results by EDGM: X ^3ð0Þ = (2.0000, 14.2602, 22.2313, 34.6581, 54.0311)
The simulation errors of four different models are presented in Table 7.8.
ð0Þ ð0Þ ð0Þ
The simulation results with X1 ; X2 and X3 confirmed the above conclusion
once again.

7.3.2 Applicable Ranges of EGM

Pn  ð1Þ 2 Pn ð1Þ 2
Proposition 7.3.1 When ðn  1Þ k¼2 z ðkÞ ! k¼2 z ðkÞ , the EGM
(1, 1) becomes invalid.
Proof By using the model parameters obtained by the least squared estimate, we
have
Pn Pn Pn ð1Þ
zð1Þ ðkÞ ð0Þ
k¼2 x ðkÞ  ðn  1Þ
ð0Þ
k¼2 z ðkÞx ðkÞ
^
a¼ k¼2
Pn 2 Pn 2
ðn  1Þ k¼2 ½zð1Þ ðkÞ  ð1Þ
k¼2 z ðkÞ
Pn ð0Þ
Pn  ð1Þ
2 Pn ð1Þ Pn ð1Þ ð0Þ
^ k¼2 x ðkÞ k¼2 z ðkÞ  k¼2 z ðkÞ k¼2 z ðkÞx ðkÞ
b¼ P 2  Pn 2
ðn  1Þ nk¼2 ½zð1Þ ðkÞ  ð1Þ
k¼2 z ðkÞ

P  2 Pn ð1Þ 2
When ðn  1Þ nk¼2 zð1Þ ðkÞ ! k¼2 z ðkÞ , ^a ! 1; ^b ! 1, so that the
model parameters cannot be determined. Hence, the EGM (1, 1) becomes invalid.
158 7 Series of GM Models

Proposition 7.3.2 When the development coefficient a of the EGM (1, 1) model
satisfies jaj  2, the GM (1, 1) model becomes invalid.
Proof From the following expression of the GM (1, 1) model
 k2  
1  0:5a b  axð0Þ ð1Þ
xð0Þ ðkÞ ¼ ; k ¼ 2; 3;    ; n
1 þ 0:5a 1 þ 0:5a

it can be seen that when a ¼ 2, xð0Þ ðkÞ ! 1; when a ¼ 2, xð0Þ ðkÞ ¼ 0; and when
ð0Þ
 k2
|a| > 2, b1þax0:5að1Þ  0:5a
becomes a constant, while the sign of 11 þ 0:5a changes with
k being even or odd. Thus, the sign of xð0Þ ðkÞ flips with k being even or odd.
The discussion above indicates that ð1; 2 [ ½2; 1Þ is the forbidden area for
the development coefficient (a) of the GM (1, 1) model. When
a 2 ð1; 2 [ ½2; 1Þ, the GM (1, 1) model loses its validity. In general, when
|a| < 2, the GM (1, 1) model is meaningful. However, for different values of a, the
prediction effect of the model is different. For the case of -2 < a < 0, let us
respectively take a = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 1.5, 1.8 to conduct a sim-
ð0Þ
ulation analysis. By taking k = 0, 1, 2, 3, 4, 5, from xi ðk þ 1Þ ¼ eak , we obtain
the following sequences:
ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ ð0Þ
If  a ¼ 0:1; X1 ¼ ðx1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ; x1 ð6ÞÞ
¼ ð1; 1:1051; 1:2214; 1:3499; 1:4918; 1:6487Þ:
ð0Þ
If a = 0.2, X2 = (1, 1.2214, 1.4918, 1.8221, 2.2255, 2.7183).
ð0Þ
If a = 0.3, X3 = (1, 1.3499, 1.8221, 2.4596, 3.3201, 4.4817).
ð0Þ
If a = 0.4, X4 = (1, 1.4918, 2.225, 3.3201, 4.9530, 7.3890).
ð0Þ
If a = 0.5, X5 = (1, 1.6487, 2.7183, 4.4817, 7.3890, 12.1825).
ð0Þ
If a = 0.6, X6 = (1, 1.8821, 3.3201, 6.0496, 11.0232, 20.0855).
ð0Þ
If a = 0.8, X7 = (1, 2.2255, 4.9530, 11.0232, 24.5325, 54.5982).
ð0Þ
If a = 1, X8 = (1, 2.7183, 7.3890, 20.0855, 54.5982, 148.4132).
ð0Þ
If a = 1.5, X9 = (1, 4.4817, 20.0855, 90.0171, 403.4288, 1808.0424).
ð0Þ
If a = 1.8, X10 = (1, 6.0496, 36.5982, 221.4064, 1339.4308, 8103.0839).
ð0Þ ð0Þ ð0Þ
Let us respectively apply X1 , X2 , …, and X9 to establish a GM (1, 1) model
and obtain the following time response sequences:
7.3 Suitable Ranges of Different GM (1, 1) 159

ð1Þ
^x1 ðk þ 1Þ ¼ 10:50754e0:09992182k  9:507541;
ð1Þ
^x2 ðk þ 1Þ ¼ 5:516431e0:1993401k  4:516431;
ð1Þ
^x3 ðk þ 1Þ ¼ 3:85832e0:297769k  2:858321;
ð1Þ
^x4 ðk þ 1Þ ¼ 3:033199e0:394752k  2:033199;
ð1Þ
^x5 ðk þ 1Þ ¼ 2:541474e0:4898382k  1:541474;
ð1Þ
^x6 ðk þ 1Þ ¼ 2:216363e0:5826263k  1:216362;
ð1Þ
^x7 ðk þ 1Þ ¼ 1:815972e0:7598991k  0:8159718;
ð1Þ
^x8 ðk þ 1Þ ¼ 1:581973e0:9242348k  0:5819733;
ð1Þ
^x9 ðk þ 1Þ ¼ 1:287182e1:270298k  0:2871823;
ð1Þ
^x10 ðk þ 1Þ ¼ 0:198197e1:432596k  0:1981966:

ð0Þ ð1Þ ð1Þ

From ^xi ðk þ 1Þ ¼ ^xi ðk þ 1Þ  ^xi ðkÞ, i ¼ 1; 2;    ; 10, we obtain

ð0Þ ð0Þ
^x1 ðk þ 1Þ ¼ 0:99918e0:09992182k ; ^x2 ðk þ 1Þ ¼ 0:99698e0:1993401k ;
ð0Þ ð0Þ
^x3 ðk þ 1Þ ¼ 0:99362e0:297769k ; ^x4 ðk þ 1Þ ¼ 0:989287e0:394752k ;
ð0Þ ð0Þ
^x5 ðk þ 1Þ ¼ 0:984248e0:4898382k ; ^x6 ðk þ 1Þ ¼ 0:97868e0:5826263k ;
ð0Þ ð0Þ
^x7 ðk þ 1Þ ¼ 0:966617e0:7598991k ; ^x8 ðk þ 1Þ ¼ 0:95419e0:9242348k ;
ð0Þ ð0Þ
^x9 ðk þ 1Þ ¼ 0:925808e1:270298k ; ^x10 ðk þ 1Þ ¼ 0:91220e1:432596k :

From the mean generation of zð1Þ ðkÞ ¼ 12 ðxð1Þ ðkÞ þ xð1Þ ðk  1ÞÞ of GM (1, 1)
model xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b, it has the effect of weakening the growth for
increasing sequences. For an exponential sequence, the established GM (1, 1) has a
small development coefficient.
ð0Þ
Let us compare the errors between the original sequence Xi and the simulation
sequence X ^ið0Þ , as seen in Table 7.9.
It can be seen that as the development coefficient increases, the simulation error
grows drastically. When the development coefficient is smaller than or equal to 0.3,
the simulation accuracy can reach above 98 %. When the coefficient is smaller than
or equal 0.5, the simulation accuracy can reach above 95 %. When the coefficient is
greater than 1, the simulation accuracy is lower than 70 %. When the coefficient is
greater than 1.5, the simulation accuracy is lower than 50 %.
Let us now further focus on the first step, second step, fifth step, and 10th step
prediction errors. See Table 7.10.
It can be seen that when the development coefficient is smaller than 0.3, the step 1
prediction accuracy is above 98 %, with both steps 2 and 5 accuracies above 97 %.
When 0:3\  a  0:5, the steps 1 and 2 prediction accuracies are all above 90 %;
160 7 Series of GM Models

Table 7.9 The simulation errors of different development coefficients (a)

P P
Development coefficient (a) 1 6i¼2 ½^xð0Þ ðkÞ  xð0Þ ðkÞ Mean relative error 1 6k¼2 Dk (%)
5 5
0.1 0.004 0.104
0.2 0.010 0.499
0.3 0.038 1.300
0.4 0.116 2.613
0.5 0.307 4.520
0.6 0.741 7.074
0.8 3.603 14.156
1 14.807 23.544
1.5 317.867 51.033
1.8 1632.240 65.454

Table 7.10 Prediction errors

a 0.1 0.2 0.3 0.4 0.5 0.6 0.8 1 1.5 1.8
Step 1 error 0.129 0.701 1.998 4.317 7.988 13.405 31.595 65.117 – –
(%)
Step 2 error 0.137 0.768 2.226 4.865 9.091 15.392 36.979 78.113 – –
(%)
Step 5 error 0.160 0.967 2.912 6.529 12.468 21.566 54.491 – – –
(%)
Step 5 error 0.855 1.301 4.067 9.362 18.330 32.599 88.790 – – –
(%)

and the step 10 prediction accuracy also above 80 %. When the development
coefficient is greater than 0.8, the step 1 prediction accuracy is below 70 %. The
horizontal bars in Table 4.5 represent that the relevant errors are greater than 100 %.
From this analysis, we can draw the following conclusions: When a  0:3, GM
(1, 1) can be applied to make mid- to long-term predictions; when 0:3\  a  0:5,
GM (1, 1) can be applied to make short- and mid-term predictions with caution;
when 0:5\  a  0:8 and GM (1, 1) is used to make short-term predictions, one
needs to be very cautious about the prediction results; when 0:8\  a  1, one
should employ the remnant GM (1, 1) model; and when a [ 1, GM (1, 1) should
not be applied.

7.4 Remnant GM (1,1) Model

When the accuracy of a GM (1, 1) model does not meet the predetermined
requirement, one can establish another GM (1, 1) model using the error sequence to
remedy the original model to improve the accuracy. We will use the remnant GM
(1, 1) of EGM (1) as an example.
7.4 Remnant GM (1, 1) Model 161

Definition 7.4.1 Assume that X ð0Þ is a sequence of raw data, X ð1Þ the accumulation
generated sequence based on X ð0Þ , and the time response formula of the GM (1, 1)
model is

b b
^xð1Þ ðk þ 1Þ ¼ ðxð0Þ ð1Þ  Þ eak þ
a a

then

b
d ^xð1Þ ðk þ 1Þ ¼ ðaÞðxð0Þ ð1Þ  Þ eak ð7:26Þ
a

is referred to as the restored value through derivatives.

Generally, d^xð1Þ ðk þ 1Þ 6¼ ^xð0Þ ðk þ 1Þ, where ^xð0Þ ðk þ 1Þ ¼ ^xð1Þ ðk þ 1Þ  ^xð1Þ ðkÞ
stands for the restored value through inverse accumulation. This very fact implies
that the GM (1, 1) is neither a differential equation nor a difference equation.
However, when jaj is sufficiently small, from 1  ea  a, it follows that
d ^xð1Þ ðk þ 1Þ  ^xð0Þ ðk þ 1Þ, meaning that the results of differentiation and difference
are quite close. Therefore, the GM (1, 1) model in this case can be seen as both a
differential equation and a difference equation.
Because the restored values through derivatives and through inverse accumu-
lation are different, to reduce possible errors caused by reciprocating operators, the
errors of X ð1Þ are often used to improve the simulated values ^xð1Þ ðk þ 1Þ of X ð1Þ .
Definition 7.4.2 Assume that eð0Þ ¼ ðeð0Þ ð1Þ; eð0Þ ð2Þ;    ; eð0Þ ðnÞÞ, where
eð0Þ ðkÞ = xð1Þ ðkÞ  ^xð1Þ ðkÞ, is the error sequence of X ð1Þ . If there is a k0 satisfying
 ð0Þ n  k0 
that 4 and  8k  k0 , the  signs of eð0Þ ðkÞ stay the same, and
ðe ðk0 Þ; eð0Þ ðk0 þ 1Þ; ;    ; eð0Þ ðnÞÞ is referred to as the error sequence of

modelability, which is and still denoted eð0Þ ¼ eð0Þ ðk0 Þ; eð0Þ ðk0 þ 1Þ;    ; eð0Þ ðnÞ
In this case, let the sequence eð1Þ ¼ ðeð1Þ ðk0 Þ; eð1Þ ðk0 þ 1Þ;    ; eð1Þ ðnÞÞ be accu-
mulation generated on eð0Þ with the following GM (1, 1) time response formula:
 
be be
^eð1Þ ðk þ 1Þ ¼ eð0Þ ðk0 Þ  exp½ae ðk  k0 Þ þ ; k  k0
ae ae

Then the simulation sequence of eð0Þ is given by ^eð0Þ ¼ ð^eð0Þ ðk0 Þ; ^eð0Þ ðk0 þ 1Þ;
   ; ^eð0Þ ðnÞÞ, where
 
ð0Þ ð0Þ be
^e ðk þ 1Þ ¼ ðae Þ e ðk0 Þ  exp½ae ðk  k0 Þ k  k0 :
ae
^ ð1Þ , the modified time response formula
Definition 7.4.3 If ^eð0Þ is used to improve X
162 7 Series of GM Models

xð0Þ ð1Þ  ba eak þ ba ; k\k0

^x ðk þ 1Þ ¼ ð0Þ
ð1Þ

x ð1Þ  ba eak þ ba  ae ðeð0Þ ðk0 Þ  baee Þ eae ðkk0 Þ ; k  k0

ð7:27Þ

is referred to as the GM (1, 1) model with error modification or simply remnant GM

(1, 1) for short, where the sign of the error modification value
 
ð0Þ be
ð0Þ
^e ðk þ 1Þ ¼ ae e ðk0 Þ  exp½ae ðk  k0 Þ
ae

needs to stay the same as those in eð0Þ .

If a modeling of the error sequence eð0Þ ¼ ðeð0Þ ðk0 Þ; eð0Þ ðk0 þ 1Þ;    ; eð0Þ ðnÞÞ of
ð0Þ
X and X ^ ð0Þ is used to modify the simulation value X ^ ð0Þ , then different methods of
restoration from X ^ ð0Þ can produce different time response sequences of error
^ ð1Þ to X
modification.
Definition 7.4.4 Let
 
b aðk1Þ
^xð0Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ ¼ ð1  ea Þ xð0Þ ð1Þ  e
a

Then the corresponding time response sequence of error modification

(

ð1  ea Þ xð0Þ ð1Þ  ba eak ;   k\k0

ð0Þ
^x ðk þ 1Þ ¼ ð0Þ

b ak ð0Þ a ðkk Þ

ð1  e Þ x ð1Þ  a e  ae e ðk0 Þ  ae e
a be e 0
; k  k0
ð7:28Þ

is called the error modification model of inverse accumulation restoration.

Definition 7.4.5 Let
 
ð0Þ ð0Þ b ak
^x ðk þ 1Þ ¼ ðaÞ x ð1Þ  e ;
a

then the corresponding time response sequence of error modification

(

ðaÞ xð0Þ ð1Þ  ba eak ;   k\k0

ð0Þ
^x ðk þ 1Þ ¼ ð0Þ

b ak ð0Þ ae ðkk0 Þ

ðaÞ x ð1Þ  a e  ae e ðk0 Þ  ae e
be
; k  k0
ð7:29Þ

is referred to as the error modification model of derivative restoration.

7.4 Remnant GM (1, 1) Model 163

In the previous discussion, all the error simulation terms in remnant GM (1, 1)
have been taken as the derivative restoration. Of course, they can be taken as
inverse accumulation restoration. That is, one can take
 
be ae ðkk0 Þ
^eð0Þ ðk þ 1Þ ¼ ð1  eae Þ eð0Þ ðk0 Þ  e ; k  k0
ae

As long as jae j is sufficiently small, the effects of different error restoration

methods on the modified ^xð0Þ ðk þ 1Þ are almost the same.
Example 7.4.1 Let

X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ð13ÞÞ

¼ ð6; 20; 40; 25; 40; 45; 35; 21; 14; 18; 15:5; 17; 15Þ

be a sequence of raw data, and the creation of a EGM (1, 1) model produce the
following time response sequence:

^xð1Þ ðk þ 1Þ ¼ 567:999e0:06486k þ 573:999

The application of inverse accumulating restoration gives:

^ ð0Þ ¼ f^xð0Þ ðkÞg13

X 2 ¼ ð35:6704; 33:4303; 31:3308; 29:3682; 27:5192; 25:7900;
24:1719; 22:6534; 21:2307; 19:8974; 18:6478; 17:4768Þ

The errors and relative errors of the results can be seen in Table 7.11.

Table 7.11 The errors and relative errors of EGM (1, 1)

No. Real data Simulated Errors Relative errors
xð0Þ ðkÞ values ^xð0Þ ðkÞ eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ Dk ¼ xjeðkÞj
ð0Þ ðkÞ (%)

2 20 35.6704 −15.6704 78.3540

3 40 33.4303 6.5697 16.4242
4 25 31.3308 −6.3308 25.3232
5 40 29.3682 10.6318 26.5795
6 45 27.5192 17.4808 38.8642
7 35 25.6901 9.2099 26.3140
8 21 24.1719 −3.1719 15.1043
9 14 22.6534 −8.6534 61.8100
10 18 21.2307 −3.2307 17.9483
11 15.5 19.8974 −4.3974 28.3703
12 17 18.6478 −1.6478 9.6926
13 15 17.4768 −2.4768 16.5120
164 7 Series of GM Models

From Table 7.11, it can be seen that the simulation error is relatively large. Thus,
it is necessary to apply a remnant model to remedy some of the errors.
Let k0 ¼ 9, we get the error sequence as follows

eð0Þ ¼ ðeð0Þ ð9Þ; eð0Þ ð10Þ; eð0Þ ð11Þ; eð0Þ ð12Þ; eð0Þ ð13ÞÞ
¼ ð8:6534; 3:2307; 4:3974; 1:6478; 2:4768Þ

which is an error sequence of modelability. Taking absolute value gives

eð0Þ ¼ ð8:6534; 3:2307; 4:3974; 1:6478; 2:4768Þ

In establishing a EGM (1, 1) for eð0Þ , we have the time response sequence of eð1Þ

^eð1Þ ðk þ 1Þ ¼ 24e0:16855ðk9Þ þ 32:7

whose restored value of derivatives is

^eð0Þ ðk þ 1Þ ¼ ð0:16855Þð24Þe0:16855ðk9Þ ¼ 4:0452e0:16855ðk9Þ

From

b
^xð0Þ ðk þ 1Þ ¼ ^xð1Þ ðk þ 1Þ  ^xð1Þ ðkÞ ¼ ð1  ea Þðxð0Þ ð1Þ  Þeak
a
¼ 38:0614e0:06486k

We can obtain the remnant model of inverse accumulating restoration


ð0Þ 38:0614e0:06486k ; k\9
^x ðk þ 1Þ ¼
38:0614e0:06486k  4:0452e0:16855ðk9Þ ; k9

where the sign of ^eð0Þ ðk þ 1Þ is the same as the original error sequence.
Based on this model, we can modify the four simulation values with k = 10, 11,
12, 13, with improved accuracy listed in Table 7.12.

Table 7.12 Improved results

No. Real data Simulated values Errors Relative errors
xð0Þ ðkÞ ^xð0Þ ðkÞ eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ Dk ¼ xjeðkÞj
ð0Þ ðkÞ (%)

10 18 17.1858 0.8142 4.52

11 15.5 16.4799 −0.9799 6.32
12 17 15.6604 1.2396 7.29
13 15 15.0372 −0.0372 0.25
7.4 Remnant GM (1, 1) Model 165

From this table, we can compute the sum of squares of errors as follows,

s ¼ eT e ¼ 3:1611

and the average relative error

1 X13
D¼ Dk ¼ 4:595 %
12 k¼10

Here, the simulation accuracy of the remnant EGM (1, 1) has obviously
increased. However, the current error sequence no longer satisfies the modeling
requirement. Therefore, if the improved accuracy is still unsatisfactory, we will
have to consider other models or some appropriate choice of data to the original
sequence.

7.5 Group of GM (1, 1) Models

In practice, one does not have to use all the available data in their modeling. Each
subsequence of the original data can be employed to establish a model. Generally
speaking, different subsequences lead to different models. Even though the same
kind of GM (1, 1) is applied, different subsequences lead to different a; b values.
These changes reflect the fact that varied circumstances and conditions have dif-
ferent effects on the system under consideration.
For example, for the grain production in China, if we use the data values col-
lected since 1949 to establish a model GM (1, 1), the development coefficient (–a)
will be on the small side. However, if only the values collected after 1978 are used,
the corresponding development coefficient (–a) will obviously increase.
Definition 7.5.1 For a given sequence X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ, if we
take xð0Þ ðnÞ as the origin of the time axis, then t < n is seen as the past, t = n the
present, and t > n the future.
Definition 7.5.2 Assume that X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ is a sequence of
raw data, let
 
b
^xð0Þ ðk þ 1Þ ¼ ð1  ea Þ xð0Þ ð1Þ  eak
a

be the restored values of inverse accumulation of the GM (1, 1) time responses

of X ð0Þ . Then:
(1) For t  n, ^xð0Þ ðtÞ is referred to as the simulated value out of the model; and
(2) When t > n, ^xð0Þ ðtÞ is known as the prediction of the model.
166 7 Series of GM Models

The main purpose of modeling is to make predictions. To improve the prediction

accuracy, one first needs to guarantee sufficiently high accuracy in his simulation,
especially for the simulation of the time moment t = n. Therefore, in general, the
data, including xð0Þ ðnÞ, used for modeling should be an equal-time-interval
sequence.
Definition 7.5.3 Assume that X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ is a sequence of
raw data, then:
(1) The GM (1, 1) model established using the entire sequence X ð0Þ is known as
the all-data GM (1, 1);
(2) 8k0 [ 1, the GM (1, 1) model established on the tail sequence
X ð0Þ = ðxð0Þ ðk0 Þ; xð0Þ ðk0 þ 1Þ;    ; xð0Þ ðnÞÞ is known as a partial-data GM (1, 1);
(3) If xð0Þ ðn þ 1Þ stands for a piece of new information, then the GM (1, 1) model
established on the prolonged sequence X ð0Þ = ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞ;
xð0Þ ðn þ 1ÞÞ is known as a new-information GM (1, 1);
(4) The GM (1, 1) model established on X ð0Þ ¼ ððxð0Þ ð2Þ;    ; xð0Þ ðnÞ; xð0Þ ðn þ 1ÞÞ
with the new information added and the oldest piece xð0Þ ð1Þ of information
removed is known as a metabolic GM (1, 1).

Example 7.5.1 Let

X ð0Þ ¼ ð60:7; 73:8; 86:2; 100:4; 123:3Þ

and xð0Þ ð6Þ ¼ 149:5 is a piece of new information. Try to establish a model with
X(0), a model of new information, and a metabolic EGM (1, 1).
Solution: (1) The model with X(0). From

X ð0Þ ¼ ð60:7; 73:8; 86:2; 100:4; 123:3Þ

We have
 
a 0:17241
^a ¼ ðBT BÞ1 BT Y ¼ ¼
b 55:889264

The time response sequence is as follows

b b
^xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þeaðk1Þ þ ¼ 384:865028e0:17241k  324:165028
a a
7.5 Group of GM (1, 1) Models 167

Then we obtained the simulation sequence of X(0) as follows

^ ð0Þ ¼ ð60:7; 72:41804; 86:04456; 102:2351; 121:4721Þ

X

The corresponding error sequence is

e ¼ ð0; 1:38196; 0:155434; 1:8351; 1:827829Þ

where eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ.

Therefore, we got the average relative error

1X 5
D¼ Dk ¼ 1:34 %
4 k¼2

where Dk ¼ xjeðkÞj
ð0Þ ðkÞ.

(2) The model of new information. In inserting a piece of new information

xð0Þ ð6Þ ¼ 149:5, the data sequence became

X ð0Þ ¼ ð60:7; 73:8; 86:2; 100:4; 123:3; 149:5Þ

We have
 
a 0:180888
^a ¼ ðBT BÞ1 BT Y ¼ ¼
b 54:254961

Its time response sequence is as follows:

b b
^xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þeaðk1Þ þ ¼ 360:63748e0:180888k  299:93748
a a

The simulation sequence of the new information sequence X(0), the corre-
sponding error sequence e, and the average relative error △ are as follows:

^ ð0Þ ¼ ð60:7; 71:50736; 85:68587; 102:6757; 123:0342; 147:429Þ

X
e ¼ ð0; 2:29264; 0:514129; 2:2757; 0:265712; 2:07041Þ
1X 6
D¼ Dk ¼ 1:51 %
5 k¼2

(3) The metabolic EGM (1, 1). In adding a piece of new information
xð0Þ ð6Þ ¼ 149:5, and deleting a piece of old information xð0Þ ð1Þ ¼ 60:7, we have
168 7 Series of GM Models

X ð0Þ ¼ ð73:8; 86:2; 100:4; 123:3; 149:5Þ

and
 
a 0:187862
^a ¼ ðBT BÞ1 BT Y ¼ ¼
b 62:830896

The corresponding time response sequence is:

b b
^xð1Þ ðkÞ ¼ ðxð0Þ ð1Þ  Þeaðk1Þ þ ¼ 408:251645e0:187862k  334:451645
a a

And the simulation sequence of the metabolic sequence X(0), the corresponding
error sequence e, and the average relative error △ are as follows:

^ ð0Þ ¼ ð73:8; 84:37234; 101:8093; 122:85; 148:2391Þ

X
e ¼ ð0; 1:827657; 1:4093; 0:45; 1:2609Þ
1X 6
D¼ Dk ¼ 1:18 %
4 k¼3

Compared with these different results, it implies that the simulation accuracy can
be improved by appropriately choosing the data to be used in the process of
modeling. From the three different error sequences, we can see that for the simu-
lation accuracy of value x(0)(5), both the new information model and the metabolic
model are better than the model in (1). This implies that the new information EGM
(1, 1) and the metabolic EGM (1, 1) have better prediction abilities than the old
model. As a matter of fact, in the development process of a grey system, there
always exists some stochastic interferences or driving forces entering the system as
time goes on, so that the consequent development of the system is accordingly
affected.
Therefore, when using the EGM (1, 1) model to do predictions, high accuracy
can be achieved only for the first or the second data values after the last origin value
x(0)(n). In general, the farther away into the future, and the farther away from the
last origin value, the weaker the prediction ability of EGM (1, 1) becomes. In
practical applications, one needs to constantly consider those interferences and
driving factors entering the system as time goes on and promptly add new pieces of
information to the original sequence X(0) and establish consequent new information
EGM (1, 1) models.
From the simulation accuracy of value x(0)(6), it can be seen that the metabolic
model is better than the new information model. From the angle of prediction, it can
be seen that the metabolic model is the best prediction model. As the system
develops further, the significance of the older data reduces so that, when new data
are added, the older data are deleted promptly, and the constantly renewing mod-
eling sequence can better reflect the current characteristics of the system.
7.5 Group of GM (1, 1) Models 169

Specifically, as the accumulation of quantitative changes increases, a jump or

sudden change in the system will occur. At this very moment, compared with the
older system, the current system is completely different. Hence, the practice of
deleting old data is very reasonable. Indeed, the ongoing replacement of old data
can avoid computation difficulties in modeling due to the fact that increased
information can increase computer storage space requirements tremendously.

7.6 The Models of GM (r, h)

7.6.1 The Model of GM (0, N)

ð0Þ ð0Þ ð0Þ ð0Þ

Definition 7.6.1 Assume that X1 ¼ ðx1 ð1Þ; x1 ð2Þ;    ; x1 ðnÞÞ is a data
sequence of a system’s characteristic variable,

ð0Þ ð0Þ ð0Þ ð0Þ

X2 ¼ ðx2 ð1Þ; x2 ð2Þ;    ; x2 ðnÞÞ
ð0Þ ð0Þ ð0Þ ð0Þ
X3 ¼ ðx3 ð1Þ; x3 ð2Þ;    ; x3 ðnÞÞ
... ... ... ... ... ...
ð0Þ ð0Þ ð0Þ ð0Þ
XN ¼ ðxN ð1Þ; xN ð2Þ;    ; xN ðnÞÞ

ð1Þ
the data sequences of relevant factors, and Xi the accumulation generated
ð0Þ
sequence of Xi , i ¼ 2; 3;    ; N. Then

ð1Þ ð1Þ ð1Þ ð1Þ

x1 ðkÞ ¼ a þ b2 x2 ðkÞ þ b3 x3 ðkÞ þ    þ bN xN ðkÞ ð7:30Þ

is called the model of GM (0, N). Because this model does not contain any
derivative, it is a static model. Although its form looks like a multivariate linear
regression model, it is essentially different from any of the statistical models. In
particular, the general multivariate linear regression model is established on the
basis of the original data sequences, while the model of GM (0, N) is constructed on
the accumulation generation of the original data.
ð0Þ ð1Þ
Theorem 7.6.1 Assume Xi and Xi ði ¼ 1; 2;    ; N) as given in Definition 7.6.1,
let
2 3 2 3
ð1Þ ð1Þ ð1Þ
1 x2 ð2Þ x3 ð2Þ 
ð1Þ
xN ð2Þ x1 ð2Þ
6 7 6 ð1Þ 7
6 ð1Þ ð1Þ ð1Þ 7 6 x1 ð3Þ 7
B ¼ 6 1 x2 ð3Þ x3 ð3Þ  xN ð3Þ 7; Y ¼ 6
6 .. 7
7
4     5 4 . 5
ð1Þ ð1Þ ð1Þ
1 x2 ðnÞ x3 ðnÞ  xN ðnÞ ð1Þ
x1 ðnÞ
170 7 Series of GM Models

then the least squares estimate of the parametric sequence ^a ¼ ½a; b1 ; b2 ;    ; bN T

is given by

^a ¼ ðBT BÞ1 BT Y:

Example 7.6.1 Let

ð0Þ ð0Þ
X1 ¼ ð2:874; 3:278; 3:307; 3:39; 3:679Þ ¼ fx1 ðkÞg51

be a data sequence of a system’s characteristic variable, and

ð0Þ ð0Þ
X2 ¼ ð7:04; 7:645; 8:075; 8:53; 8:774Þ ¼ fx2 ðkÞg51

the data sequences of a relevant factor. Try to establish the model of GM (0, 2).
Solution: Assume the model of GM (0, 2) as follows:

ð1Þ ð1Þ
X1 ¼ bX2 þ a

From
2 ð1Þ 3 2 3 2 ð1Þ 3 2 3
x2 ð2Þ 1 14:685 1 x1 ð2Þ 6:152
6 ð1Þ 7 6 6 7
7 6 22:76 1 7 x ð3Þ 7 6 9:459 7
ð1Þ
6 x ð3Þ 1 7; Y ¼ 6
B ¼ 6 2ð1Þ 7¼4 5 6 1ð1Þ 7 ¼ 6 7
4 x2 ð4Þ 1 5 31:29 1 4 x1 ð4Þ 5 4 12:849 5
ð1Þ 40:064 1 ð1Þ 16:528
x2 ð5Þ 1 x1 ð5Þ

We have
 
^b ¼ b ¼ ðBT BÞ1 BT Y ¼ 0:412435
a 0:482515

It follows that

ð1Þ ð1Þ
^x1 ðkÞ ¼ 0:412435x2 ðkÞ  0:482515

Therefore, the simulation results are as shown in Table 7.13.

The average relative error is

X5
1X 5
jeðkÞj
¼1
D Dk ¼ ¼ 2:475 %
4 k¼2 4 k¼2 xð0Þ ðkÞ
7.6 The Models of GM (r, h) 171

Table 7.13 Simulation results with errors

Ordinality Real data Simulated Errors Relative errors
xð0Þ ðkÞ values ^xð0Þ ðkÞ eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ Dk ¼ xjeðkÞj
ð0Þ ðkÞ (%)

2 3.278 3.153 0.125 3.8

3 3.307 3.331 −0.024 0.7
4 3.390 3.518 −0.128 3.8
5 3.679 3.619 0.06 1.6

7.6.2 The Model of GM (1, N)

ð0Þ ð1Þ
Definition 7.6.2 Assume that Xi and Xi (i ¼ 1; 2;    ; N) as given in Definition
ð1Þ ð0Þ ð1Þ
7.6.1. Let Xi be the accumulated sequences of Xi , i ¼ 1; 2; . . .; N, and Z1 the
ð1Þ
adjacent neighbor average sequence of X1 . Then,

ð0Þ ð1Þ
X
N
ð1Þ
x1 ðkÞ þ az1 ðkÞ ¼ bi xi ðkÞ ð7:31Þ
i¼2

is called the model of GM (1, N).

ð1Þ
The constant (−a) is known as the system’s development coefficient, bi xi ðkÞ
the driving term, bi the driving coefficient, and ^a ¼ ½a; b1 ; b2 ;    ; bN T the sequence
of parameters.
ð0Þ ð1Þ ð1Þ
Theorem 7.6.2 For the previously defined terms Xi , Xi , and Z1 ,
i ¼ 1; 2; . . .; N, let
2 3 2 3
ð1Þ ð1Þ ð0Þ
z1 ð2Þ x2 ð2Þ 
ð1Þ
xN ð2Þ x1 ð2Þ
6 7 6 ð0Þ 7
6 ð1Þ
z1 ð3Þ x2 ð3Þ
ð1Þ ð1Þ 7 6 x1 ð3Þ 7
B¼6  xN ð3Þ 7; Y ¼ 6 7
4     5 6 .. 7
4 . 5
ð1Þ ð1Þ ð1Þ
z1 ðnÞ x2 ðnÞ    xN ðnÞ ð0Þ
x1 ðnÞ

Then the least squares estimate of the sequence ^a ¼ ½a; b1 ; b2 ;    ; bN T of

parameters satisfies

^a ¼ ðBT BÞ1 BT Y:

Example 7.6.2 Let

ð0Þ ð0Þ
X1 ¼ ð2:874; 3:278; 3:307; 3:39; 3:679Þ ¼ fx1 ðkÞg51
172 7 Series of GM Models

is a data sequence of a system’s characteristic variable, and

ð0Þ ð0Þ
X2 ¼ ð7:04; 7:645; 8:075; 8:53; 8:774Þ ¼ fx2 ðkÞg51

the data sequences of a relevant factor. Try to establish the model of GM (1, 2).
Solution: Assume that the model of GM (1, 2) is as follows:

ð0Þ ð1Þ ð1Þ

x1 ðkÞ þ az1 ðkÞ ¼ bx2 ðkÞ

From

ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ

X1 ¼½x1 ð1Þ; x1 ð2Þ; x1 ð3Þ; x1 ð4Þ; x1 ð5Þ
¼ð2:874; 6:152; 9:459; 12:849; 16:528Þ
ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ
X2 ¼½x2 ð1Þ; x2 ð2Þ; x2 ð3Þ; x2 ð4Þ; x2 ð5Þ
¼ð7:04; 14:685; 22:76; 31:29; 40:064Þ

We have

ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ

Z1 ¼½z1 ð2Þ; z1 ð3Þ; z1 ð4Þ; z1 ð5Þ
¼ð4:513; 7:8055; 11:154; 14:6885Þ

It follows that
2 ð1Þ ð1Þ 3 2 3 2 ð0Þ 3
z1 ð2Þ x2 ð2Þ 4:513 14:685 x1 ð2Þ
6 ð1Þ ð1Þ 7 6 7 6 ð0Þ 7
6 z1 ð3Þ x2 ð3Þ 7 6 7:8055 22:76 7; Y ¼ 6 x ð3Þ 7
B ¼ 6 ð1Þ 7¼4 5 6 1ð0Þ 7
4 z1 ð4Þ ð1Þ
x2 ð4Þ 5 11:154 31:29 4 x1 ð4Þ 5
ð1Þ ð1Þ 14:6885 40:064 ð0Þ
2 z1 ð5Þ3 x2 ð5Þ x1 ð5Þ
3:278
6 3:307 7
¼64 3:390 5
7

3:679

Therefore, we have
 
a 2:2273
^a ¼ ¼ ðBT BÞ1 BT Y ¼
b 0:9068

and

ð0Þ ð1Þ ð1Þ

x1 ðkÞ þ 2:2273z1 ðkÞ ¼ 0:9068x2
7.6 The Models of GM (r, h) 173

Table 7.14 Simulation results with errors

Ordinality Real data Simulated Errors Relative errors
xð0Þ ðkÞ values ^xð0Þ ðkÞ eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ Dk ¼ xjeðkÞj
ð0Þ ðkÞ (%)

2 3.278 3.265 0.013 0.4

3 3.307 3.254 0.053 1.6
4 3.390 3.530 −0.140 4.1
5 3.679 3.614 0.065 1.8

That is,

ð0Þ ð1Þ ð1Þ

^x1 ðkÞ ¼ 2:2273z1 ðkÞ þ 0:9068x2

The simulation results are as shown in Table 7.14.

The average relative error is

X5
1X 5
jeðkÞj
¼1
D Dk ¼ ¼ 1:975 %
4 k¼2 4 k¼2 xð0Þ ðkÞ

7.6.3 The Grey Verhulst Model

The GM (1, 1) model is suitable for sequences that show an obvious exponential
pattern and can be used to describe monotonic changes. As for non-monotonic
wavelike development sequences, or saturated sigmoid sequences, one can consider
establishing a grey Verhulst model.
Definition 7.6.3 Assume that X ð0Þ is a sequence of raw data, X ð1Þ the accumulation
sequence of X ð0Þ , and Z ð1Þ the adjacent neighbor average sequence of X ð1Þ . Then,

xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b½zð1Þ ðkÞa ð7:32Þ

is known as the power model of GM (1, 1). Also,

dxð1Þ =dt þ axð1Þ ¼ bðxð1Þ Þa ð7:33Þ

is known as the whitenization equation of the power model of GM (1, 1) (Deng

1985).
Theorem 7.6.3 The solution of the whitenization equation of the power model of
GM (1, 1) is
174 7 Series of GM Models

 Z 1a
1

ð1Þ ð1aÞat ð1aÞat

x ðtÞ ¼ e ½ð1  aÞ be dt þ c ð7:34Þ

Theorem 7.6.4 Let X ð0Þ , X ð1Þ , and Z ð1Þ be defined as above. Let
2 3 2 3
zð1Þ ð2Þ ½zð1Þ ð2Þa xð0Þ ð2Þ
6 zð1Þ ð3Þ ½zð1Þ ð3Þa 7 6 xð0Þ ð3Þ 7
6 7 6 7
B¼6 .. .. 7; Y ¼ 6 . 7
4 . . 5 4 .. 5
zð1Þ ðnÞ ½zð1Þ ðnÞa xð0Þ ðnÞ

Then the least squares estimate of the parametric sequence ^a ¼ ½a; bT of the
power model of GM (1, 1) is

^a ¼ ðBT BÞ1 BT Y:

Definition 7.6.4 When the power a ¼ 2 in the power model of GM (1, 1), the
resultant model

xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ bðzð1Þ ðkÞÞ2 ð7:35Þ

is known as the grey Verhulst model and

dxð1Þ =dt þ axð1Þ ¼ bðxð1Þ Þ2 ð7:36Þ

is known as the whitenization equation of the grey Verhulst model (Deng 1985).
Theorem 7.6.5 (1) The solution of the Verhulst whitenization equation is

1 axð1Þ ð0Þ
xð1Þ ðtÞ ¼ ¼
eat ½xð1Þ1ð0Þ  ba ð1  eat Þ eat ½a  bxð1Þ ð0Þð1  eat Þ

That is

axð1Þ ð0Þ
xð1Þ ðtÞ ¼ ð7:37Þ
bxð1Þ ð0Þ þ ½a  bxð1Þ ð0Þeat

(2) The time response sequence of the grey Verhulst model is

axð1Þ ð0Þ
^xð1Þ ðk þ 1Þ ¼ ð7:38Þ
bxð1Þ ð0Þ þ ½a  bxð1Þ ð0Þeak

The Verhulst model is mainly used to describe and study processes with satu-
rated states (or sigmoid processes). For instance, this model is often used in the
prediction of human populations, biological growth, reproduction, and economic
7.6 The Models of GM (r, h) 175

Table 7.15 Expenditures on the research of a certain kind of torpedo (in million Yuan)
Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Expenditure 496 779 1187 1025 488 255 157 110 87 79

Table 7.16 Accumulated expenditures (in ten thousand Yuan)

Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Expenditure 496 1275 2462 3487 3975 4230 4387 4497 4584 4663

life span of consumable products. From the solution of the Verhulst equation, it can
be seen that when t ! 1, if a [ 0, then xð1Þ ðtÞ ! 0; if a\0, then xð1Þ ðtÞ ! ab. That
is, there is a sufficiently large t such that for any k [ t, both xð1Þ ðk þ 1Þ and xð1Þ ðkÞ
will be sufficiently close to each other. In this case, xð0Þ ðk þ 1Þ ¼ xð1Þ ðk þ 1Þ
xð1Þ ðkÞ  0, which means that the system approaches distinction.
In practice, one often faces sigmoid processes in the original data sequences.
When such an instance appears, we can simply take the original sequence as X ð1Þ
with its accumulation generation as X ð0Þ to establish a grey Verhulst model to
directly simulate X ð1Þ .
Example 7.6.3 Assume that the expenditures on the research of a certain kind of
torpedo are given in Table 7.15. Try to employ the grey Verhulst model to simulate
the data and make predictions (Liang 2005).
The accumulated expenditures are given in Table 7.16.
From Theorem 7.6.5, we compute the parameters as follows:

0:98079
^a ¼ ½a; bT ¼
0:00021576

so that the whitenization equation is

dxð1Þ =dt  0:98079xð1Þ ¼ 0:00021576ðxð1Þ Þ2 :

By taking xð1Þ ð0Þ ¼ xð0Þ ð1Þ ¼ 496, we obtain the time response sequence

axð1Þ ð0Þ 486:47

^xð1Þ ðk þ 1Þ ¼ ¼ :
bxð1Þ ð0Þ þ ½a  bxð1Þ ð0Þeak 0:10702  0:87378e0:98079k

On the basis of this formula, we produce the simulated values ^xð1Þ ðkÞ as shown
in Table 7.17.
From Table 7.17, we can obtain the average relative error
176 7 Series of GM Models

Table 7.17 The simulation results with errors

Ordinality Actual data Simulated data Error Relative error
xð0Þ ðkÞ ^xð0Þ ðkÞ eðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ Dk ¼ xjeðkÞj
ð0Þ ðkÞ

2 1275 1119.1 155.9 0.12226

3 2462 2116 346 0.14053
4 3487 3177.5 309.5 0.08876
5 3975 3913.7 61.3 0.01541
6 4230 4286.2 −56.2 0.01328
7 4387 4444.8 −57.8 0.01318
8 4497 4507.4 −10.4 0.0023
9 4584 4531.3 52.7 0.0115
10 4663 4540.3 122.7 0.02631

1X 10
D¼ Dk ¼ 4:3354 %
9 k¼2

and predict the research expenditure for the year of 2005 on the special kind of
torpedo as

ð0Þ ð1Þ ð1Þ

^x1 ð11Þ ¼ ^x1 ð11Þ  ^x1 ð10Þ ¼ 9:0342:

This value indicates that the research work on the torpedo is nearing its
conclusion.

7.6.4 The Models of GM (r, h)

In this subsection, we focus on the investigation of the structure of the models of

GM (r, h), and its relationships with models EGM (1, 1), GM (1, N), GM (0, N), and
the grey Verhulst model
ð0Þ ð0Þ ð0Þ ð0Þ
Definition 7.6.5 Assume that Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ, i ¼ 1; 2;    ; h,
ð0Þ
where X1 stands for a data sequence of a system’s characteristic, and
ð0Þ
Xi ,i ¼ 2; 3;    ; h data sequences of relevant factors. Let

ð1Þ ð1Þ ð1Þ ð0Þ

að1Þ^x1 ðkÞ ¼ ^x1 ðkÞ  ^x1 ðk  1Þ ¼ ^x1 ðkÞ
ð1Þ ð1Þ ð1Þ ð0Þ ð0Þ
að2Þ^x1 ðkÞ ¼ að1Þ^x1 ðkÞ  að1Þ^x1 ðk  1Þ ¼ ^x1 ðkÞ  ^x1 ðk  1Þ
... ... ... ... ... ...
ð1Þ ð1Þ ð1Þ ð0Þ ð0Þ
aðrÞ^x1 ðkÞ ¼ aðr1Þ^x1 ðkÞ  aðr1Þ^x1 ðk  1Þ ¼ aðr2Þ^x1 ðkÞ  aðr2Þ^x1 ðk  1Þ
7.6 The Models of GM (r, h) 177

and zð1Þ ðkÞ ¼ 12 ðxð1Þ ðkÞ þ xð1Þ ðk  1ÞÞ, then

ð1Þ
X
r1
ð1Þ ð1Þ
X
h1
ð1Þ
aðrÞ^x1 ðkÞ þ ai aðriÞ x1 ðkÞ þ ar z1 ðkÞ ¼ bj xj þ 1 ðkÞ þ bh ð7:39Þ
i¼1 j¼1

is referred to as the model of GM (r, h). The GM (r, h) model is a rth order grey
model in h variables.
Definition 7.6.5 In the model of GM (r, h), ^a ¼ ½a1 ; a2 ;    ; ar T is referred
Ph1 ð1Þ
to as the development coefficient vector, j¼1 bj xj þ 1 ðkÞ the driving term, and
^ T
b ¼ ½b1 ; b2 ;    ; bh  the vector of driving coefficients.
ð0Þ ð0Þ
Theorem 7.6.6 Let X1 be a data sequence of a system’s characteristic, Xi ,
ð1Þ
i ¼ 2; 3;    ; h, the data sequences of relevant factors, Xi the accumulation
ð0Þ ð1Þ ð1Þ
sequence of Xi , Z1 the adjacent neighbor average sequence from X1 , and
ð1Þ ð1Þ
aðriÞ X1 the ( r  i)th order inverse accumulation sequence of X1 . Define
2 ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ
3
aðr1Þ x1 ð2Þ aðr2Þ x1 ð2Þ  að1Þ x1 ð2Þ z1 ð2Þ x2 ð2Þ 
ð1Þ
xh ð2Þ 1
6 7
6 aðr1Þ xð1Þ ð3Þ aðr2Þ xð1Þ ð3Þ ð1Þ ð1Þ ð1Þ ð1Þ 7
6  að1Þ x1 ð3Þ z1 ð3Þ x2 ð3Þ    xh ð3Þ 1 7
B ¼6 1 1
7;
6              7
4 5
ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ
aðr1Þ x1 ðnÞ aðr2Þ x1 ðnÞ  að1Þ x1 ðnÞ z1 ðnÞ x2 ðnÞ 
ð1Þ
xh ðnÞ 1
2 ð1Þ
3
aðrÞ x1 ð2Þ
6 7
6 aðrÞ xð1Þ ð3Þ 7
6 1 7
Y¼6 7
6 .
. 7
4 . 5
ð1Þ
aðrÞ x1 ðnÞ

then the parametric sequence ^c ¼ ½^a; ^bT ¼ ½a1 ; a2 ;    ; ar ; b1 ; b2 ;

   ; bh T of the least squares estimate satisfies

^a ¼ ðBT BÞ1 BT Y:

The model of GM (r, h) is the general form of grey systems models. In

particular,
(1) When r = 1 and h = 1, the previous (7.39) reduces to:

ð1Þ ð1Þ ð1Þ ð1Þ

dx1 =dt þ a1 x1 ¼ b1 and að1Þ x1 ðkÞ þ a1 z1 ðkÞ ¼ b1

which is the model of EGM (1, 1).

(2) When r = 1 and h = N, the previous (7.39) takes the form of
178 7 Series of GM Models

ð0Þ ð1Þ
X
N
ð1Þ
x1 ðkÞ þ a1 z1 ðkÞ ¼ bi xi ðkÞ
i¼2

which is the GM (1, N) model.

(3) When r = 0 and h = N, the previous model (7.39) is
ð1Þ ð1Þ ð1Þ ð1Þ
x1 ðkÞ ¼ b1 x2 ðkÞ þ b2 x3 ðkÞ þ    þ bN1 xN ðkÞ þ bN

which is the GM (0, N) model.

(4) When r = 1 and h = 1, and b1 in the model of EGM (1, 1) is changed to
bðzð1Þ ðkÞÞ2 , then we have the following grey Verhulst model

xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ bðzð1Þ ðkÞÞ2 :

Based on this discussion, it can be seen that models EGM (1, 1), GM (1, N), GM
(0, N), etc., are all special cases of model GM (r, h). So, it is very important to
further the study of model GM (r, h).

7.7 Practical Applications

Example 7.7.1 Let us look at the revenue predictions of private enterprises at

Changge County, Henan Province, The People’s Republic of China, which we
mentioned in Example 4.3.1. For the years from 1983 to 1986, the overall business
revenue of private enterprises in Changge county was recorded as

X ¼ ð10155; 12588; 23480; 35388Þ

We obtained the following second-order buffered sequence

XD2 ¼ ð27260; 29547; 32411; 35388Þ

in Example 4.3.1 by a second-order average weakening buffer operator (AWBO) as

follows:

1
xðkÞd ¼ ½xðkÞ þ xðk þ 1Þ þ    þ xðnÞ; k ¼ 1; 2;   ; n
n  kþ1

We denote the XD2 as X(0), that is, let

Xð0Þ ¼ ð27260; 29547; 32411; 35388Þ:

Then the 1-AGO sequence X(1) of X(0) is as follows

7.7 Practical Applications 179

X ð1Þ ¼ ðxð1Þ ð1Þ; xð1Þ ð2Þ; xð1Þ ð3Þ; xð1Þ ð4ÞÞ ¼ ð27260; 56807; 89218; 124606Þ:

Assume that

xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b

Based on the least squares method, we obtain the estimated values for a and b as
follows:

^a ¼ 0:089995; ^b ¼ 25790:28

Thus, the resultant whitenization equation of EGM (1, 1) is given by

dxð1Þ
 0:089995xð1Þ ¼ 25790:28
dt

and its time response sequence is


^xð1Þ ðk þ 1Þ ¼ 313834e0:089995k  286574
^xð0Þ ðk þ 1Þ ¼ ^xð1Þ ðk þ 1Þ  ^xð1Þ ðkÞ

From these results, we obtain the simulated sequence

^ ¼ ð^xð1Þ; ^xð2Þ; ^xð3Þ; ^xð4ÞÞ ¼ ð27260; 29553; 32337; 35381Þ

X

with the sequence of errors

eð0Þ ¼ ðeð0Þ ð1Þ; eð0Þ ð2Þ; eð0Þ ð3Þ; eð0Þ ð4ÞÞ ¼ ð0; 6; 74; 7Þ

The sequence of relative errors

 ð0Þ   ð0Þ   ð0Þ   ð0Þ 
e ð1Þ  e ð2Þ   e ð3Þ  e ð4Þ
D ¼½ ð0Þ ;  ð0Þ ;  ð0Þ ;  ð0Þ 
x ð1Þ x ð2Þ x ð3Þ x ð4Þ
¼ð0; 0:0002; 0:00228; 0:0002Þ

And the average relative error

Xn4
 ¼1
D Dk ¼ 0:00067 ¼ 0:067 %\0:01
4 k¼1
D4 ¼ 0:0002 ¼ 0:02 %\0:01

Therefore, the accuracy of our simulation is in level one.

^
Now, we can compute the absolute degree e of grey incidences of X and X.
180 7 Series of GM Models

 
X 3 
 1 
jsj ¼  ½xðkÞ  xð1Þ þ ½xð4Þ  xð1Þ ¼ 11502
 k¼2 2 
 
X 3 
 1 
j^sj ¼  ½^xðkÞ  ^xð1Þ þ ½^xð4Þ  ^xð1Þ ¼ 11430:5
 k¼2 2 
 
X 3 
 1 
j^s  sj ¼ ½xðkÞ  xð1Þ  ð^xðkÞ  ^xð1ÞÞ þ ½xð4Þ  xð1Þ  ð^xð4Þ  ^xð1ÞÞ ¼ 71:5
 k¼2 2 

Thus,

1 þ jsj þ j^sj 1 þ 11502 þ 11430:5

e¼ ¼ ¼ 0:997 [ 0:90
1 þ jsj þ j^sj þ j^s  sj 1 þ 11502 þ 11430:5 þ 71:5

That is, the degree of incidence is in level one.

Compute the ratio of mean square deviations C:

1X 4
1X 4
x ¼ xðkÞ ¼ 31151:5; S21 ¼ ðxðkÞ  xÞ2 ¼ 37252465; S1 ¼ 6103:48
4 k¼1 4 k¼1
1X 4
1X 4
e ¼ eðkÞ ¼ 18:75; S22 ¼ ðeðkÞ  eÞ2 ¼ 4154:75; S2 ¼ 64:46
4 k¼1 4 k¼1

It follows that

S2 64:46
C¼ ¼ ¼ 0:01\0:35
S1 6103:48

which is in level one.

Compute the small error probability. From

0:6745S1 ¼ 4116:80

jeð1Þ  ej ¼ 18:75; jeð2Þ  ej ¼ 24:75; jeð3Þ  ej ¼ 55:25; jeð4Þ  ej ¼ 11:75

Therefore

p ¼ PðjeðkÞ  ej\0:6745S1 Þ ¼ 1 [ 0:95

With our accuracy checks in place, we can apply the grey model

^xð1Þ ðk þ 1Þ ¼ 313834e0:089995k  286574
^xð0Þ ðk þ 1Þ ¼ ^xð1Þ ðk þ 1Þ  ^xð1Þ ðkÞ
7.7 Practical Applications 181

to make predictions. Here, we list five predicted values as follows:

^ ð0Þ ¼½^xð0Þ ð5Þ; ^xð0Þ ð6Þ; ^xð0Þ ð7Þ; ^xð0Þ ð8Þ; ^xð0Þ ð9Þ
X
¼ð38714; 42359; 46348; 50712; 55488Þ

These predictions indicated an average 9.4 % annual growth. When we look

back today, this predicted rate of growth agreed very well with the recorded values
over the time span of our predictions.
Chapter 8
Combined Grey Models

Along with the disciplinary development of systems science and systems engineering,
methods and modeling techniques established for systems evaluation, prediction,
decision-making, and optimization are enriched constantly. Generally, each method
and every model have their strengths and weaknesses, so in practical applications
several different methods and modeling techniques are combined to form hybrid
methods or techniques in order to successfully deal with the problems at hand. Such
combinations and mixtures are used to capitalize upon the strengths and advantages of
different methods so that they complement each other and at the same time improve
the weaknesses of individual methods and modeling techniques. This explains why
combined or mixed systems are superior to individual component methods.
Additionally, the availability of many different methods and modeling techniques also
provides us with different ways to deal with information and systems. Therefore, how
to combine and mix different methods and techniques has become a research direction
with wide-ranging applicability in areas of data mining and knowledge discovery.
Grey systems theory and methods strongly complement many of the traditional
technologies and soft computing techniques. In this chapter, we explore various
combinations, mixtures and applications of grey systems models and models
developed in econometrics, production functions, artificial neural networks, linear
regression, Markov models, and rough sets.

8.1 Grey Econometrics Models

8.1.1 Determination of Variables Using the Principles

of Grey Incidence

In analyzing systems due to the complications of mutually crossing influences of

the endogenous variables, at the very start of modeling, the first problem that needs

© Springer Science+Business Media Singapore 2017 183

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_8
184 8 Combined Grey Models

to be addressed is how to select the variables that will be part of the eventual model.
To revolve this problem, the researcher needs not only rely on his qualitative
analysis of the system, but also have sufficiently adequate tools for conducting
quantitative analysis. Grey incidences provide an effective method for this class of
problems.
Let y be an endogenous variable of the system of our concern (for systems with
many endogenous variables, these variables can be studied individually), and
x1 ; x2 ;


; xn be pre-images of influencing factors that are correlated either posi-
tively or negatively to y: First study the degree of incidence ei between y and xi ,
i ¼ 1; 2;


; n: For a chosen lower threshold value e0 , when ei \e0 , remove the
variable xi out of consideration. By doing so, some of the system’s endogenous
variables with weak degrees of incidence with y can be removed from further
consideration. Assume that the remaining illustrative variables of y are
xi1 ; xi2 ;


; xim . Next, consider the degrees eij ik ðij ; ik ¼ i1 ; i2 ;


;im Þ of incidence
between these remaining variables. For a chosen threshold value e00 , when eij ik  e00 ,
the variables xij and xik are seen as the same kind so that the remaining variables are
divided into several subsets. Now, choose one representative from each of these
subsets to enter into the eventual model. By going through this possess, the
resultant econometrics model can be greatly simplified without losing the needed
power of explanation. At the same time, to a certain degree the difficult problem of
collinearity of the variables can be avoided.

8.1.2 Grey Econometrics Models

In econometrics, there are many different kinds of models, such as linear regression
models in one or multiple variables, nonlinear models, systems of equations, among
others. When estimating the parameters of these models, one often faces phe-
nomena that are difficult to explain. For instance, the coefficients of the major
illustrative variables are nearly zero; the signs of some estimated values of the
parameters do not agree with reality or contradict theoretical economic analysis;
small vibrations in a few individual observations cause drastic changes in many
other estimated parametric values. Among the main reasons underlying these dif-
ficulties are:
(1) During the time period the observations are done, the internal structure of the
system goes through major changes;
(2) There is a problem of collinearity between the illustrative variables; and
(3) There are randomness and noise in the observed data.
For the first two scenarios, there is a need to repeat the investigation of the model
structure or a need to recheck the illustrative variables. For the third scenario, one
can consider establishing models using the GM (1, 1) simulated values of the
original observations to eliminate the effect of the randomness or noise existing in
8.1 Grey Econometrics Models 185

the available data. The combined grey econometrics model, obtained this way, can
more accurately reflect the relationship between the system’s variables. At the same
time, the prediction results made on the endogenous variables of the grey econo-
metrics model system, which is based on the GM (1, 1) predicted values of the
illustrative variables, possess more solid scientific foundation. Besides, by com-
paring the results of grey predictions of the endogenous variables with those
obtained out of econometrics models, one can further improve the reliability of the
predictions.
The steps for establishing and applying grey econometrics models are as
follows:
Step 1: Design the theoretical model. Study the economic activity of interest
closely. According to the purpose of the investigation, select the variables that will
potentially enter the model. Discover the relationships between these variables
based on theories of economic behavior and experience and/or analyze the sampled
data. Develop the mathematical expressions, which are the theoretical model, that
describe the relationships between these variables. This stage is the most important
and difficult phase of the entire modeling process, and the following work need to
be done:
(1) Study relevant theories of economics
Theoretical models summarize the fundamental characteristics and laws of
development of the objective matters. They are abstract pictures of reality.
Therefore, in the stage of model design, one first needs to conduct a qualitative
analysis using economic theories. With different theories, various models can be
established. For instance, according to the theory of equilibrium of labor markets,
the rate y of wage increase is related to the unemployment rate x1 and inflation rate
x2 , that is, y ¼ f ðx1 ; x2 Þ: The greater the unemployment rate increases, the smaller
the rate of wage increase due to the fact that the supply of labor is clearly greater
than the demand. This is the well-established Alban W. Phillips curve, which has
been widely accepted and applied in the economic models of Western countries.
However, this model may not necessarily hold true in the socialist market economy
of China. As a second example, according to Keynes’s theory of consumption, it is
believed that, on average, when income grows, people tend to increase their con-
sumption. However, the degree of increase in consumption is not as high as that of
income. Assume that y stands for consumption, and x for income. Then, a math-
ematical expression for the relationship between these variables is

y ¼ f ðxÞ ¼ b0 þ b1 x þ e

where the parameter b1 ¼ dy=dx stands for the marginal consumption tendency,
and e a random noise, representing the inherent randomness of consumption.
According to Keynes, 0\b1 \1: However, Simon Kuznets does not agree with
Keynes’s opinion of a declining marginal consumption tendency. His work indi-
cates that there is a stable proportion of increase between consumption and income.
That is, the previous model is only a product of Keynes’s theory.
186 8 Combined Grey Models

(2) Variables and the form of the eventual model

The established model should reflect the objective economic activity. However,
it is impossible for such a reflection to include all details. This is why we need
reasonable assumptions. Employing the method of this section to select the major
variables to be included in the model using grey incidences will help to eliminate
minor relationships and factors. It focuses on the dominant connections while
simplifying the eventual model, making it convenient to handle and apply.
The specific works of this stage of model design include: (i) Determine which
variables to include, which ones are dependent variables, and which ones are
independent. Here, each independent variable is also known as illustrative variable.
(ii) Determine the number of parameters to be included in the model and their
(positive or negative) signs. (iii) Determine the mathematical form of the model
expression. Is it linear or nonlinear?
(3) Collection and organization of statistical data
After having decided on which variables to consider, one needs to collect all the
relevant data. That is the foundation of establishing models. Generally speaking, all
the collected raw data need to be statistically categorized and organized so that they
become the empirical evidence of the characteristics of the problem of concern and
are systematically usable for the purpose of modeling. The basic types of statistical
data, as discussed in Chap. 3, include behavioral sequences, time series, index
sequences, horizontal sequences, among others.
Step 2: Establish the GM (1, 1) model and obtain its simulated values. In order
to eliminate the random effect or error noise existing in the observational values of
individual variables of the model, establish the GM (1, 1) models for the individ-
ually observed sequences and then apply the simulated values of these GM (1, 1)
models as the base sequences on which to construct the eventual model.
Step 3: Estimate the parameters. After having designed the econometrics model,
the next task is to estimate the parameters, which are the constant coefficients of the
quantitative relationship between the chosen variables of the model. They connect
the individual variables within the model. More specifically, these parameters
explain how independent variables affect the dependent variable. Before using
observed data to make estimations, these parameters are unknown. After the form
of the model is established on the basis of the GM (1, 1) model, simulated
sequences solve the estimated values of the parameters using an appropriate
method, such as that of least squares estimate. As soon as the parameters are clearly
specified, the relationships between model variables become known and the model
can be determined.
The estimated values of the parameters provide realistic and empirical contents
and verification for the theories of economics. For instance, in the previously
mentioned consumption model, if the estimated value of parameter b1 is ^b1 ¼ 0:8; it
not only classifies the realistic content of the marginal consumption tendency, but
also provides a piece of evidence for the assumption of Keynes’s theory of con-
sumption that this parameter is between 0 and 1.
8.1 Grey Econometrics Models 187

Step 4: Test the model. After the parameters are estimated, the abstract model
becomes specific and determined. However, to determine whether or not the model
agrees with objective reality, and whether or not it can explain realistic economic
processes, it still has to go through tests. The tests consist of two aspects, the test of
economic meanings and statistical tests. The test of economic meanings checks
whether or not the individual estimated values of the parameters agree with eco-
nomic theories and relevant experiences. Statistical tests check the reliability of the
estimate, the effectiveness of the data sequence simulation, the correctness of
various econometrics assumptions, as well as the overall structure of the model and
its prediction ability using the principles of statistical reasoning. It is only after the
model passes through these tests that it can be applied in practice. If the model does
not pass the tests, then the model needs to be modified and improved.
Step 5: Apply the established model. Grey econometrics models have been
mainly employed to analyze economic structures, evaluate policies and decisions,
simulate economic systems, and predict economic development. Each application
process is also a process of verifying the model and its underlying theory. If the
prediction contains small errors, it means that the model is of high accuracy and
quality, with a strong ability to explain reality and an underlying theory that agrees
with reality. Otherwise, the model and the economic theory on which the model
was initially developed need to be modified.
Combined grey econometrics models can be employed not only to situations of
known system structures, but also to situations of system structures that need further
study and exploration. Combined grey econometrics models have produced satis-
factory results in practical applications. To this end, please consult Liu and Lin
(2006, pp. 247–254) to see how applications are carried out.

8.2 Combined Grey Linear Regression Models

Combined grey linear regression models can improve the weakness of original
linear regression models where no exponential growth is considered. They can also
improve the weakness of GM (1, 1) models that do not involve enough linear
factors. Thus, such combined models are suited for studying sequences with both
linear tendencies and exponential growth tendencies. For such a sequence, the
modeling process can be described as follows.


Definition 8.2.1 Assume that X ð0Þ ¼ xð0Þ ð1Þ; xð0Þ ð2Þ; . . .; xð0Þ ðnÞ is a sequence of


raw data. Its first order accumulation sequence is X ð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ; . . .;
xð1Þ ðnÞg:

^xð1Þ ðkÞ ¼ C1 emk þ C2 k þ C3 ð8:1Þ

is called a combined grey linear regression model, where v; C1 ; C2 ; C3 , are pa-

rameters that need to be estimated.
188 8 Combined Grey Models

In fact, combined grey linear regression model (8.1) is a simulation model of

X ð1Þ , which can be seen as the sum of a linear regression model of y ¼ ak þ b and
an exponential model of y ¼ C1 eak þ C2 .
From the model GM (1, 1), we can obtain

b b
^xð1Þ ðk þ 1Þ ¼ ðxð0Þ ð1Þ  Þeak þ ð8:2Þ
a a

Let C1 ¼ ðxð0Þ ð1Þ  baÞ; C3 ¼ ba ; which can be written as shown below:

^xð1Þ ðk þ 1Þ ¼ C1 eak þ C3 ð8:3Þ

By adding a linear term C2k to formula (8.3), we can obtain the same formula as
(8.1).
Lemma 8.2.1 Assume that X ð0Þ and X ð1Þ are the same as in Definition 8.2.1, then
the parameter v in formula (8.1) can be estimated by the following formula (8.4):
Pn3 Pn2m
V~ m ðkÞ
^¼
V m¼1 k¼1
ð8:4Þ
ðn  2Þðn  3Þ=2

where V ~m ðkÞ ¼ ln½ym ðk þ 1Þ=ym ðkÞ; ym ðkÞ ¼ xð1Þ ðk þ m þ 1Þ  xð1Þ ðk þ mÞ 

x ðk þ 1Þ þ xð1Þ ðkÞ; k; m ¼ 1; 2; . . .; n  3
ð1Þ

Theorem 8.2.1 Assume that X ð0Þ and X ð1Þ are the same as in Definition 8.2.1. Let
2 ð1Þ 3 2 m 3
x ð1Þ 2 3 e 1 1
6 xð2Þ ð2Þ 7 C1 6 e2m 2 1 7
6 7
X ð1Þ ¼ 6 . 7; C4 C2 5; A6 4 ... .. .. 7;
4 .. 5
C3 . . 5
x1 ðnÞ enm n 1

then we have the matrix form (8.5) of (8.1):

X ð1Þ ¼ AC ð8:5Þ

Therefore, we have

C ¼ ðAT AÞ1 AT X ð1Þ ð8:6Þ

With the estimated values of parameters v; C1 ; C2 ; C3 , the following formula

(8.7) can be used as a simulating or forecasting model:
^
^xð1Þ ðkÞ ¼ C1 eVk þ C2 k þ C3 ð8:7Þ
8.2 Combined Grey Linear Regression Models 189

Table 8.1 The original sequence of recorded subsides

Time 9502 9504 9506 9508 9510 9512 9602 9604
Amount of subside 12 22 31 43 51 57 75 83

From Eq. (8.7), it can be seen that if C1 ¼ 0; then the first order accumulation
sequence stands for a linear regression model. If C2 ¼ 0; then the accumulation
sequence stands for a GM (1, 1) model. This new model improves the weaknesses
of the original linear regression model with no exponential growth and that of the
GM (1, 1) model where no linear factors are considered.
By applying the inverse accumulation generation operator to Eq. (8.7), we can
obtain the simulated and predicted values X^ ð0Þ of the original sequence.

Example 8.2.1 At a certain observation station of ore and rock movement, the
sequence of recorded subsides of a specific location from February 1995 to April
1996 is given in Table 8.1. Try to make predictions for the sinking dynamics of this
specific location.
Solution: Due to the small amount of available data, grey systems models are the
most appropriate models for this prediction task. However, grey systems models
employ exponential functions to simulate accumulation generated sequences. They
are generally only suitable for modeling situations of exponential development, as it
is difficult for such models to describe linear tendencies of change. Therefore, in
this case study we will apply a grey linear exponential regression model to predict
the subsides of the specified location.
The original sequence of data is

X ð0Þ ¼ ð12; 22; 31; 43; 51; 57; 75; 83Þ:

Its first order accumulation sequence is

X ð1Þ ¼ ð12; 34; 65; 108; 159; 216; 291; 374Þ:

For different m values, from Eqs. (8.6) and (8.7) we obtain the estimated value
^ ¼ 0:02058096 for v: Also, from Eq. (6.10), we obtain the estimated value of C:
V

C ¼ ðAT AÞ1 AT X ð1Þ ¼ ð21750:995; 439:9523; 21751:078Þ:

Thus, the combined model of the first order accumulation generation sequence is

^xð1Þ ðkÞ ¼ 21750:995e0:02058096k  439:9523k  21751:078:

190 8 Combined Grey Models

Tab 8.2 Simulated and predicted values and their errors

Time 9502 9504 9506 9508 9510 9512 9602 9604 9606 9608
xð0Þ ðkÞ 12 22 31 43 51 57 75 83
^xð0Þ ðkÞ 12.34 21.75 31.35 41.15 51.15 61.36 71.79 82.43 93.29 104.38
Error (%) −2.85 1.15 −1.12 4.31 −0.30 −7.66 4.28 0.69

Out of this model, we obtain the simulated and predicted values for each of the
time moments as listed in Table 8.2.

8.3 Grey Cobb-Douglas Model

In this section, we study the Cobb-Douglas or production function model. Let K be

the capital input, L the labor input, and Y the production output. Then,

Y ¼ A0 ect K a Lb

is known as the C-D production function model, where a stands for capital elas-
ticity, c labor elasticity, and the parameter for the progress of technology. The
log-linear form of this production function model is given below:

ln Y ¼ ln A0 þ ct þ a ln K þ b ln L

For given time series data of the production output Y; capital input K; and labor
input L;

Y ¼ ðyð1Þ; yð2Þ;


; yðnÞÞ; K ¼ ðkð1Þ; kð2Þ;


; kðnÞÞ; and

L ¼ ðlð1Þ; lð2Þ;


; lðnÞÞ;

one can employ the method of multivariate least squares estimate to approximate
the parameters ln A0 , c, a, and b.
When Y; K; and L represent the time series of a specific department, district, or
business, it is often the case that, due to severe fluctuations existing in the data, the
estimated parameters contain errors leading to incorrect results. For instance, the
estimated coefficient c for progress of technology is too small or becomes a neg-
ative number; the estimated values a and b for elasticity go beyond their reasonable
ranges. Under such circumstances, if one considers using the GM (1, 1) simulated
data of Y; K; and L as the original data for their least squares estimates, then to a
certain degree they can eliminate some of the random fluctuations, produce more
reasonable estimated parameter values, and obtain a model that can more accurately
reflect the relationship between the production output and labor, and capital inputs
and the progress of technology.
8.3 Grey Cobb-Douglas Model 191

Definition 8.3.1 Assume that

Y^ ¼ ð^yð1Þ; ^yð2Þ;


; ^yðnÞÞ;

K^ ¼ ð^kð1Þ; ^kð2Þ;


; ^kðnÞÞ; and
^ ¼ ð^lð1Þ; ^lð2Þ;


; ^lðnÞÞ:
L

are respectively the GM (1, 1) simulated sequences of Y; K; and L: Then

Y^ ¼ A0 ect K
^ aL
^b is known as the grey model of production function.
In the grey production function model although no grey parameters appear
explicitly, it stands for an expression that combines the idea of grey systems
modeling into the C-D production function model. That is, this model possesses a
very deep intension of the greyness. It embodies the non-uniqueness principle of
solutions and the absoluteness principle of greyness. This is why, in practical
applications, this model has produced satisfactory results. To this end, please
consult Liu and Lin (2006, pp. 256–258) to see how applications are carried out.

8.4 Grey Artificial Neural Network Models

8.4.1 BP Artificial Neural Model and Computational

Schemes

Each artificial neural network is made up of a large amount of elementary infor-

mation processors, known as neurons or nodes. The model with multi-layered
nodes, or the scheme known as error back propagation, represents the currently well
developed and widely employed artificial neural network system and computational
method. It translates the input-output problem of an available sample into a non-
linear optimization problem. It is a powerful tool that can be employed to uncover
the laws and patterns hidden in large amounts of data. The use of artificial neural
networks to simulate data sequences has several latent advantages. First, it has the
ability to model multiple kinds of functions, including nonlinear functions, piece-
wise defined functions, among others. Secondly, artificial neural networks are
unlike the traditional methods of distinguishing data sequences, which, to work
properly, must have presumed types of functional relationships between data
sequences. This means that artificial neural networks can establish the needed
functional relationship by using the attributes and intension naturally existing in the
provided data variables, without presuming the kinds of distributions the parameters
satisfy. Thirdly, this method possesses the advantage of making use of available
information very efficiently, while avoiding the problem of losing the real meanings
and pictures of the data due to various combinations, such as additions of positive
and negative values of data mining methods. That is, the artificial neural networks
method is especially useful for improving the GM (1, 1) model.
192 8 Combined Grey Models

Fig. 8.1 A back propagation

neural network

Figure 8.1 shows a back propagation network with three layers. The network
consists of an input layer, an implicit (or latent) layer, and an output layer. An entire
process of learning consists of forward and backward propagation. The particular
scheme of learning is given below:
(1) Apply random numbers to initialize Wij (the connection weight between nodes
i and j of different layers) and hj (the threshold value of node jÞ;
(2) Feed in the preprocessed training samples fXPL g and fP YPK g;
(3) Compute the output of the nodes of each layer, Opj ¼ f i ðWij Ipi  hj Þ for the
pth sample point, where Ipi stands for the output of node i and the input of node j;
(4) Compute the information error of each layer. For the input layer,
dP pk ¼ Opk ðypk  Opk Þð1  Opk Þ; for the latent layer, Opi ¼ Opi ð1  Opi Þ
i dpi Wij ;
(5) For the backward propagation, the modifiers of the weights are Wij ðt þ 1Þ ¼
adpi Opi þ Wij ðtÞ; and the modifiers of the thresholds hj ðt þ 1Þ ¼ hj ðtÞ þ bdpi ,
where a stands for the learning factor and b the momentum factor for accel-
erated convergence; and
P P
(6) Calculate the error Ep ¼ ð p k ÞðOpk  Ypk Þ2 =2.

8.4.2 Steps in Grey BP Neural Network Modeling

The steps to establish a grey BP neural network model are as follows:


Step 1: Assume that a time series xð0Þ ðiÞ ; i ¼ 1; 2;


; n; is given. We then
obtain the restored values ^xð0Þ ðtÞ; i ¼ 1; 2;


; n; using the outputs of the
GM (1, 1) model
Step 2: Establish the back propagation network model for the error sequence

ð0Þ 
e ðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ ;k ¼ 1; 2;


; n
If the order of prediction is S; it means that we use the information of eð0Þ ði  1Þ;
eð0Þ ði  2Þ; …, eð0Þ ði  SÞ to predict the value at the ith moment; we will treat
8.4 Grey Artificial Neural Network Models 193

eð0Þ ði  1Þ; eð0Þ ði  2Þ; …, eð0Þ ði  SÞ as the input sample points of the back
propagation network training, while using the value of eð0Þ ðiÞ as the expected
prediction of the back propagation network training. By using the back propagation
computational scheme outlined earlier, train this network through enough amount
of cases of error sequences so that output values (along with empirical test values)
are produced in ways that correspond to different input vectors. The resultant
weights and thresholds represent the correct internal representations through the
self-learning and adaptation of the network. A well trained back propagation net-
work model can be an effective tool for error sequence prediction.

ð0Þ 
Step 3: Determine the simulation values of e ðkÞ ¼ xð0Þ ðkÞ  ^xð0Þ ðkÞ ;


k ¼ 1; 2;


; n: Assume that the simulation sequence is ^eð0Þ ðkÞ ;
k ¼ 1; 2;


; n; which is obtained by the BP neural network



Step 4: Based on ^xð0Þ ðiÞ and ^eð0Þ ðkÞ ;i; k ¼ 1; 2;


; n; we have the following
result

^xð0Þ ði; kÞ ¼ ^xð0Þ ðiÞ þ ^eð0Þ ðkÞ ð8:8Þ

which is the predicted sequence of the grey artificial neural network model.
Example 8.4.1 Given the actual yearly investments in environmental protection
over a period of time of a certain location, and the GM (1, 1) simulations and
relevant errors in Table 8.3, establish an artificial neural network model for the
error sequence.
Solution: Based on and using the GM (1, 1) error sequence data given in Table 8.3,
we apply the previously outlined method to establish a back propagation network
model. Our projected back propagation network will have three characteristic pa-
rameters, one latent layer, within which there are 6 nodes, and one input layer
within which there is one node. Let the learning rate be 0.6, the convergence rate
0.001, and the variance limited within the range of 0.01. Let us conduct the training
and testing of the network on a computer. Then, Table 8.4 lists the simulation
results of the combined back propagation network model.

Table 8.3 The GM (1, 1) simulations and errors

Year Investment xð0Þ ðiÞ GM (1, 1) simulation ^xð0Þ ðiÞ Errors eð0Þ ðkÞ
1985 110.20 110.20 0
1986 146.34 164.39 −19.05
1987 185.36 187.65 −2.29
1988 221.14 214.22 6.92
1989 255.16 244.54 10.52
1990 289.18 279.17 9.01
1991 320.54 319.69 1.85
1992 352.79 363.81 −11.02
194 8 Combined Grey Models

Table 8.4 Simulation results of the grey artificial neural network model
Year Actual value xð0Þ ðiÞ Simulated value ^xði; kÞ Relative errors (%)
1988 221.14 221.12 0.01
1989 255.16 255.29 0.05
1990 289.18 289.11 0.02
1991 320.54 320.79 0.08
1992 352.79 352.70 0.03

8.5 Grey Markov Model

8.5.1 Grey Moving Probability Markov Model

Definition 8.5.1 Assume that fXn ; n 2 Tg is a stochastic process. If for any whole
number n 2 T and any states i0 ; i1 ;


; in þ 1 2 I; the following conditional proba-
bility satisfies

PðXn þ 1 ¼ in þ 1 jX0 ¼ i0 ; X1 ¼ i1 ;


; Xn ¼ in Þ ¼ PðXn þ 1 ¼ in þ 1 jXn ¼ in Þ ð8:9Þ

then fXn ; n 2 Tg is known as a Markov chain. Equation (8.9) is seen without any
post-effect. It means that the future state of the system at t ¼ n þ 1 is only related to
the current state at t ¼ n; without any influence from any other earlier state
t  n  1:
For any n 2 T and states i; j 2 I; the following

pij ðnÞ ¼ PðXn þ 1 ¼ jjXn ¼ iÞ ð8:10Þ

is known as the transition probability of the Markov chain. If the transition prob-
ability pij ðnÞ in this equation does not have anything to do with the index n; then
fXn ;n 2 Tg is known as a homogeneous Markov chain. For such a Markov chain,
the transition probability pij ðnÞ is often denoted as pij . Because our discussion will
be mainly on homogeneous Markov chains, the word “homogeneous” will be
omitted. When all the transition probabilities pij ðnÞ are placed in a matrix, such as
P ¼ ½pij ; this matrix is referred to as the transition probability matrix of the sys-
tem’s state.
Proposition 6.1 The entries of the transition probability matrix P satisfy
ij  0; i; j 2 I; and
(1) pP
(2) j2I pij ¼ 1; i 2 I:

ðnÞ
The probability pij ¼ PðXm þ n ¼ jjXm ¼ iÞ; i; j 2 I; n  1 is known as the nth
ðnÞ
step transition probability of the given Markov chain, and PðnÞ ¼ ½pij  the nth step
transition probability matrix.
8.5 Grey Markov Model 195

Proposition 6.2 The nth step transition probability matrix PðnÞ satisfies
ðnÞ
(1) pij  0; i; j 2 I;
P ðnÞ
(2) pij ¼ 1; i 2 I; and
j2I
(3) PðnÞ ¼ Pn .

Any Markov chain with grey transition probabilities is known as a grey Markov
chain. When studying practical problems, due to a lack of sufficient information, it
is often difficult to determine the exact values of the transition probabilities. In such
cases, it might be possible to determine the grey ranges pij ðÞ of these uncertain
probabilities based on available information. When the transition probability matrix
~
is grey, the entries of its whitenization PðÞ ¼ ½P~ ij ðÞ are generally required to
satisfy the following properties:
~ ðÞ  0; i; j 2 I; and
(1) P
Pij
~
(2) j2I Pij ðÞ ¼ 1; i 2 I:

Proposition 6.3 Assume that the initial distribution of a finite-state grey Markov
chain is PT ð0Þ = ðp1 ; p2 ;


; pn Þ and the transition probability matrix PðÞ ¼
½Pij ðÞ: Then, the system’s distribution of the next sth state is

PT ðsÞ ¼ PT ð0ÞPs ðÞ ð8:11Þ

That is, when the system’s initial distribution and the transition probability
matrix are known, one can predict the system’s distribution for any future state.

8.5.2 Grey State Markov Model

Assume that a stationary process X ð0Þ satisfies the condition of Markov chains. If
we divide it into n states and each of the states i is expressed by

i ¼ ½ai ; bi ;


ði ¼ 1; 2;


; sÞ

where ai ; bi are constants and determined according to the states. The steps to
establish a grey state Markov model are outlined next.
Step 1: Determine the states for a stationary process X ð0Þ which satisfies the
condition of Markov chains

i ¼ ½ai ; bi ; ði ¼ 1; 2;


; sÞ
196 8 Combined Grey Models

Step 2: Compute the initial probability distribution. Assume that there are s
different states 1 ; 2 ;


; s . If state i ði ¼ 1; 2;


; sÞ occurs Mi times in total
in M experimentations, then the frequency of Mi can be calculated by

Mi
fi ¼ ði ¼ 1; 2;


; sÞ:
M

We can use fi ði ¼ 1; 2;


; sÞ as an approximation of the initial probability

pi ði ¼ 1; 2;


; sÞ; that is, let fi  pi ði ¼ 1; 2;


; sÞ:
Step 3: Compute the transition probability. Just like the initial probability, we
take the frequency as an approximation of the transition probability.
Firstly, we calculate the one step transition frequency of i ! j (from state i
transfer to state j through one step) by

fij ¼ f ðj ji Þ

If state i ði ¼ 1; 2;


; sÞ occurs Mi times in total in M experimentations, let Mij

be the number of transfers to the state j from Mi state i . Then we have

Mij
fij ¼
Mi

Then, if fij  pij , we have the transition probability matrix P ¼ ðpij Þs s .

Similarly, we can calculate the approximation of m steps transition probability as
follows (8.12):

Mij ðmÞ
pij ðmÞ ¼ ; ði ¼ 1; 2;


; sÞ ð8:12Þ
Mi

where Mij ðmÞ is the number of transfers to the state j from Mi state i through m
steps.
Step 4: Prediction using the transition probability. Assume that the object of
prediction is located at state k , then consider the kth row of P: If

max pkj ¼ pkl

j

then it can be inferred that, at the next time moment, the system will most likely
transform from state k to state l . If there are two or more entries in the kth row of
P that are equal or roughly equal, then the direction of change in the system’s state
is difficult to determine. In this case, one needs to look at the two-step or n-step
transition probability matrix Pð2Þ or PðnÞ , where n  3:
8.6 Combined Grey-Rough Model 197

8.6 Combined Grey-Rough Model

Grey systems theory and rough set theory are two mathematical tools developed to
address uncertain and incomplete information. To a certain degree they complement
each other. They both apply the idea of lowering the preciseness of expression of
the available data to gain the extra generality of the expression. In particular, grey
systems theory employs the method of grey sequence generations to reduce the
accuracy of data expressions, while rough set theory makes use of the idea of data
scattering to uncover patterns hidden in the data by ignoring unnecessary details.
Neither grey systems theory nor rough set theory requires any prior knowledge,
such as probability distribution or degree of membership. On one hand, rough set
theory investigates rough, non-intersecting classes and concepts of roughness, with
emphasis placed on the indistinguishability of objects. On the other hand, grey
systems theory focuses on grey sets with clear extension and unclear intension, with
emphasis placed on uncertainties caused by insufficient information. Thus, if rough
set theory and grey systems methodology are mixed, their individual weaknesses
both in theory and application can be improved so that greater theoretical strength
and practical applicability can be achieved.

8.6.1 Rough Membership, Grey Membership and Grey

Numbers

Rough set theory can be seen as an expansion of the classic set theory. It makes use
of rough membership functions to define rough sets, where each membership
function is explained and understood as those of conditional probabilities.
The concepts of rough approximation sets and rough membership functions of
the rough set theory are closely related to those of greyness of grey numbers. When
either lX ðxÞ ¼ 0 or lX ðxÞ ¼ 1; the object is assured either to belong or not to
belong to set X. In such cases, the classification is definite and clear; the involved
greyness is the smallest. If 0\lX ðxÞ\1; then object x belongs to set X with the
degree of confidence lX ðxÞ: In this case, object x projects a kind of grey state of
transition between definitely being in set X and definitely not being in X. When
lX ðxÞ = 0.5, the probability of object x to either belong to set X or not to belong to
X is 50 %. For this situation, the degree of uncertainty is the highest. That is, the
degree of greyness is the highest. When the rough membership function lX ðxÞ is
near 1 or 0, the uncertainty for object x to belong or not to belong to set X is
decreased, and the corresponding degree of greyness should also decrease. The
closer to 0.5 the rough membership is, the greater the uncertainty for object x to
belong or not to belong to set X; the corresponding degree of greyness is also
greater in such cases. We categorize all rough membership functions into two
groups: upper and lower rough membership functions, where a rough membership
function is upper if its values come from the interval [0.5, 1], denoted lX ðxÞ; the
198 8 Combined Grey Models

corresponding grey membership function is also referred to as upper and denoted by

gX ðxÞ: A lower rough membership function is one that takes values from the
interval [0, 0.5], denoted lX ðxÞ: The corresponding grey membership function is
referred to as a lower grey membership function, denoted gX ðxÞ:
Evidently, upper, lower and general rough membership functions satisfy the
following properties:
(1) lX ðxÞ ¼ 1  lX ðxÞ;
(2) lX [ Y ðxÞ ¼ lX ðxÞ þ lY ðxÞ  lX \ Y ðxÞ; and
(3) maxð0; lX ðxÞ þ lY ðxÞ  1Þ  lX \ Y ðxÞ  minð1; lX ðxÞ þ lY ðxÞÞ:
Based on the discussion above, we introduce the following definition of grey
membership functions using the concept of rough membership functions.
Definition 8.6.1 Assume that x is an object with its field of discourse U: That is,
x 2 U: Let X be a subset of U: Then mappings from U to the closed interval [0,1]:

lX : U ! ½0:5; 1; lj ! gX ðxÞ 2 ½0; 1; and

lX : U ! ½0; 0:5; lj ! gX ðxÞ 2 ½0; 1

are respectively referred to as upper and lower grey membership functions of X;

where lX  lX ; gX ðxÞ and gX ðxÞ are respectively referred to as upper and lower grey
membership functions of object x with respect to X:
The defined concept of grey membership functions based on rough membership
functions is depicted in Fig. 8.2.
Definition 8.6.2 Assume that x 2 U; X
U; the grey number scale of the uncer-
tainty for x to belong to X is gc , the grey number scale of the upper grey mem-
bership function gX ðxÞ is gc , and the grey number scale of the lower grey
membership function gX ðxÞ is gc . Then the greyness scales gc and gc of the upper
grey number and the lower grey number of the greyness scales gc of different grey
numbers are respectively given as outlined below.
The greyness of white numbers (gc ¼ 0Þ: if lX ðxÞ = 0, then gc = 0; if lX ðxÞ = 1,
then gc = 0.

Fig. 8.2 A conceptual

depiction of grey membership
functions
8.6 Combined Grey-Rough Model 199

For first class grey numbers ðgc ¼ 1Þ : if lX ðxÞ 2 (0, 0.1], then gc = 1;
if lX ðxÞ 2 [0.9,1), then gc = 1.
For second class grey numbers ðgc ¼ 2Þ : if lX ðxÞ 2 (0.1, 0.2], then gc = 2;
if lX ðxÞ 2 [0.8, 0.9), then gc = 2.
For third class grey numbers ðgc ¼ 3Þ : if lX ðxÞ 2 (0.2, 0.3], then gc = 3;
if lX ðxÞ 2 [0.7,0.8), then gc = 3.
For fourth class grey numbers ðgc ¼ 4Þ : if lX ðxÞ 2 (0.3, 0.4], then gc = 4;
if lX ðxÞ 2 [0.6,0.7), then gc = 4.
For fifth class grey numbers ðgc ¼ 5Þ : if lX ðxÞ 2 (0.4, 0.5), then gc = 5;
if lX ðxÞ 2 (0.5, 0.6), then gc = 5.
The greyness of black numbers ðgc [ 5Þ : if lX ðxÞ = 0.5, then gc = gc > 5.
When lX ðxÞ 2 [0, 1], gX ðxÞ = 0 and gX ðxÞ ¼ 1: In this case, there is no
uncertain information, so it is referred to as the greyness of white numbers. That is,
gc = gc = gc = 0. When lX ðxÞ = 0.5, gX ðxÞ = gX ðxÞ ¼ 1; the degree of uncertainty
for object x to belong or not to belong to set X reaches its maximum, which is
referred to as the greyness of black numbers gc [ 5: From Definition 8.6.1, it
follows that the higher the greyness of a grey number, the less clear the information
is; the lower the greyness of a grey number, the clearer the information is.
From Definition 8.6.1, it can be readily obtained that lX ðxÞ ¼ 1  lX ðxÞ: If we
use the greyness of the upper grey number to represent the degree of uncertainty for
object x to belong to set X, and the greyness of the lower grey number to illustrate
the degree of uncertainty for object x not to belong to set X, then these two degrees
of uncertainty are supplementary.
According to Definition 8.6.2, the scale of the greyness of a grey number is
determined by the grey interval to which the maximum rough membership value of
the information granularity could belong. Thus, the whitenizations of grey numbers
of different degrees of greyness are defined as the maximum possible rough
membership value of the grey numbers of corresponding scales. For example, if the
possible maximum rough membership value of a certain conditional subset com-
puted out of the available decision-making table is lX ðxÞ = 0.75, because
0.75 2 [0.7, 0.8), then lX ðxÞ = 0.75 stands for the white value of such a grey
number whose upper greyness is gc = 3.

8.6.2 Grey Rough Approximation

Definition 8.6.3 Assume that S ¼ ðU; A; V; f Þ; A ¼ C [ D; X

U; P
C; and the
greyness scale gc  5 of a grey number. Then
 
jIP ðxÞ \ Xj
apr gPc ðXÞ ¼ [  gc ð8:13Þ
jIP ðxÞj
200 8 Combined Grey Models

and
 
jIP ðxÞ \ Xj
apr gPc ðXÞ ¼ [ [ gc ð8:14Þ
jIP ðxÞj

are respectively referred to as the gc -lower approximation and gc -upper approxi-

mation of X with respect to IP , where the upper rough membership function cor-
responding to the upper scale gc of grey-number greyness satisfies lX ðxÞ 2 ð0:5; 1;
and the lower rough membership function corresponding to the lower scale gc of
grey-number greyness satisfies lX ðxÞ 2 ½0; 0:5Þ:
The gc -lower approximation of the set X
U under the grey-number greyness
scale gc equals the union of all the equivalence classes of U that belong to X; with
grey-number greyness scales less than or equal to the upper grey-number greyness
scale gc . The gc -upper approximation is equal to the intersection of all the equiv-
alence classes of U that belong to X; with grey-number greyness scales greater than
the lower grey-number greyness scale gc .
Definition 8.6.4 The quality of gc -classification is
n o
jX \ IP ðxÞj
j[ jIP ðxÞj  gc j
cgPc ðP; DÞ ¼ ð8:15Þ
jUj

The classification quality cgPc ðP; DÞ measures the percentage of the knowledge in the
field of discourse that can be clearly classified for a given grey-number greyness
scale gc  5, in the totality of current knowledge.
For a given grey-number greyness scale gc  5, let approximate reduction
redPgc ðC; DÞ stand for the set of attributes with the minimum condition that still
produces clear classification without containing any extra attributes.
In rough set theory, the classification of the elements located along the boundary
regions is not clear. Whether or not an element in such a region can be clearly
classified is determined most commonly by the pre-fixed greyness scale. The
concept of grey rough approximation so defined is analogous to that of variable
precision rough approximation. When the interval grey numbers in which the upper
greyness scale gc and the lower greyness scale gc of the grey-number greyness gc of
the grey rough approximation respectively belong to their corresponding white
values, the grey rough approximation is consequently transformed into rough
approximation under the meaning of variable precision rough sets. Evidently,
variable precision rough approximation can be seen as a special case of grey rough
approximation. When compared to models of variable precision rough sets of the
sets of variable precision, whether or not elements in a relatively rough set X can be
correctly classified is mostly determined by the pre-fixed maximum critical confi-
dence threshold parameter b. This is where classification can be done if smaller than
or equal to the upper bound of b, and indistinguishability appears when this upper
bound is surpassed. However, the parameter of the maximum critical confidence
8.6 Combined Grey-Rough Model 201

threshold b in general is difficult to determine beforehand, especially for large data

sets. In other words, the parameter of maximum critical confidence threshold b
generally stands for a grey number. Thus, the concept of interval grey numbers
provides a practical quantitative tool which appoints upper and lower endpoints.
For cases where we cannot obtain much information about the degree of accuracy
of the actual data, this method of representation becomes extremely useful.
Proposition 8.6.1 Given the greyness scale gc  5; the following hold true:
(1) apr gPc ðX [ YÞ  apr gPc ðXÞ [ apr gPc ðYÞ;
(2) apr gPc ðX \ YÞ
apr gPc ðXÞ \ apr gPc ðYÞ;
(3) apr gPC ðX [ YÞ  apr gPC ðXÞ [ apr gPC ðYÞ; and
(4) apr gPc ðX \ YÞ
apr gPc ðXÞ \ apr gPc ðYÞ:

Proof (1) For any X

U and Y
U; and given the greyness scale gc , we have

jIP ðxÞ \ ðX [ YÞj jIP ðxÞ \ Xj


jIP ðxÞj jIP ðxÞj

and

jIP ðxÞ \ ðX [ YÞj jIP ðxÞ \ Yj

 :
jIP ðxÞj jI P ðxÞj

Therefore, apr gPc ðX [ YÞ apr gPc ðXÞ [ apr gPc ðYÞ:

(2) For any X, Y
U, and given the greyness scale gc  5; we have

jIP ðxÞ \ ðX \ YÞj jIP ðxÞ \ Xj


jIP ðxÞj jIP ðxÞj

and

jIP ðxÞ \ ðX \ YÞj jIP ðxÞ \ Yj

 :
jIP ðxÞj jIP ðxÞj

Therefore, apr gPc ðX \ YÞ

apr gPc ðXÞ \ apr gPc ðYÞ: Similarly, we can prove (3) and
(4). QED.
Proposition 8.6.2 apr gPc ðXÞ
apr gPc ðXÞ:
Proof Let x 2 apr gPc ðXÞ: Because the equivalence relation IP is reflective, we have
x 2 IP ðxÞ: From Definition 8.6.2, it follows that gc  5 and that the interval grey
number to which the rough membership value corresponding to the upper
grey-number greyness scale belongs is greater than the interval grey number to
202 8 Combined Grey Models

which the rough membership value corresponding to the lower grey-number

greyness scale belongs. Hence, we have x 2 apr gPc ðXÞ and consequently apr gPc ðXÞ

apr gPc ðXÞ: QED.

8.6.3 Combined Grey Clustering and Rough Set Model

When employing the expansion dominant rough set model to probabilistic

decision-making, one needs to have a multi-criteria decision-making table.
However, in many practical applications involving uncertain multi-criteria
decision-making, the researcher has to rely on existing data sets to generate his
multi-criteria information table instead of being able to obtain their own
multi-criteria decision making table. For instance, we can easily collect the financial
data of a publically-traded company, such as income per share, net asset per share,
net profit, reliability, operating profit, and so on. Based on the collected financial
data, we can establish a multi-criteria information table. Given that such a com-
pany’s style of decision-making is unknown ahead of time, it is difficult to classify
it according to whether it presents a high risk, moderate risk, or low risk
decision-making style. Thus, it is also difficult, if not impossible, to generate a
relevant multi-criteria decision-making table. Therefore, dominant rough set models
and expanded dominant rough set models cannot be directly employed to conduct
decision-making analysis of these problems. However, the method of grey clus-
tering of grey systems theory generally groups objects into different preference
categories by considering attribute preference information and decision-makers’
preference behaviors. In particular, the method of grey fixed weight clustering
provides an effective way to transform a multi-criteria information table, which is
made of preferred attributes of various dimensions, into a multi-criteria
decision-making table. For instance, based on the collected financial data of
companies, the distributions of the preferred attributes’ values of the criteria, and
the preferred behaviors of the decision-makers, we can establish possibility func-
tions. On this basis, we can group the companies into different risk classes, such as
high risk, moderate risk, and low risk class.
When considering the strengths of the methods of dominant rough sets and grey
fixed weight clustering, we can construct a hybrid method combining grey fixed
weight clustering and dominant rough sets, where grey fixed weight clustering can
be seen as a processing tool used before the method of dominant rough sets is
employed. The purpose of doing so is to generalize the dominant rough sets to a
method that can be employed to conduct decision-making analysis based on
multi-criteria information tables, and to extract the most precise expression of
knowledge from the multi-criteria information table.
By following the steps below, one can establish the needed model combining
grey fixed weight clustering and dominant rough sets:
8.6 Combined Grey-Rough Model 203

(1) Develop a system of knowledge expressions using the values of preferred

conditional attributes (criteria);
(2) Determine the ordered decision-making evaluation grey classes g according to
the specific circumstances;
(3) Establish the possibility function for the field of each criterion. Let the
possibility function of the kth subclass of the jth criteria be fik ð
Þ
ðj ¼ 1; 2;


; m; k ¼ 1; 2;


; gÞ;
(4) Determine the clustering weight gj , j = 1, 2, …, m, for each criterion;
(5) Based on the observed value xij, i = 1, 2, …, n, j = 1, 2, …, m, of object i with
P
respect to criterion j, compute the coefficients rki ¼ m j¼1 fj ðxij Þgj of the grey
k

fixed weight clustering i = 1, 2, …, n, k = 1, 2, …, g;

(6) Obtain the clustering coefficient vector

X
m X
m X
m
ri ¼ ðr1i ; r2i ; . . .; rgi Þ ¼ ð fj1 ðxij Þgj ; fj2 ðxij Þgj ; . . . fjg ðxij Þgj Þ;
j¼1 j¼1 j¼1

(7) Generate the clustering coefficient matrix

2 3
r11 r21 . . . rg1
X 6 r12 r22 . . . rg2 7
6 7
¼ ðrki Þ ¼ 6 . .. .. .. 7 ;
4 .. . . . 5
r1n r2n . . . rgn

(8) Based on the clustering coefficient matrix R, determine the classes to which

individual objects belong. If max1  k  g frki g ¼ rki , then object i belongs to
grey class k ;
(9) Establish the decision-making table using preferred conditional attributes and
preferred decision-making grey classes; and
(10) Employ the method of dominant rough sets to conduct decision-making
analysis.

8.7 Practical Applications

Example 8.7.1 Let us look at how to choose regional key technologies using a
hybrid model combining the methods of grey fixed weight clustering and dominant
rough sets. For a specific geographic area, the evaluation criteria system and rel-
evant evaluation values for its key technologies are given in Table 8.5.
Based on the evaluations of relevant experts on 11 key technologies candidates,
we generate the knowledge expression system as shown in Table 8.6.
204 8 Combined Grey Models

Table 8.5 The criteria system for evaluating key regional technologies
Code Meaning of criterion Criterion Evaluation values
weight
a1 Time lag of technology 0.1 A: >10 years; B: 5–10 years; C:
3–5 years; D: <3 years
a2 Time length technological 0.09 A: >10 years; B: 5–10 years; C:
bottleneck existed 3–5 years; D: <3 years
a3 Ability to create own 0.14 A: complete own right; B: partial right;
knowledge right C: no right at all
a4 Coverage of technology 0.09 A: widely applicable; B: applied in
profession; C: special technique
a5 Promotion and lead of 0.11 A: strong; B: relatively strong; C:
technological fields general; D: weak
a6 Time needed for 0.07 A: within 1 year; B: 1–3 years; C:
technology transfer 4–5 years; D: >5 years
a7 Input/output ratio 0.13 A: high; B: relatively high; C: normal;
D: low
a8 Effect on environmental 0.12 A: strong; B: relatively strong; C:
protection normal; D: weak

Table 8.6 The knowledge U a1 a2 a3 a4 a5 a6 a7 a8

system on key regional
technologies n1 B B C B D C B A
n2 D D B B C B C D
n3 D D B B C A D A
n4 B C C B B B A C
n5 B B B B C B B C
n6 D D B B B B B C
n7 D D B C D A C C
n8 C B B C C C B C
n9 B B B B A B A B
n10 C B B B B B B B
n11 B B B B C B C B

In the following graph we present a decision-making analysis for this region’s

key technologies candidates.
For the evaluation criteria of the region’s key technologies, the preference orders
are the same as A > B > C > D. Quantify the set of criteria evaluations by letting
the set be V ¼ ðA; B; C; DÞ ¼ ð7; 5; 3; 1Þ: According to practical needs, we divide
each criterion into three grey classes of decision-making: the class of weak need for
key technologies (coded with 1), the class of general need for key technologies
(coded with 2), and the class of strong need for key technologies (coded with 3). Let
us take the possibility function of the class of weak need as the measurement of the
low bound, that of the class of general need as the moderate measurement, and that
8.7 Practical Applications 205

Fig. 8.3 Possibility functions

of the three grey classes

of the class of strong need as the measurement of the upper bound. For details, see
Fig. 8.3. Based on decision-making goals and specific distributions of experts’
evaluation values, we introduce the possibility functions for each grey class as
shown in Table 8.7.
P
From formula r ki ¼ m j¼1 fj ðxij Þ
gj , we can compute the clustering coefficient
k

for each grey class of each key technology. Based on such coefficients we can
establish the evaluation decision-making (Table 8.8) for choosing key technologies
for the region.
Because the values of all the conditional attributes have the preference order
A > B > C > D, these attributes contain preference information. Based on the
decision-making attributes, the comprehensive evaluation can be divided into three
preference ordered classes: Cl1 = {1}, Cl2 = {2}, Cl3 = {3}. Based on this result,
we divide the field of discourse and obtain the following unions of the
decision-making classes:
Cl1 ¼ Cl1 , with the comprehensive evaluation 1 (the need for key technologies is
weak);
Cl2 ¼ Cl1 [ Cl2 , with the comprehensive evaluation  2 (the need for key
technologies is at most moderate);
Cl2 ¼ Cl2 [ Cl3 , with the comprehensive evaluation  2 (the need for key
technologies is at least moderate);
Cl3 ¼ Cl1 [ Cl2 [ Cl3 , with the comprehensive evaluation  3 (the need for key
technologies is at most strong); and
Cl3 ¼ Cl3 , with the comprehensive evaluation 3 (the need for key technologies is
strong).
A reduction found by using the method of dominant rough sets is {a2, a7}. The
sets D  and D  of the least amounts of preference rules generated from this
reduction are respectively given in Tables 8.9 and 8.10.
Based on the set D  of preference decision-making rules generated by
employing our hybrid model that combines grey fixed weight clustering and
 206

Table 8.7 Possibility functions for key regional technologies

Criterion name Weak need class (1) Moderate need class (2) Strong need class (3)
8 8 8
Time lag of technology < 1 1  x\2 < x  2 2  x\3 < 0:5x  1:5 3  x\5
f ðxÞ ¼ 3  x 2  x\3 f ðxÞ ¼ 4  x 3  x\4 f ðxÞ ¼ 1 5x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Time length technological bottleneck < 1 1  x\2 < x  2 2  x\3 < 0:5x  1:5 3  x\5
existed f ðxÞ ¼ 3  x 2  x\3 f ðxÞ ¼ 4  x 3  x\4 f ðxÞ ¼ 1 5x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Ability to create own knowledge right < 1 3  x\4 < x  4 4  x\5 < x  5 5  x\6
f ðxÞ ¼ 5  x 4  x\5 f ðxÞ ¼ 6  x 5  x\6 f ðxÞ ¼ 1 6x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Coverage of technology < 1 3  x\4 < x  4 4  x\5 < x  5 5  x\6
f ðxÞ ¼ 5  x 4  x\5 f ðxÞ ¼ 6  x 5  x\6 f ðxÞ ¼ 1 6x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Promotion and lead of technological fields < 1 1  x\2 < 0:5x  1 2  x\4 < 0:5x  2 4  x\6
f ðxÞ ¼ 2  0:5x 2  x\4 f ðxÞ ¼ 3  0:5x 4  x\6 f ðxÞ ¼ 1 6x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Time needed for technology transfer < 1 1  x\2 < 0:5x  1 2  x\4 < 0:5x  2 4  x\6
f ðxÞ ¼ 2  0:5x 2  x\4 f ðxÞ ¼ 3  0:5x 4  x\6 f ðxÞ ¼ 1 6x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 8
Input/output ratio < 1 1  x\3 < x  3 3  x\4 < 0:5x  2 4  x\6
f ðxÞ ¼ 4  x 3  x\4 f ðxÞ ¼ 3  0:5x 4  x\6 f ðxÞ ¼ 1 6x7
: : :
0 otherwise 0 otherwise 0 otherwise
8 8 
Effect on environmental protection < 1 1  x\3 < 0:5x  1:5 3  x\5 0:5x  25 5x7
f ðxÞ ¼
f ðxÞ ¼ 2:5  0:5x 3  x\5 f ðxÞ ¼ 3:5  0:5x 5  x\7 0 otherwise
: :
0 otherwise 0 otherwise
8 Combined Grey Models
8.7 Practical Applications 207

Table 8.8 Evaluation decision-making table for key regional technologies

U a1 a2 a3 a4 a5 a6 a7 a8 Cl
n1 B B C B D C B A 3
n2 D D B B C B C D 1
n3 D D B B C A D A 1
n4 B C C B B B A C 3
n5 B B B B C B B C 2
n6 D D B B B B B C 1
n7 D D B C D A C C 1
n8 C B B C C C B C 2
n9 B B B B A B A B 3
n10 C B B B B B B B 2
n11 B B B B C B C B 2

Table 8.9 Set D  of preference rules

Rule Confidence Support
number
If the length of time for technology bottleneck to exist  C and 100 % 2
input/output ratio = A, then the urgency for needing key
technologies = 3 (strong)
If the length of time for technology bottleneck to exist  B and 100 % 5
input/output ratio  C, then the urgency for needing key
technologies  2 (moderate)
If the length of time for technology bottleneck to exist = D, then 100 % 4
the urgency for needing key technologies = 1 (weak)

Table 8.10 Set D  of preference rules

Rule Confidence Support
number
If the length of time for technology bottleneck to exist  C and 100 % 2
input/output ratio = A, then the need for key technologies = 3
(strong)
If the length of time for technology bottleneck to exist  B and 100 % 1
input/output ratio  C, then the need for key technologies  2
(moderate)
If the length of time for technology bottleneck to exist = D, then 100 % 4
the need for key technologies = 1 (weak)

dominant rough sets, all the 11 key technologies considered are correctly classified.
That is, the quality of classification is 100 %. Based on the set D  of preference
decision-making rules, a total of 7 key technologies are classified correctly so that
the quality of classification is 63.6 %.
Chapter 9
Techniques for Grey Systems Forecasting

9.1 Introduction

No matter what needs to be done, one should always get familiar with the situation,
think through the details, make educated predictions, and lay out a detailed plan
before he could potentially arrive at his desired successful conclusions. For matters
as great as international affairs, national events and citizens’ lives, the development
of regional or business entities, and for matters as small as daily work or living
arrangements, scientifically sound predictions are needed everywhere.
Prediction is about foretelling the possible course of development of societal
events, political matters, economic ups and downs, and so on, using scientific
methods and techniques based on attainable historical and present data so that
appropriate actions can be planned and carried out. In short, prediction is about
making scientific inferences regarding the evolution of materials and events ahead
of time. General prediction includes not only static inference about unknown
matters based on what is known within a specific time frame, but also dynamic
inference about the future based on history and the present state of affairs of a
certain matter. A specific prediction is a dynamic forecast within which a scientific
inference about the future evolution of a certain event is given.
Grey prediction makes scientific, quantitative forecasts about the future states of
systems based on understandings of unascertained characteristics of such systems.
It makes use of sequence operators on the original data sequences in order to
generate, treat, and excavate the hidden laws of systems evolution, so that grey
systems models can be established to predict future outcomes. All the methods of
the grey systems theory studied so far can be employed to make predictions. For a
given problem, the appropriate prediction model is chosen by making use of the
conclusions of a sufficiently and carefully done qualitative analysis. Also, the
choice of models should vary along with changing conditions. Each model chosen
has to be tested through many different methods in order to decide its

© Springer Science+Business Media Singapore 2017 209

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_9
210 9 Techniques for Grey Systems Forecasting

appropriateness and effectiveness. Only the models that pass various tests can be
meaningfully employed to make predictions.
Definition 9.1.1 Let X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ be a sequence of raw
^ ð0Þ ¼ ð^xð0Þ ð1Þ; ^xð0Þ ð2Þ;    ; ^xð0Þ ðnÞÞ the simulated data out of a chosen pre-
data, X


diction model, eð0Þ ¼ ðeð1Þ; eð2Þ; . . .eðnÞÞ = xð0Þ ð1Þ  ^xð0Þ ð1Þ; xð0Þ ð2Þ  ^xð0Þ ð2Þ,

…, xð0Þ ðnÞ  ^xð0Þ ðnÞ the error sequence, and
     
 eð1Þ   eð2Þ   eðnÞ 
D ¼  ð0Þ ;  ð0Þ ;    ;  ð0Þ  ¼ fDk gn1
    
x ð1Þ x ð2Þ x ðnÞ

the relative error sequence. Then:

 
 eðkÞ 
(1) For k  n, Dk ¼ xð0Þ ðkÞ
is known as relative error of the simulation at point k,
 Pn
and D ¼ n k¼1 Dk the average relative error;
1

(2) 1  D  is known as the average relative accuracy, and 1  Dk the simulation

accuracy at point k, k ¼ 1; 2;    ; n; and
(3) For a given a, when D\a and Dn \a hold true, the prediction model is said to
be error-satisfactory.

Definition 9.1.2 Let e stand for the absolute degree of incidence between the raw
^ ð0Þ . If for a given e0 [ 0 the absolute degree e of
data X ð0Þ and the simulated values X
incidence satisfies e [ e0 , then the simulation model is said to be incidence
satisfactory.
^ ð0Þ , and eð0Þ are the same as
Definition 9.1.3 Assume that the sequences X ð0Þ , X
above, and consider the relevant means and variances

1X n
1X n
x ¼ xð0Þ ðkÞ; S21 ¼ ðxð0Þ ðkÞ  xÞ2
n k¼1 n k¼1

and
1X n
1X n
e ¼ eðkÞ ; S22 ¼ ðeðkÞ  eÞ2 :
n k¼1 n k¼1

(1) If for a given C0 [ 0, the ratio of root-mean-square deviation (RMSD) is

C ¼ SS21 < C0 , then the model is said to be RMSD ratio satisfactory.
(2) If p ¼ PðjeðkÞ  eÞj\0:6745S1 Þ is seen as a small error probability and for a
given p0 [ 0, when p [ p0 , then the model is said to be small-error proba-
bility satisfactory.
The discussion above shows three different ways to test a chosen model. Each of
them is based on observations of the error to determine the accuracy of the model.
9.1 Introduction 211

Table 9.1 Commonly used scales of accuracy for model testing

Threshold Relative Degree of Variance Small error
accuracy scale error a incidence e0 ratio C0 probability p0
1st level 0.01 0.90 0.35 0.95
2nd level 0.05 0.80 0.50 0.80
3th level 0.10 0.70 0.65 0.70
4th level 0.20 0.60 0.80 0.60

For both the mean relative error D and the simulation error, the smaller they are, the
better. With regards to the degree of incidence e, the greater it is the better. As for
the RMSD ratio C, the smaller the value is, the better. This is because a small C
indicates that S2 is relatively small, while S1 is relatively large. This means that the
error variance is small while the variance of the original data is large, so that the
errors are relatively more concentrated with little fluctuation compared to the
original data. Therefore, for better simulation results, the smaller S2 is when
compared to S1 , the better. With regards to small error probability p, as soon as a set
of a, e0 , C0 , and p0 values are chosen, a scale of accuracy for testing models is
determined. The most commonly used scales of accuracy for testing models are
listed in Table 9.1.
In most applications published so far in the area of grey systems, the most
commonly used is the criterion of relative errors.

9.2 Interval Forecasting

If a given sequence of raw data is chaotic and it is difficult for any model to pass the
accuracy test, the researcher will then have trouble producing accurate quantitative
predictions. In this case, one can consider providing a range for future values to fall
within.
Definition 9.2.1 Let XðtÞ be a zigzagged line. If there are smooth and continuous
curves fu ðtÞ and fs ðtÞ, satisfying that for any t, fu ðtÞ\XðtÞ\fs ðtÞ, then fu ðtÞ is
known as the lower bound function of XðtÞ, fs ðtÞ the upper bound function, and
S = fðt; XðtÞÞjXðtÞ 2 ½fu ðtÞ; fs ðtÞg the value domain of XðtÞ.
If the upper and lower bound of XðtÞ are the same kind of function, then S is
known as a uniform domain. S is known as a predicted domain if t > n. When S is a
uniform domain with exponential functions as its upper and lower bounds fu ðtÞ and
fs ðtÞ, then S is known as a uniform exponential domain. If a uniform domain S has
linear upper and lower bound functions fu ðtÞ and fs ðtÞ, then S is known as a uniform
linear domain or a straight domain for short. If for t1 \t2 , fs ðt1 Þ  fu ðt1 Þ\fs ðt2 Þ 
fu ðt2 Þ always holds true, then S is known as a trumpet-like domain.
Example 9.2.1 Let X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ be a sequence of raw data,
and its accumulation generation be X ð1Þ ¼ ðxð1Þ ð1Þ; xð1Þ ð2Þ;    ; xð1Þ ðnÞÞ. Define
212 9 Techniques for Grey Systems Forecasting

rmax ¼ max fxð0Þ ðkÞg; rmin ¼ min fxð0Þ ðkÞg

1kn 1kn

and respectively take the upper and lower bound functions fu ðn þ tÞ and fs ðn þ tÞ of
X ð1Þ as follows:

fu ðn þ tÞ ¼ xð1Þ ðnÞ þ trmin ; fs ðn þ tÞ ¼ xð1Þ ðnÞ þ trmax :

That is, both the upper and lower bound functions of a proportional band are
increasing straight lines of time with slopes rmin and rmax , respectively.
Then S ¼ fðt; XðtÞÞjt [ n; XðtÞ 2 ½fu ðtÞ; fs ðtÞg is known as the proportional
domain (see Fig. 9.1).
ð0Þ
Example 9.2.2 For a sequence X ð0Þ of raw data, let Xu be the sequence corre-
ð0Þ
sponding to the curve that connects all the low points of X ð0Þ , and Xs the sequence
corresponding to the curve of all the upper points of X ð0Þ . Assume that
 
bu bu
^xuð1Þ ðk þ 1Þ ¼ xuð0Þ ð1Þ  expðau kÞ þ
au au

and
 
bs bs
^xsð1Þ ðk þ 1Þ ¼ xsð0Þ ð1Þ  expðas kÞ þ
as as

ð0Þ ð0Þ
are respectively the GM (1, 1) time response sequences of Xu and Xs . Then

S ¼ fðt; XðtÞÞXðtÞ 2 ½X ^sð1Þ ðtÞg
^uð1Þ ðtÞ; X

is known as a wrapping domain (see Fig. 9.2).

Example 9.2.3 For a given sequence X ð0Þ of raw data, let us take m different
sub-sequences to establish m GM (1, 1) models with the corresponding parameters

Fig. 9.1 A trumpet-like

domain
9.2 Interval Forecasting 213

Fig. 9.2 A wrapping domain

ai ¼ ½ai ; bi T ; i ¼ 1; 2;    ; m. Let
^
amax ¼ max fai g, amin ¼ min fai g, and
1im 1im

 
bmin bmin
^xuð1Þ ðk þ 1Þ ¼ xuð0Þ ð1Þ  expðamin kÞ þ
amin amin
 
bmax bmax
^xsð1Þ ðk þ 1Þ ¼ xsð0Þ ð1Þ  expðamax kÞ þ
amax amax

 ^uð1Þ ðtÞ; X
^sð1Þ ðtÞg is known as a development
Then S ¼ fðt; XðtÞÞXðtÞ 2 ½X
domain. The wrapping domain and development domain are exponential domains.
Definition 9.2.2 For a sequence X ð0Þ ¼ ðxð0Þ ð1Þ; xð0Þ ð2Þ;    ; xð0Þ ðnÞÞ of raw data,
let fu ðtÞ and fs ðtÞ be a upper and a lower bound function of the accumulation
sequence X ð1Þ of X ð0Þ . For any k [ 0,

1
^xð0Þ ðn þ kÞ ¼ ½fu ðn þ kÞ þ fs ðn þ kÞ
2
ð0Þ
is known as basic prediction value, and ^xu ðn þ kÞ ¼ fu ðn þ kÞ and
ð0Þ
^xs ðn þ kÞ ¼ fs ðn þ kÞ, respectively, the lowest and highest predicted values.
Example 9.2.4 The data (in tens of thousands) for car sales in a certain city are
given as follows:

X ð0Þ ¼ðxð0Þ ð1Þ; xð0Þ ð2Þ; xð0Þ ð3Þ; xð0Þ ð4Þ; xð0Þ ð5Þ; xð0Þ ð6ÞÞ
¼ð5:0810; 4:6110; 5:1177; 9:3775; 11:0574; 11:3524Þ

where xð0Þ ð1Þ = 5.0810 is the annual sales for the year of 2010, …, and
xð0Þ ð6Þ = 11.3524 for the year of 2015. Try to make a prediction using development
domain.
214 9 Techniques for Grey Systems Forecasting

Solution: Take the following sub-sequences

ð0Þ
X1 ¼ðxð0Þ ð1Þ; xð0Þ ð2Þ; xð0Þ ð3Þ; xð0Þ ð4Þ; xð0Þ ð5Þ; xð0Þ ð6ÞÞ
¼ð5:0810; 4:6110; 5:1177; 9:3775; 11:0574; 11:3524Þ

ð0Þ
X2 ¼ðxð0Þ ð1Þ; xð0Þ ð2Þ; xð0Þ ð3Þ; xð0Þ ð4Þ; xð0Þ ð5ÞÞ
¼ð5:0810; 4:6110; 5:1177; 9:3775; 11:0574Þ

ð0Þ
X3 ¼ðxð0Þ ð2Þ; xð0Þ ð3Þ; xð0Þ ð4Þ; xð0Þ ð5Þ; xð0Þ ð6ÞÞ
¼ð4:6110; 5:1177; 9:3775; 11:0574; 11:3524Þ

ð0Þ
X4 ¼ðxð0Þ ð3Þ; xð0Þ ð4Þ; xð0Þ ð5Þ; xð0Þ ð6ÞÞ
¼ð5:1177; 9:3775; 11:0574; 11:3524Þ

Based on each of these sub-sequences, let us establish the corresponding the

models of EGM (1, 1):

dxð1Þ
þ ai xð1Þ ¼ bi ; i ¼ 1; 2; 3; 4
dt

Their individual parameters ^ai ¼ ½ai ; bi T ; i ¼ 1; 2; 3; 4, are given below:

a1 ¼ ½a1 ; b1 T ¼ ½0:2202; 3:4689T ; ^a2 ¼ ½a2 ; b2 T ¼ ½0:3147; 2:1237T

^

a3 ¼ ½a3 ; b3 T ¼ ½0:2013; 5:0961T ; ^a4 ¼ ½a4 ; b4 T ¼ ½0:0911; 8:7410T

^

Because

amin ¼ min fai g ¼ minf0:2202; 0:3147; 0:2013; 0:0911g ¼ 0:0911 ¼ a4

1i4

amax ¼ max fai g ¼ maxf0:2202; 0:3147; 0:2013; 0:0911g ¼ 0:3147 ¼ a2

1i4

the upper bound time response sequence of the development domain is

8  
>
< ^xð1Þ ðk þ 1Þ ¼ xð0Þ ð1Þ  b2 ea2 k þ b2 ¼ 11:8293e0:3147k  6:7483
s
a2 a2
>
: ^xð0Þ ðk þ 1Þ ¼ ^xð1Þ ðk þ 1Þ  ^xð1Þ ðkÞ
s s s

ð0Þ
That is, ^xs ðk þ 1Þ ¼ 11:8293e0:3147k  11:8293e0:3147k0:3147 ¼ 3:1938e0:3147k .
ð0Þ ð0Þ
Thus, the highest predicted values are ^xs ð7Þ = 21.1029, ^xs ð8Þ = 28.9078, and
9.2 Interval Forecasting 215

ð0Þ ð0Þ
^xs ð9Þ = 39.5993. Because the starting value of X4 is xð0Þ ð3Þ, the lower bound
time response sequence of the development domain is
8  
>
< ^xð1Þ ðk þ 3Þ ¼ xð0Þ ð3Þ  b4 ea4 k þ b4 ¼ 101:0672e0:0911k  95:9495
u
a4 a4
>
: ^ð0Þ ð1Þ ð0Þ
xu ðk þ 3Þ ¼ ^xu ðk þ 3Þ  ^xu ðk þ 2Þ

That is, ^xð0Þ

u ðk þ 3Þ ¼ 101:0672e
0:0911k
 101:0672e0:0911k0:0911 ¼ 8:8003e0:0911k .
ð0Þ ð0Þ
Therefore, we obtain the lowest predicted values ^xu ð7Þ = 12.6694, ^xu ð8Þ =
ð0Þ
13.8777, and ^xu ð9Þ = 15.2014. From the highest and lowest predicted values, we
obtain the basic prediction values:

1
^xð0Þ ð7Þ ¼ ½^xsð0Þ ð7Þ þ ^xuð0Þ ð7Þ ¼ 16:8862
2
1
^xð0Þ ð8Þ ¼ ½^xsð0Þ ð8Þ þ ^xuð0Þ ð8Þ ¼ 21:3928
2
1
^xð0Þ ð9Þ ¼ ½^xsð0Þ ð9Þ þ ^xuð0Þ ð9Þ ¼ 27:4004
2

Based on the qualitative analysis of the estimated amount of car ownership in the
given city and the improvement in public transportation systems, we conclude that
the lowest predicted values are the most reliable.

9.3 Grey Disaster Forecasting

The basic idea of grey disaster prediction is essentially the prediction of abnormal
values. The kinds of values that are considered abnormal are commonly determined
based on individuals’ experiences. The objective of grey disaster predictions is to
provide the time moments of the forthcoming abnormal values so that relevant
parties can prepare for the worst ahead of time.
Definition 9.3.1 Let X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ be a sequence of raw data. Then
(1) For a given upper abnormal (or catastrophe) value n, the sub-sequence of X

Xn ¼ ðx½qð1Þ; x½qð2Þ;    ; x½qðmÞÞ ¼ fx½qðiÞjx½qðiÞ  n; i ¼ 1; 2;    ; mg

is known as the upper catastrophe sequence.

216 9 Techniques for Grey Systems Forecasting

(2) For a given lower abnormal (or catastrophe) value f, the sub-sequence

Xf ¼ ðx½qð1Þ; x½qð2Þ;    ; x½qðlÞÞ ¼ fx½qðiÞjx½qðiÞ  f; i ¼ 1; 2;    ; lg

is known as the lower catastrophe sequence. Together, these upper and lower
catastrophe sequences are referred to as catastrophe sequences. Because the idea
behind the discussion of catastrophe sequences is the same, in the following dis-
cussion we will not distinguish between upper and lower catastrophe sequences.
Definition 9.3.2 Assume that X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ is a sequence of raw data.
The following sub-sequence of X

Xn ¼ ðx½qð1Þ; x½qð2Þ;    ; x½qðmÞÞ  X

is a catastrophe sequence. Then,

Qð0Þ ¼ ðqð1Þ; qð2Þ;    ; qðmÞÞ

will be referred to as the catastrophe date sequence.

Disaster prediction is about finding patterns, if any, through the study of
catastrophe date sequences to predict future dates of occurrences of catastrophes. In
grey system theory, each disaster prediction is realized through establishing GM
(1, 1) models for relevant catastrophe date sequences.
Definition 9.3.3 If Qð0Þ ¼ ðqð1Þ; qð2Þ;    ; qðmÞÞ is a catastrophe date sequence,
the following

Qð1Þ ¼ ðqð1Þð1Þ ; qð2Þð1Þ ;    ; qðmÞð1Þ Þ

is the 1-AGO sequence of the catastrophe date sequence Qð0Þ , Z ð1Þ is the adjacent
neighbor mean generated sequence of Qð1Þ , and

qðkÞ þ azð1Þ ðkÞ ¼ b

is referred to as a catastrophe model of GM (1, 1). For the available sequence

X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ of raw data, if n stands for the present and the last entry
qðmÞð  nÞ in the corresponding catastrophe date sequence Qð0Þ represents when the
last catastrophe occurred, then the predicted value ^qðm þ 1Þ represents the next
forthcoming catastrophe and for any k [ 0, ^qðm þ kÞ stands for the predicted date
for the kth catastrophe to occur in the future.
Example 9.3.1 The following sequence gives the annual average precipitations (in
mm) of a certain region for 17 years, where x(1), x(2), …, x(17) are respectively the
data for the years of 1998, 1999, …, 2014:
9.3 Grey Disaster Forecasting 217

X ¼ðxð1Þ; xð2Þ; xð3Þ; xð4Þ; xð5Þ; xð6Þ; xð7Þ; xð8Þ; xð9Þ; xð10Þ
xð11Þ; xð12Þ; xð13; xð14Þ; xð15Þ; xð16Þ; xð17ÞÞ
¼ð390:6; 412:0; 320:0; 559:2; 380:8; 542:4; 553:0; 310:0; 561:0; 300:0
632:0; 540:0; 406:2; 313:8; 576:0; 586:6; 318:5Þ

Take n ¼ 320 mm as a lower abnormal (drought) value. Carry out a drought

prediction for this specific region.
Solution: If n ¼ 320, we obtain the following lower catastrophe sequence

Xn ¼ ðxð3Þ; xð8Þ; xð10Þ; xð14Þ; xð17ÞÞ ¼ ð320:0; 310:0; 300:0; 313:8; 318:5Þ

with the corresponding catastrophe date sequence

Qð0Þ ¼ ðqð1Þ; qð2Þ; qð3Þ; qð4Þ; qð5ÞÞ ¼ ð3; 8; 10; 14; 17Þ

and it’s 1-AGO sequence

Qð1Þ ¼ ð3; 11; 21; 35; 52Þ

The mean sequence based on consecutive neighbors of Q(1) is given by

Z ð1Þ ¼ ð7; 16; 28; 43:5Þ

Let qðkÞ þ azð1Þ ðkÞ ¼ b. From

2 3 2 3
7 1 8
6 16 17 6 10 7
B¼6
4 28
7;Y ¼6 7
15 4 14 5
43:5 1 17

it follows that
   
a 0:25361
^a ¼ ¼ ðBT BÞ1 BT Y ¼
b 6:258339

Therefore, the GM (1, 1) ordinality response sequence of the catastrophe date

sequence is

^qð1Þ ðk þ 1Þ ¼ 27:667e0:25361k  24:667

^qðk þ 1Þ ¼ ^qð1Þ ðk þ 1Þ  ^qð1Þ ðkÞ

218 9 Techniques for Grey Systems Forecasting

That is,

^
qðk þ 1Þ ¼ 27:667e0:25361k  24:667e0:25361ðk1Þ ¼ 6:1998e0:25361k

Thus, we can obtain a simulated sequence for Qð0Þ as follows:

^ ð0Þ ¼ð^qð1Þ; ^qð2Þ; ^qð3Þ; ^qð4Þ; ^qð5ÞÞ

Q
¼ð6:1998; 7:989; 10:296; 13:268; 17:098Þ

From

eðkÞ ¼ qðkÞ  ^qðkÞ; k ¼ 1; 2; 3; 4; 5

we obtain the error sequence as follows:

eð0Þ ¼ðeð1Þ; eð2Þ; eð3Þ; eð4Þ; eð5ÞÞ

¼ð3:1998; 0:011; 0:296; 0:732; 0:098Þ

And from
 
 eðkÞ 
Dk ¼  ; k ¼ 1; 2; 3; 4; 5
qðkÞ

it follows that the sequence of relative errors is

D ¼ ðD2 ; D3 ; D4 ; D5 Þ ¼ ð0:1 %; 2:96 %; 5:1 %; 0:6 %Þ

From this sequence, we calculate the average relative error

X5
 ¼1
D Dk ¼ 2:19 %
4 k¼2

With 1  D ¼ 97:81 % as the average relative accuracy, and 1  D5 ¼ 99:4 %.

Therefore, we can use

^qðk þ 1Þ ¼ 6:1998e0:25361k

to carry out our predictions. Because

^
qð5 þ 1Þ ¼ ^qð6Þ  22; ^qð6Þ  ^qð5Þ  22  17 ¼ 5

we predict that in 5 years, counting from the time of the last drought in 2019, there
might be a drought. In order to improve the accuracy of our prediction, we can take
several different abnormal values to build various models to make predictions.
9.4 Wave Form Forecasting 219

9.4 Wave Form Forecasting

When the available data sequence vibrates widely with large magnitudes, it is often
difficult, if not impossible, to find an appropriate simulation model. In this case, one
can consider making use of the pattern of fluctuation of the data to predict the future
development of the wavy movement. This kind of prediction is known as a wave
form forecasting.
Definition 9.4.1 Let X ¼ ðxð1Þ; xð2Þ;    ; xðnÞÞ be the sequence of raw data, then

xk ¼ xðkÞ þ ðt  kÞ½xðk þ 1Þ  xðkÞ

is known as a k-piece zigzagged line of the sequence X, and

fxk ¼ xðkÞ þ ðt  kÞ½xðk þ 1Þ  xðkÞjk ¼ 1; 2;    ; n  1g

the zigzagged line, still denoted by using X.

Definition 9.4.2 Assume that X is a zigzagged line, let

rmax ¼ max fxðkÞgandrmin ¼ min fxðkÞg:

1kn 1kn

Then
(1) For any 8n 2 ½rmin ; rmax , X ¼ n is known as the n-contour (line); and
(2) The solutions ðti ; xðti ÞÞði ¼ 1; 2;   Þ of system of equations

(
X ¼ fxðkÞ þ ðt  kÞ½xðk þ 1Þ  xðkÞjk ¼ 1; 2;    ; n  1g
X¼n

is called the n-contour points. The n-contour point is the intersection of the zig-
zagged line X and the n-contour line.

of X there is a
n-contour point, then the
Proposition 9.4.1 If on the ith segment
nxðiÞ
coordinates of this point are given by i þ xði þ 1ÞxðiÞ ; n :

Proof The equation of i-piece zigzagged line of the sequence X is as follows:

X ¼ xðiÞ þ ðti  iÞ½xði þ 1Þ  xðiÞ

From
(
X ¼ xðiÞ þ ðti  iÞ½xði þ 1Þ  xðiÞ
X¼n
220 9 Techniques for Grey Systems Forecasting

We have

n  xðiÞ
ti ¼ i þ
xði þ 1Þ  xðiÞ

Definition 9.4.3 Let Xn ¼ ðP1 ; P2 ;    ; Pm Þ be the sequence of n-contour points of

X such that point Pi is located on the ith segment. Let

n  xðti Þ
qðiÞ ¼ ti þ ; i ¼ 1; 2;    ; m
xðti þ 1Þ  xðti Þ

Then Qð0Þ ¼ ðqð1Þ; qð2Þ;    ; qðmÞÞ is known as the n-contour time moment
sequence. By establishing a GM (1, 1) model using this n-contour moment
sequence, one can produce the predicted values for future n-contour time moments:

^qðm þ 1Þ; ^qðm þ 2Þ;    ; ^qðm þ kÞ:

Definition 9.4.4 The lines X ¼ ni ði ¼ 0; 1; 2;    ; sÞ, where n0 ¼ rmin , n1 ¼ 1s

ðrmax  rmin Þ þ rmin , … ni ¼ si ðrmax  rmin Þ þ rmin , …, ns1 ¼ s1 s ðrmax 
rmin Þ þ rmin , ns ¼ rmax are known as equal time distanced contours. When taking
contour lines, one needs to make sure that the corresponding contour moments
satisfy the conditions for establishing valid GM (1, 1) models.
Definition 9.4.5 Let X ¼ ni ði ¼ 1; 2;    ; sÞ be s different contours,

ð0Þ
Qi ¼ ðqi ð1Þ; qi ð2Þ;    ; qi ðm1 Þ; i ¼ 1; 2;    ; s;

stand for the sequence of ni -contour time moments, and

qi ðmi þ 1Þ; ^qi ðmi þ 2Þ;    ; ^qi ðmi þ ki Þ; i ¼ 1; 2;    ; s;

^

the GM (1, 1) predicted ni -contour time moments. If there are i 6¼ j such that

^qi ðmi þ li Þ ¼ ^qj ðmj þ lj Þ;

then these values are known as a pair of invalid moments.

Proposition 9.4.2 Let q^i ðmi þ jÞ, j ¼ 1; 2;    ; ki , i ¼ 1; 2;    ; s, be the GM (1, 1)
predicted ni -contour time moments. After deleting all invalid predictions, order the
rest in terms of their magnitudes as follows:

^qð1Þ\^qð2Þ\. . .\^qðns Þ;
9.4 Wave Form Forecasting 221

where ns  k1 þ k2 þ    þ ks . If X ¼ n^qðkÞ is the contour line corresponding to

^qðkÞ. Then the predicted wavy curve of X ð0Þ is given below:

^ ð0Þ ¼ fn^qðkÞ þ ½t  ^qðkÞ½n^qðk þ 1Þ  n^qðkÞ jk ¼ 1; 2;    ; ns g

X¼X

9.5 System Forecasting

9.5.1 The Five-Step Modeling Process

Generally, when studying a system one should first establish a mathematical model
through which the overall functionality of the system, abilities of coordination,
incidence relations, causal relations, and dynamic relationships between different
parts can be quantitatively investigated. This kind of study has to be guided by an
early qualitative analysis, and there must be close connection between the quanti-
tative and qualitative studies. As for the development of the system’s model, one
generally goes through the following five steps: development of thoughts, analysis
of relevant factors, quantification, dynamics, and optimization. This is the so-called
five-step modeling (Deng 1985).
Step 1: Develop thoughts and form concepts. Through an initial qualitative
analysis, one clarifies his goal, possible paths and specific procedures, and
then verbally and precisely describes the desired outcomes. This is the
initial language model of the problem (see Fig. 9.3).
Step 2: Examine all the factors involved in the language model and their mutual
relationships in order to pinpoint the causes and conclusions. Then,
construct a line-drawing to depict the causal relationships (Fig. 9.3). Each
pair (or a group) of causes and effect form a link. A system might be made
up of many of such links. At the same time, a quantity can be a cause of a
link and also a consequence of another link. When several of these links
are connected, one obtains a line drawing of many links that organically
form the system of our concern (Fig. 9.4).
Step 3: Quantitatively study each causality link and obtain an approximate
quantitative relationship, which is a quantified model.

Fig. 9.3 Depicted causal relationships

222 9 Techniques for Grey Systems Forecasting

Fig. 9.4 Line drawing of an abstract system

Step 4: For each link, collect additional input-output data, on which dynamic GM
models are established. Such dynamic models are higher level quantitative
models. They can further reveal the relationships between input and
output, and their laws of transformation. They are the foundation of
systems analysis and optimization.
Step 5: Systematically investigate the established dynamic models by adjusting
their structures, mechanisms, and parameters, in order to arrive at the
purpose of optimizing the outcome and realizing the desired conclusions.
Models obtained in this way are known as optimal models.
The procedure of five-step modeling is such a holistic process that at five dif-
ferent stages five different kinds of models are established: language models, net-
work models, quantified models, dynamic models, and optimized models. In the
entire process of modeling, the conclusions of the next level should be repeatedly
fed back so that the modeling exercise itself becomes a feedback system making the
model system as perfect as possible.

9.5.2 System Models for Prediction

For a system with many mutually related factors and many autonomous controlling
variables, no single model can reflect adequately the development and change of the
system. To effectively study such a system and to predict its future behaviors, one
should consider establishing a system of models.
Definition 9.5.1 Assume that

ð0Þ ð0Þ ð0Þ ð0Þ

Xi ¼ ðxi ð1Þ; xi ð2Þ;    ; xi ðnÞÞ; i ¼ 1; 2; . . .; m;

are sequences of raw data for the state variables of a system, and

ð0Þ ð0Þ ð0Þ ð0Þ

Uj ¼ ðuj ð1Þ; uj ð2Þ;    ; uj ðnÞÞ; j ¼ 1; 2; . . .; s;
9.5 System Forecasting 223

are sequences of data of the control variables. Then the following

ð0Þ ð1Þ ð1Þ

ð1Þ ð1Þ ð1Þ
x1 ¼ a11 z1 þ a12 x2 þ    þ a1m xm þ b11 u1 þ b12 u2 þ    þ b1s usð1Þ
ð0Þ ð1Þ ð1Þ
ð1Þ ð1Þ ð1Þ
x2 ¼ a21 x1 þ a22 z2 þ    þ a2m xm þ b21 u1 þ b22 u2 þ    þ b2s usð1Þ
. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .
ð0Þ ð1Þ ð1Þð1Þ ð1Þ ð1Þ
xm ¼ am1 x1 þ am2 x2 þ    þ amm zm þ bm1 u1 þ bm2 u2 þ    þ bms usð1Þ
ð1Þ ð1Þ ð1Þ
du1 ð1Þ du2 ð1Þ dus
¼ c 1 u1 þ d1 ; ¼ c2 u2 þ d2 ; . . .; ¼ cs usð1Þ þ ds
dt dt dt

are known as system models for prediction. As a matter of fact, each system model
for prediction consists of m DGM (1, m + s) and s EGM (1, 1) models. If we write
the previous system models for prediction using the terminology of matrices, we
have

X ð0Þ ¼ AX ð1Þ þ BU ð1Þ
U ð0Þ ¼ CU ð1Þ þ D

ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ

where X ð1Þ ¼ ðx1 ; x2 ;    ; xm ÞT , U ð1Þ ¼ ðu1 ; u2 ;    ; us ÞT , A ¼ ½akl m m ,
  
B ¼ bpq m s , C ¼ diag cj s s , and D ¼ dj s 1 .
X is known as the state vector, U the control vector, A the state matrix B the
control matrix, C the development matrix, and D the grey effect vector.
Proposition 9.5.1 For the previous system models for prediction, the time response
sequences are given as follows:
ð0Þ ð1Þ ð1Þ ð1Þ ð1Þ
^xi ðkÞ ¼ ai1 xi ðkÞ þ ai2 x2 ðkÞ þ    þ aim xð1Þ ð1Þ
m ðkÞ þ bi1 u1 ðkÞ þ bi2 u2 ðkÞ þ    þ bis us ðkÞ;
i ¼ 1; 2; . . .; m

ð0Þ ð0Þ dj
uj ðkÞ ¼ ð1  ecj Þðuj ð1Þ  Þecj ðk1Þ
^ ; j ¼ 1; 2; . . .; s
cj

9.6 Practical Applications

Example 9.6.1 Let us look at a wavy curve prediction for the (synthetic) stock
index of Shanghai stock exchange. Using the stock index data of the stock index
weekly closes of Shanghai stock exchange, the time series plot from February 21,
1997, through to October 31, 1998, is shown in Fig. 9.5.
224 9 Techniques for Grey Systems Forecasting

Fig. 9.5 Shanghai stock

exchange index (Feb. 21,
1997, to Oct. 31, 1998)

Let us take

n1 ¼ 1140; n2 ¼ 1170; n3 ¼ 1200; n4 ¼ 1230; n5 ¼ 1260;

n6 ¼ 1290; n7 ¼ 1320; n8 ¼ 1350; n9 ¼ 1380:

Then the corresponding ni -contour time moment sequences are given below:
(1) For n1 ¼ 1140,
ð0Þ
Q1 ¼ fq1 ðkÞg71 ¼ ð4:4; 31:7; 34:2; 41; 42:4; 76:8; 78:3Þ
(2) For n2 ¼ 1170,
ð0Þ
Q2 ¼ fq2 ðkÞg12
2 ¼ ð5:2; 19:8; 23; 25:6; 26:9; 31:2; 34:8; 39:5; 44:6; 76; 76:2; 79:2Þ

(3) For n3 ¼ 1200,

ð0Þ
Q3 ¼ fq3 ðkÞg11
3 ¼ ð5:9; 19:5; 24:8; 25:2; 26:5; 30:3; 46:2; 53:4; 55:4; 75:5; 79:7Þ

(4) For n4 ¼ 1230,

ð0Þ
Q4 ¼ fq4 ðkÞg10
4 ¼ ð6:5; 19:2; 28:3; 29:5; 49:7; 50:8; 56:2; 76:4; 82:9; 85Þ

(5) For n5 ¼ 1260,

ð0Þ
Q5 ¼ fq5 ðkÞg75 ¼ ð7; 14:2; 16:5; 16:4; 18:8; 56:7; 75:2Þ
(6) For n6 ¼ 1290,
ð0Þ
Q6 ¼ fq6 ðkÞg56 ¼ ð8:3; 13:4; 16:9; 56:2; 74:6Þ
(7) For n7 ¼ 1320,
ð0Þ
Q7 ¼ fq7 ðkÞg67 ¼ ð8:8; 12:8; 60:2; 71:8; 72:7; 73:6Þ
9.6 Practical Applications 225

(8) For n8 ¼ 1350,

ð0Þ
Q8 ¼ fq8 ðkÞg68 ¼ ð9:6; 12:5; 61:8; 69:8; 70:9; 71:8Þ
(9) For n9 ¼ 1380,
ð0Þ
Q9 ¼ fq9 ðkÞg49 ¼ ð10:8; 12:4; 64:1; 69Þ
ð0Þ ð1Þ
Applying the 1-AGO on Qi ði ¼ 1; 2;    ; 9Þ produces Qi ði ¼ 1; 2;    ; 9Þ,
whose EGM (1, 1) response sequences are respectively given by:

ð1Þ ð1Þ
q1 ðk þ 1Þ ¼ 113:91e0:215k  109:51; ^q2 ðk þ 1Þ ¼ 98:58e0:159k  93:83;
^

ð1Þ ð1Þ
^
q3 ðk þ 1Þ ¼ 102:08e0:166k  96:18; ^q4 ðk þ 1Þ ¼ 151:66e0:160k  145:16;

ð1Þ ð1Þ
q5 ðk þ 1Þ ¼ 13e0:435k  6; ^q6 ðk þ 1Þ ¼ 21:94e0:539k  13:64;
^

ð1Þ ð1Þ
q7 ðk þ 1Þ ¼ 185:08e0:192k  176:28; ^q8 ðk þ 1Þ ¼ 193:19e0:186k  182:57;
^

ð1Þ
^q9 ðk þ 1Þ ¼ 45:22e0:490k  35:39:

ð1Þ ð1Þ
By letting ^
qi ðk þ 1Þ ¼ ^qi ðk þ 1Þ  ^qi ðkÞ, we obtain the following ni -contour
prediction sequences, i ¼ 1; 2;    ; 9;

^ ð0Þ ¼ ð^q1 ð12Þ; ^q1 ð13ÞÞ ¼ ð99:8; 127:7Þ

Q 1

^ ð0Þ ¼ ð^q2 ð13Þ; ^q2 ð14Þ; ^q2 ð15ÞÞ ¼ ð96:8; 116:7; 131:4Þ
Q 2

^ ð0Þ ¼ ð^q3 ð12Þ; ^q3 ð13Þ; ^q3 ð14ÞÞ ¼ ð95:7; 114:2; 133:8Þ
Q 3

^ ð0Þ ¼ ð^q4 ð11Þ; ^q4 ð12Þ; ^q4 ð13ÞÞ ¼ ð110:9; 134:2; 152:8Þ
Q 4

^ ð0Þ ¼ ð^q5 ð8Þ; ^q5 ð9ÞÞ ¼ ð94:2; 148:8Þ

Q 5

^ ð0Þ ¼ ð^q6 ð6ÞÞ ¼ ð135:5Þ

Q 6

^ ð0Þ ¼ ð^q7 ð7Þ; ^q7 ð8Þ; ^q7 ð9ÞÞ ¼ ð101:9; 123:4; 149:5Þ
Q 7

^ ð0Þ ¼ ð^q8 ð7Þ; ^q8 ð8Þ; ^q8 ð9ÞÞ ¼ ð105; 119:8; 144:6Þ
Q 8

^ ð0Þ ¼ ð^q9 ð5ÞÞ ¼ ð122:3Þ

Q 9
226 9 Techniques for Grey Systems Forecasting

Fig. 9.6 The predicted wavy

curve of Shanghai stock
exchange index (Nov. 1998 to
March 2000)

Based on these predictions, we construct the predicted wavy curve for the
Shanghai stock exchange index for the time period from November 1998 to the end
of 1999 (see Fig. 9.6).
Chapter 10
Grey Models for Decision-Making

10.1 Introduction

Deciding on what actions to take based on actual circumstances and pre-determined

goals is known as decision-making. The essential meaning of decision-making is to
make a decision or to choose a course of actions. Decision-making not only plays
an important part in various kinds of management activities, but also appears
throughout every person’s daily life. The concept of decision-making can be
divided into two categories: general and specific. In the general category, each
decision-making stands for an entire process of activities, including posting ques-
tions, collecting data, establishing a goal, making, analyzing, and evaluating a plan
of action, implementing the plan, feedback, and modifying the plan. In the specific
category, decision-making only represents the step of choosing a specific plan of
action out of the entire decision-making process. Also, some scholars understand
decision-making as choosing and picking a plan of action underuncertain uncertain
conditions. In this case, the choice can be most likely influenced by the decision
maker’s prior experience, attitude, and willingness to take a certain amount of risk.
Grey decision-making is about making a decision using decision models that
involve grey elements or that combine general decision model and grey systems
models. Its focus of study is on the problem of choosing a specific plan.
In this chapter, we define an event as the problem waiting to be resolved, the
eventevent needing to be handled, and the current state of a system’s behavior.
Events are where we begin our investigation.
Definition 10.1.1 Events, countermeasures, objectives, and effects are known as
the four key elements of decision-making.
Definition 10.1.2 The totality of all events within the range of a research is known as
the set set of events of events of the study, denoted A ¼ fa1 ; a2 ; . . .an g, where ai ,
i ¼ 1; 2; 3;    ; n; stands for the ith event. The totality of all possible countermeasures

© Springer Science+Business Media Singapore 2017 227

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_10
228 10 Grey Models for Decision-Making

is known as the set of countermeasures, denoted B ¼ fb1 ; b2 ;    ; bm g with bj , j ¼

1; 2; . . .m; be the jth countermeasure.


Definition 10.1.3 The Cartesian product A  B ¼ ðai ; bj Þjai 2 A;bj 2 B of the
event set A and the countermeasureset set B is known as the set of decision
schemes, written as S ¼ A  B; where each ordered pair sij ¼ ai ; bj ; for any
ai 2 A; bj 2 B; is known as a decision scheme.
For example, in the decision-making on what to plant in agriculture, weather
conditions can be used as the set of events, with a normal year denoted as a1 , a
drought year as a2 , and a flood year as a3 . Then, the set of events is

A ¼ fa1 ; a2 ; a3 g

Different strains of crops can be seen as countermeasures, with corn denoted as

b1 , Chinese sorghum as b2 , soybeans as b3 , sesame b4 , potatoes and yams as b5 ,   ;
then the countermeasure set is given as

B ¼ fb1 ; b2 ; b3 ; b4 ; b5 ;   g

Therefore, the set of decision scheme is

S ¼ A  B ¼ fs11 ; s12 ;    ; s15 ;    ; s21 ;    ; s25 ;    ;    ; s31 ;    ; s35 ;   g

where sij ¼ ðai ; bj Þ.

Here, events and countermeasures are simple. Therefore, the constructed deci-
sion schemes are relatively simple, too. In practical decision-making, events are
often complicated, consisting of many kinds of simple events, so the counter-
measures are complicated, too. Hence, the resultant decision schemes can be
extremely complicated.
Let us continue to use the previous agricultural decision-making example. The
set of events is the organic body consisting of weather, soil, irrigation, fertilizer,
agricultural chemicals, work force, and technology. The countermeasures are not
simply the individual strains of crops, but various proportional combinations of
many different strains of crops. Let us define a1 as an event characterized by a
normal year, loam, 50 % effective irrigation area, sufficient fertilizer and agricul-
tural chemicals, sufficient work force, and medium level of technology.
Additionally, let us define a2 as an event characterized by a drought year, black
earth, 50 % effective irrigation area, sufficient fertilizer and work force, lack of
agricultural chemicals, and medium level of technology. Then, we have the set of
events:

A ¼ fa1 ; a2 ; . . . g

Let us write b1 as the countermeasure including 30 % corn + 10 % Chinese sor-

ghum + 20 % soybeans + 15 % sesame + 15 % potatoes and yams + 10 % others.
10.1 Introduction 229

Also, let us write b2 as the countermeasure including 10 % corn + 20 % Chinese

sorghum + 30 % soybeans + 30 % sesame + 10 % others. Then, we have the
countermeasure set:

B ¼ fb1 ; b2 ; . . . g

Now, the decision scheme s11 ¼ ða1 ; b1 Þ is that, under the conditions of a normal
year, loam, 50 % effective irrigation area, with sufficient fertilizer and agricultural
chemicals, sufficient workforce, and medium level of technology, we should plant
30 % corn, 10 % Chinese sorghum + 20 % soybeans + 15 % sesame + 15 %
potatoes and yams + 10 % others.
Let us look at the example of teaching scheduling. The collection of all course
offerings of a fixed semester at a certain school can be seen as the set of events; all
teaching faculty of this school, and various teaching methods, such as laboratory,
interns, and multimedia, are seen as the set of countermeasures. Based on the
circumstances, one teacher can teach several courses, or several teachers teach one
course together. The work load could be 100 % teaching, or 60 % teaching, 20 %
laboratory, 10 % interns, and 10 % multimedia and others.
For a given decision scheme sij 2 S; evaluating the effects under a set of
pre-determined objectives and deciding on what to take and what to let go based on
evaluation is the decision-making we discuss in this chapter. In the following
sections, we will study several different kinds of grey decision-making methods.

10.2 Grey Target Decisions


Definition 10.2.1 Let S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg be a a set of decision
ðk Þ
schemes, uij the effect value of decision scheme sij with respect to objective k; and
ðk Þ ðk Þ
R the set of all real numbers. Then uij : S 7! R; defined by sij 7! uij , is known as the
effect mapping of S with respect to object k.
ðk Þ ðk Þ
Definition 10.2.2 If uij ¼ uih , then we say that the countermeasures bj and bh of
event ai are equivalent with respect to objective k, written as bj ffi bh ; and the set

ðk Þ
Bi ¼ fbjb 2 B; b ffi bh g

is known as the effect equivalence class of countermeasure bh of event ai with

respect to objective k.
Definition 10.2.3 If k is such an objective that the greater the effect value is the
ðk Þ ðk Þ
better, and uij [ uih , then we say that the countermeasure bj is superior to bh in
ðk Þ
terms of event ai with respect to objective k, written as bj  bh . The set Bih ¼
230 10 Grey Models for Decision-Making

fbjb 2 B; b  bh g is known as the superior set of countermeasure bh of event ai

with respect to objective k.
Similarly, we can define the concept of superior classes of countermeasures for
situations where the closer to a fixed moderate value the effect value is the better,
and where the smaller the effect value is the better.
ðk Þ ðk Þ
Definition 10.2.4 If uij ¼ uih , then events ai and aj are said to be equivalent in
terms of the countermeasure bh with respect to objective k, written ai ffi aj . The set

ðk Þ
Ajh ¼ faja 2 A; a ffi ai g

is known as the effect equivalence class of events of the countermeasure bh with

respect to objective k.
Definition 10.2.5 If k is such an objective that the greater the effect value is the
ðk Þ ðk Þ
better, and uih [ ujh , then we say that event ai is superior to event aj in terms of
countermeasure bh with respect to objective k, denoted ai  aj . The set

 
Ajh ¼ aa 2 A; a  aj
ðk Þ

is known as the superior class of event aj in terms of countermeasure bh with

respect to objective k.
Similarly, the concept of superior classes can be defined for situations where the
closer to a fixed moderate value the effect value is the better, and where the smaller
the effect value is the better.
ðk Þ ðk Þ
Definition 10.2.6 If uij ¼ uhl , then scheme sij is equivalent to scheme shl under
objective k, denoted sij ffi shl . The set

SðkÞ ¼ fsjs 2 S; s ffi shl g

is known as the effect equivalence class of scheme shl under objective k.

Definition 10.2.7 If k is such an objective that the greater the effect value is the
ðk Þ ðk Þ
better, and uij [ uhl , then scheme sij is said to be superior to scheme shl under
objective k, denoted sij  shl . The set

ðk Þ
Shl ¼ fsjs 2 S; s  shl g

is known as the effect superior class of scheme shl under objective k.

Similarly, the concept of superior classes for scheme effects can be defined for
scenarios where the closer to a fixed moderate value the effect value of a scheme is
the better, and where the smaller the effect value of the scheme is the better.
10.2 Grey Target Decisions 231

Proposition 10.2.1 Assume that S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg 6¼ ∅ and
n  o

U ðkÞ ¼ uij ai 2 A; bj 2 B is the set of effects under objective k, and SðkÞ the
ðk Þ

set of effect equivalence classes of schemes under objective k. Then the mapping

 ðkÞ
uðkÞ : SðkÞ ! U ðkÞ , defined by SðkÞ 7! uij , is bijective.
ðk Þ ðk Þ
Definition 10.2.8 Let d1 and d2 be the upper and lower threshold values of the
n  o
 ðk Þ ðk Þ
decision effects of sij under objective k. Then S ¼ r d1  r  d2
1
is known as
ðk Þ ðk Þ ðk Þ
the one-dimensional grey target of objective k, uij 2 ½d1 ; d2  a satisfactory effect
under objective k, the corresponding sij a desirable scheme with respect to objective
k, and bj a desirable countermeasure of event ai with respect to objective k.
ðk Þ
Proposition 10.2.2 Assume that uij stands for the effect value of scheme sij with
ðk Þ
respect objective k. If uij 2 S1 , that is, sij is a desirable desirable scheme with
ðk Þ
respect to objective k. Then for any s 2 Sij , s is also a desirable scheme. That is,
when sij is desirable, all schemes in its effect superior class are desirable.
The discussion above applies to cases involving a single objective. Nevertheless,
grey targets of decision-making with multi-objectives can also be addressed.
ð1Þ ð1Þ
Definition 10.2.9 Assume that d1 and d2 are the threshold values of decision
ð2Þ ð2Þ
effects of objective 1, d1 and d2 the threshold values of decision effects of
objective 2. Then
n  o
 ð1Þ ð1Þ ð2Þ ð2Þ
S2 ¼ ðr ð1Þ ; r ð2Þ Þd1  r ð1Þ  d2 ; d1  r ð2Þ  d2

is known as a grey target ofntwo-dimensional

o decision-making. If the effect vector
ð1Þ ð2Þ
of scheme sij satisfies uij ¼ uij ; uij 2 S2 , then sij is seen as a desirable scheme
with respect to objectives 1 and 2, and bj a desirable countermeasure for event ai
with respect to objectives 1 and 2.
ð1Þ ð1Þ ð2Þ ð2Þ ðsÞ ðsÞ
Definition 10.2.10 Assume that d1 ; d2 ; d1 ; d2 ;    ; d1 ; d2 are respectively
the threshold values of decision effects under objectives 1, 2, …, s. Then the
following region of the s-dimensional Euclidean space
n  o
 ð1Þ ð1Þ ð2Þ ð2Þ ðsÞ ðsÞ
Ss ¼ ðr ð1Þ ; r ð2Þ ;    ; r ðsÞ Þd1  r ð1Þ  d2 ; d1  r ð2Þ  d2 ;    ; d1  r ðsÞ  d2

is known as a grey target of an s–dimensional decision-making. If the effect vector

of scheme sij satisfies
232 10 Grey Models for Decision-Making

ð1Þ ð2Þ ðsÞ

uij ¼ ðuij ; uij ;    ; uij Þ 2 Ss

ðk Þ
where uij stands for the effect value of the scheme sij with respect to objective k,
k ¼ 1; 2;    ; s; then sij is known as a desirable scheme with respect to objectives
1, 2, …, s, and bj a desirable countermeasure of event ai with respect to objectives
1, 2, …, s.
Intuitively, the grey targets of a decision-making essentially represent the
location of satisfactory effects in terms of relative optimization. In many practical
circumstances, it is impossible to obtain the absolute optimization so that people are
happy if they can achieve a satisfactory outcome. Of course, based on the need, one
can gradually shrink the grey targets of his decision-making to a single point in
order to obtain the ultimate optimal effect, where the corresponding scheme is the
most desirable, and the corresponding countermeasure the optimal countermeasure.
Definition 10.2.11 The following equation
n  o
 ð1Þ ð2Þ ðsÞ
Rs ¼ ðr ð1Þ ; r ð2Þ ;    ; r ðsÞ Þðr ð1Þ  r0 Þ2 þ ðr ð2Þ  r0 Þ2 þ    þ ðr ðsÞ  r0 Þ2  R2

is known as an s-dimensional spherical grey target centered at r0 ¼

ð1Þ ð2Þ ðsÞ ð1Þ ð2Þ ðsÞ
ðr0 ; r0 ;    ; r0 Þ with radius R. The vector r0 ¼ ðr0 ; r0 ;    ; r0 Þ is seen as the
optimum effect vector.
ð1Þ ð2Þ ðsÞ
For r1 ¼ ðr1 ; r1 ;    ; r1 Þ 2 R;
h i1=2
ð1Þ ð1Þ ð2Þ ð2Þ ðsÞ ðsÞ
jr1  r0 j ¼ ðr1  r0 Þ2 þ ðr1  r0 Þ2 þ    þ ðr1  r0 Þ2

is known as the bull’s-eye distance of vector r1 . The values of this distance reflect
the superiority of the corresponding decision effect vectors.
ð1Þ
Definition 10.2.12 Let sij and shl be two different schemes, and uij ¼ ðuij ;
ð2Þ ðsÞ ð1Þ ð2Þ ðsÞ
uij ;    ; uij Þ and uhl ¼ ðuhl ; uhl ;    ; uhl Þ their effect vectors, respectively. If
 
uij  r0  juhl  r0 j ð10:1Þ

then scheme shl is said to be superior to sij , denoted shl  sij . When the equal sign in
Eq. (10.1) holds true, schemes sij and shl are said to be equivalent, written shl ffi sij .
If for i ¼ 1; 2;    ; n and j ¼ 1; 2;    ; m; uij 6¼ r0 always holds true, then the
optimum scheme does not exist, and the event does not have any optimum coun-
termeasure If the optimum scheme does not exist, however, there  are h and l such
that for any i ¼ 1; 2;    ; n and j ¼ 1; 2;    ; m; juhl  r0 j  uij  r0  holds true,
that is, for any sij 2 S; shl  sij holds, then shl is known as a quasi-optimum scheme,
ah a quasi-optimum event and a quasi-optimum countermeasure.
10.2 Grey Target Decisions 233

Theorem 10.2.1 Let S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg be a set of schemes, and
n  o
 ð1Þ ð2Þ ðsÞ
Rs ¼ ðr ð1Þ ; r ð2Þ ;    ; r ðsÞ Þðr ð1Þ  r0 Þ2 þ ðr ð2Þ  r0 Þ2 þ    þ ðr ðsÞ  r0 Þ2  R2

an s-dimensional spherical grey target The S becomes an ordered set with “su-
periority” as its order relation
.
Theorem 10.2.2 There must be quasi-optimum scheme in the set of decision
schemes of ðS; Þ:
Proof This is a restatement of Zorn’s Lemma in set theory.
Example 10.2.1 Consider event a1 of reconstructing an old building. There are
three possibilities: b1 = renovate the building completely; b2 = tear down the
building and reconstruct another; and b3 = simply maintain what the building is by
fixing up minor problems. Let us make a grey target decision using three objectives:
cost, functionality, and construction speed.
Solution: Let us denote the cost as objective 1, the functionality as objective 2, and
the construction speed as objective 3. Then, we have the following three decision
schemes:
s11 ¼ ða1 ; b1 Þ ¼ (reconstruction, renovation),
s12 ¼ ða1 ; b2 Þ ¼ (reconstruction, new building), and
s13 ¼ ða1 ; b3 Þ ¼ (reconstruction, maintenance).
Evidently, different decision schemes with respect to different objectives have
different effects; and the standards for measuring the effects are also accordingly
different. For instance, regarding cost, the lesser the better; for functionality, the
higher the better; and for speed, the faster the better. Let us divide the effects of the
decision schemes into three classes: good, okay, and poor.
The effect vectors of the decision schemes are respectively defined as follows:
 
ð1Þ ð1Þ ð3Þ
u11 ¼ u11 ; u11 ; u11 ¼ ð2; 2; 2Þ;
 
ð1Þ ð2Þ ð3Þ
u12 ¼ u12 ; u12 ; u12 ¼ ð3; 1; 3Þ; and
 
ð1Þ ð2Þ ð3Þ
u13 ¼ u13 ; u13 ; u13 ¼ ð1; 3; 1Þ:

Let the bull’s eye be located at r0 ¼ ð1; 1; 1Þ and compute the the bull’s-eye
distances
234 10 Grey Models for Decision-Making

 2  2  2
1=2
ð1 Þ ð1Þ ð2Þ ð2 Þ ð3Þ ð3Þ
ju11  r0 j ¼ u11  r0 þ u11  r0 þ u11  r0
h i1=2
¼ ð2  1Þ2 þ ð2  1Þ2 þ ð2  1Þ2 ¼ 1:73
 2  2  2
1=2
ð1 Þ ð1Þ ð2Þ ð2 Þ ð2Þ ð3Þ
ju12  r0 j ¼ u12  r0 þ u12  r0 þ u12  r0
h i1=2
¼ ð3  1Þ2 þ ð1  1Þ2 þ ð3  1Þ2 ¼ 2:83
 2  2  2
1=2
ð1 Þ ð1Þ ð2Þ ð2 Þ ð3Þ ð3Þ
ju13  r0 j ¼ u13  r0 þ u13  r0 þ u13  r0
h i1=2
¼ ð1  1Þ2 þ ð3  1Þ2 þ ð1  1Þ2 ¼2

where ju11  r0 j is the smallest. So, the effect vector u11 ¼ ð2; 2; 2Þ of the decision
scheme s11 enters the grey target. Hence, renovation is a satisfactory decision.

10.3 Other Approaches to Grey Decision

10.3.1 Grey Incidence Decision

The bull’s-eye distance between a decision effect vector and the center of the target
measures the superiority of the scheme in comparison with other schemes. At the
same time, the degree of incidence between the effect vector of a decision scheme
and the optimum effect vector can be seen as another way to evaluate the superiority
of a decision scheme.

Definition 10.3.1 Let S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg be a set of decision
ð1Þ ð2Þ ðsÞ
schemes, and ui0 j0 ¼ fui0 j0 ; ui0 j0 ;    ; ui0 j0 g the optimum effect vector. If the decision
scheme corresponding to ui0 j0 satisfies ui0 j0 62 S; then ui0 j0 is known as an imagined
optimum effect vector, and si0 j0 the imagined optimum scheme.
Proposition 10.3.1 Let S be the same as above and the effect value of scheme sij is
ð1Þ ð2Þ ðsÞ
uij = fuij ; uij ;    ; uij g, for i ¼ 1; 2;    ; n; j ¼ 1; 2;    ; m:
(1) When k is an objective such n that o the greater its effect value is the better, let
ðk Þ ðk Þ
ui0 j0 ¼ max1  i  n;1  j  m uij ;
(2) When k is an objective such that the closer to a fixed moderate value u0 its
ðk Þ
effect value is the better, let ui0 j0 ¼ u0 ; and
(3) When k is an objective such
n o that the smaller itsn effect value is theobetter, let
ðk Þ ðk Þ ð1Þ ð2Þ ðsÞ
ui0 j0 ¼ min1  i  n;1  j  m uij ; then ui0 j0 ¼ ui0 j0 ; ui0 j0 ;    ; ui0 j0 is the
imagined optimum effect vector.
10.3 Other Approaches to Grey Decision 235

Proposition 10.3.2 Assume the same as in Proposition 10.3.1 and let ui0 j0 ¼
n o
ð1Þ ð2Þ ðsÞ
ui0 j0 ; ui0 j0 ;    ; ui0 j0 be the imagined optimum effect vector, eij the absolute
degree of grey incidence between uij and ui0 j0 , for i ¼ 1; 2;    ; n; j ¼ 1; 2;    ; m: If
for any i 2 f1; 2;    ; ng and j 2 f1; 2;    ; mg satisfying i 6¼ i1 and j 6¼ j1 , ei1 j1 eij
always holds true, then ui1 j1 is a quasi-optimum effect vector and si1 j1
a quasi-optimum decision scheme.
Grey incidence decisions can be made by following the following steps:
Step 1: Determine the set of events A ¼ fa1 ; a2 ;    ; an g and the set of
countermeasures B ¼ fb1 ; b2 ;    ; bm g: And then construct the set of
decision schemes S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg:
Step 2: Choose the objectives 1, 2, …, s, for the decision-making.
Step 3: Compute the effect values uðkÞ of the individual decision scheme sij , i ¼
ij
1; 2;    ; n; j ¼ 1; 2;    ; m; with respect to objective k, obtained in the
decision effect sequence uðkÞ

ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ

uðkÞ ¼ ðu11 ; u12 ;    u1m ; u21 ; u22 ;    u2m ;    ; un1 ; un2 ;    unm
ðkÞ
Þ; k
¼ 1; 2;    ; s:

Step 4: Compute the average image of the decision effect sequence uðkÞ with
respect to objective k, which is still written the same as

ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ

uðkÞ ¼ ðu11 ; u12 ;    u1m ; u21 ; u22 ;    u2m ;    ; un1 ; un2 ;    unm
ðkÞ
Þ; k
¼ 1; 2;    ; s

Step 5: Basedn on the results o of Step 4, write out the effect vector
ð1Þ ð2Þ ðsÞ
uij ¼ uij ; uij ;    ; uij of decision scheme sij , for i ¼ 1; 2;    ; n;
j ¼ 1; 2;    ; m: n
Step 6: Compute the imagined optimum effect vector u ¼ uð1Þ ; uð2Þ ;    ;
i0 j0 i0 j0 i0 j0
ðsÞ
ui0 j0 g:
Step 7: Calculate the absolute degree eij between uij and ui0 j0 , i ¼ 1; 2;    ; n;
j ¼ 1; 2;    ; m:

Step 8: From max1  i  n;1  j  m eij ¼ ei1 j1 , the quasi-optimum effect vector ui1 j1
and the quasi-optimum decision scheme si1 j1 are obtained.
Example 10.3.1 Let us look at grey incidence decision-making regarding the
evaluation of looms.
236 10 Grey Models for Decision-Making

Table 10.1 Objective values for the looms

Model Projectile loom Air jet loom Rapier loom
Weft-insertion rate (m/min) 1000 1200 800
Efficiency (%) 92 90 92
Total investment (10 K US$) 880 336 612
Total energy consumption (W/a) 374 924 816
Total land needed (m2) 1760 1092 2124
Total manpower (person) 18 22 24
Quantity of weft yarn waste (cm/weft) 5 6 10
Cost of parts (10 K ¥/a) 37 35 75
Noise (dB) 85 91 91
Quality Best Good Fine
Adaptability good Better Best

Solution: Let us denote the event of evaluating loom models by a1 . Then the event
set is A ¼ fa1 g: There are three loom models under consideration: Model 1: pur-
chase projectile loom, which is treated as countermeasure b1 ; Model 2: select air jet
loom, which is treated as countermeasure b2 ; Model 3: choose rapier loom, which is
treated as countermeasure b3 . Thus, the set of countermeasure

is B ¼ fb1 ; b2 ; b3 g;
and the set of decision schemes is S ¼ sij ¼ ðai ; bj Þai 2 A; bj 2 B ¼
fs11 ; s12 ; s13 g:
Now, let us determine the objectives. According to the functionality of looms,
eleven objectives are chosen. The weft-insertion rate (m/min) of the looms is
objective 1. The efficiency of the looms is objective 2. The total investment (in ten
thousand US$) on the looms is objective 3. The total energy cost (W/a) is objective
4. The total area (m2) of the land to be occupied by the looms is objective 5. The
total manpower (person) is objective 6. The quantity of weft yarn waste (cm/weft) is
objective 7. The cost of replacement parts (ten thousand Yuan/a) is objective 8.
Noise (dB) is objective 9. The quality of the produced fabric is objective 10. And,
the adaptability of the type of loom is objective 11.
Under the assumptions that the above-mentioned three loom models produce the
same kind of grey fabric meeting the same set of requirements, and that these looms
will produce the same amount of annual output, let us conduct the associated
computations for the said loom models. Our quantitative calculations lead to rel-
evant values for the objectives, some of which determined from the literature and
field investigations (see Table 10.1).
In the following equations, we compute decision effect sequences U k ðk ¼
1; 2;    ; 11Þ with respect to the objectives.
10.3 Other Approaches to Grey Decision 237

ð1Þ ð1Þ ð1Þ

For objective 1, we have U ð1Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð1000; 1200; 800Þ:
ð2Þ ð2Þ ð2Þ
For objective 2, we have U ð2Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð92; 90; 92Þ:
ð3Þ ð3Þ ð3Þ
For objective 3, we have U ð3Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð880; 336; 612Þ:
ð4Þ ð4Þ ð4Þ
For objective 4, we have U ð4Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð374; 924; 816Þ:
ð5Þ ð5Þ ð5Þ
For objective 5, we have U ð5Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð1760; 1092; 2124Þ:
ð6Þ ð6Þ ð6Þ
For objective 6, we have U ð6Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð18; 22; 24Þ:
ð7Þ ð7Þ ð7Þ
For objective 7, we have U ð7Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð5; 6; 10Þ:
ð8Þ ð8Þ ð8Þ
For objective 8, we have U ð8Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð37; 35; 75Þ:
ð9Þ ð9Þ ð9Þ
For objective 9, we have U ð9Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð85; 91; 91Þ:
ð10Þ ð10Þ ð10Þ
For objective 10, we have U ð10Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ðbest, good,fine):
ð11Þ ð11Þ ð11Þ
For objective 11, we have U ð11Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ðgood, better,best):
Quantify the last two qualitative objectives as follows:

ð10Þ ð10Þ ð10Þ

U ð10Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð9; 8; 7Þ
ð11Þ ð11Þ ð11Þ
U ð11Þ ¼ ðu11 ; u12 ; u13 Þ ¼ ð8; 7; 9Þ

We now compute the average images of the decision effect sequences for each of
the objectives:

U ð1Þ ¼ ð1; 1:2; 0:8Þ; U ð2Þ ¼ ð1:01; 0:98; 1:01Þ; U ð3Þ ¼ ð1:44; 0:55; 1:01Þ
U ð4Þ ¼ ð0:53; 1:31; 1:16Þ; U ð5Þ ¼ ð1:06; 0:66; 1:28Þ; U ð6Þ ¼ ð0:84; 1:03; 1:13Þ
U ð7Þ ¼ ð0:71; 0:86; 1:43Þ; U ð8Þ ¼ ð0:76; 0:71; 1:53Þ; U ð9Þ ¼ ð0:96; 1:02; 1:02Þ
U ð10Þ ¼ ð1:13; 1; 0:87Þ; and U ð11Þ ¼ ð1; 0:87; 1:13Þ

We also compute the effect vectors Uij of decision schemes sij , i ¼ 1; j ¼ 1; 2; 3:

ð1Þ ð2Þ ð11Þ

U11 ¼ ðu11 ; u11 ;    ; u11 Þ ¼ ð1; 1:01; 1:44; 0:53; 1:06; 0:84; 0:71; 0:76; 0:96; 1:13; 1Þ;
ð1Þ ð2Þ ð11Þ
U12 ¼ ðu12 ; u12 ;    ; u12 Þ ¼ ð1:2; 0:98; 0:55; 1:31; 0:66; 1:03; 0:86; 0:71; 1:02; 1; 0:87Þ; and
ð1Þ ð2Þ ð11Þ
U13 ¼ ðu13 ; u13 ;    ; u13 Þ ¼ ð0:8; 1:01; 1:01; 1:16; 1:28; 1:13; 1:43; 1:53; 1:02; 0:87; 1:13Þ

According to the principle of constituting optimum reference sequences, from

the average images of the decision effect sequences of the objectives, it follows that:
n o
ð1Þ ð1Þ
For objective 1, the greater the effect value is the better, so Ui0 j0 ¼ max uij ¼
ð1Þ
u12 ¼ 1:2;
n o
ð2Þ ð2Þ
For objective 1, the higher the effect value is the better, so Ui0 j0 ¼ max uij ¼
ð2Þ
u11 ¼ 1:01;
238 10 Grey Models for Decision-Making
n o
ð3Þ ð3Þ
For objective 3, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð3Þ
u12 ¼ 0:55;
n o
ð4Þ ð4Þ
For objective 4, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð4Þ
u11 ¼ 0:53;
n o
ð5Þ ð5Þ
For objective 5, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð5Þ
u11 ¼ 0:66;
n o
ð6Þ ð6Þ
For objective 6, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð6Þ
u11 ¼ 0:84;
n o
ð7Þ ð7Þ
For objective 7, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð7Þ
u11 ¼ 0:71;
n o
ð8Þ ð8Þ
For objective 8, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð8Þ
u12 ¼ 0:71;
n o
ð9Þ ð9Þ
For objective 9, the smaller effect value is the better, so Ui0 j0 ¼ min uij ¼
ð9Þ
u11 ¼ 0:96;
n o
ð10Þ ð10Þ
For objective 10, the higher effect value is the better, so Ui0 j0 ¼ max uij ¼
ð10Þ
u11 ¼ 1:13; and
n o
ð11Þ ð11Þ
For objective 11, the higher effect value is the better, so Ui0 j0 ¼ max uij ¼
ð11Þ
u13 ¼ 1:13:
That is, we obtain the following optimum reference sequence:

ð1Þ ð2Þ ð11Þ

Ui0 j0 ¼ ðui0 j0 ; ui0 j0 ;    ; ui0 j0 Þ
¼ ð1:2; 1:01; 0:55; 0:53; 0:66; 0:84; 0:71; 0:71; 0:96; 1:13; 1:13Þ

From uij and ui0 j0 , we compute the absolute degrees of grey incidence:

e11 ¼ 0:628; e12 ¼ 0:891; e13 ¼ 0:532

From

 the definition of grey incidence decision-making, it follows that because
max eij ¼ e12 ¼ 0:891; U12 is the quasi-optimum vector and s12 the
quasi-optimum decision scheme That is to say, in terms of producing general grey
fabric, the air jet loom is the best choice among the available loom models.
10.3 Other Approaches to Grey Decision 239

10.3.2 Grey Development Decision

Grey development decision-making is done based on the development tendency or

the future behaviors of the decision scheme of concern. It does not necessarily place
specific emphasis on the current effect of the scheme. Instead it focuses more on the
change of the decision effect over time. This method of decision-making can be and
has been employed for long-term planning as well as the decision-making of large
scale engineering projects and urban planning. It looks at problems from the angle
of development while attempting to make feasible arrangements and avoiding
repetitious constructions so that great savings of capital and manpower can be
achieved. What we have discussed earlier are static decision schemes with a fixed
time moment. Because we now involve the concept of time, as time moves, con-
stantly changing decision effects are considered.
Definition 10.3.2 Assume that A ¼ fa1 ; a2 ;    ; an g is a set of  events, B ¼
fb1 ; b2 ;    ; bm g a set of countermeasures, and S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg
the set of decision schemes. Then,

ðkÞ ðkÞ ðkÞ ðkÞ

uij ¼ ðuij ð1Þ; uij ð2Þ;    ; uij ðhÞÞ

is known as the decision effect time series of scheme sij with respect to objective k.
Definition 10.3.3 Let the decision effect time series of the scheme sij with respect
to objective k be

ðkÞ ðkÞ ðkÞ ðkÞ

uij ¼ ðuij ð1Þ; uij ð2Þ;    ; uij ðhÞÞ
h iT
ðkÞ ðkÞ ðkÞ
^
aij ¼ aij ; bij the least squares estimate of the parameters of the EGM (1, 1)
ðkÞ
model of uij . Then the inverse accumulation restoration of the EGM (1, 1) time
ðkÞ
response of uij is given by
" #
h i ðkÞ
bij
ðkÞ ðkÞ ðkÞ ðkÞ
^
uij ðl þ 1Þ ¼ 1 expðaij Þ  uij ð1Þ  ðkÞ
expðaij  lÞ
aij

Assume that the restored sequence through inverse accumulation of the EGM (1, 1)
time response of the decision effect time series of the scheme sij with respect to
objective k is
" #
h i ðkÞ
bij
ðkÞ ðkÞ ðkÞ ðkÞ
^
uij ðl þ 1Þ ¼ 1 expðaij Þ  uij ð1Þ  ðkÞ
expðaij  lÞ
aij
240 10 Grey Models for Decision-Making

When objective k satisfies that the effect value is the better, if

n o
ðkÞ ðkÞ
(1) max1  i  n;1  j  m aij ¼ ai0 j0 , then si0 j0 is known as the optimum
scheme of development n coefficients
o with respect to objective k;
ðkÞ ðkÞ
(2) max1  i  n;1  j  m ^uij ðh þ lÞ ¼ ^ui0 j0 ðh þ lÞ; then si0 j0 is known as the opti-
mum scheme of predictions with respect to objective k.
Similarly, the concepts of optimum schemes of development coefficients and
predictions can be defined for cases of objectives satisfying that the smaller the
effect value is the better, and that the closer to a moderate value the effect value is
the better, respectively. In particular, for objectives satisfying that the smaller the
effect value is the better, one only needs to replace “max” in the items (1) and
(2) above by “min”; if k is an objective satisfying that the closer to a fixed moderate
value the effect value is the better, one can determine the moderate value of the
development coefficients or predicted values at first; then define the optimum
scheme based on the distances of the development coefficients or predicted values
to the moderate value.
In practical applications, one may face the scenarios that either both the optimum
scheme of development coefficients and predictions are the same, or that they are
different. Even so, the following theorem tells us that eventually these optimum
schemes would converge into one.
Theorem 10.3.1 Assume that k is such an objective that the greater its effect value
is the better, si0 j0 is the optimum scheme of development coefficients, that is,
n o
ðkÞ ðkÞ ðkÞ
ai0 j0 ¼ max1  i  n;1  j  m aij ; and ^ui0 j0 ðh þ l þ 1Þ is the predicted value for
the decision effect of si0 j0 . Then there must be l0 [ 0 such that
n o
ðkÞ ðkÞ
^
ui0 j0 ðh þ l0 þ 1Þ ¼ max ^uij ðh þ l0 þ 1Þ
1  i  n;1  j  m

That is, in a sufficiently distant future, si0 j0 will also be the optimum scheme of
predictions.
Proof See Liu and Lin (2006, pp. 340–341) for details.
Similar results hold true for those objectives satisfying either that the smaller the
effect value is the better or that the closer to a fixed moderate value the effect value
is the better.
At this junction, careful readers might have noticed that Theorem 10.3.1 does
not state the case that there are some increasing and decreasing sequences among
decision effect time series at the same time. As a matter of fact, for objectives
satisfying that the greater the effect value is the better, there is no need to consider
decreasing decision effect time series. For objectives satisfying that the smaller the
effect value is the better, all increasing decision effect time series are deleted in
advance in all discussions. As for objectives satisfying that the closer to a moderate
value the effect value is the better, one can consider only either increasing or
decreasing decision effect time series depending on the circumstances involved.
10.3 Other Approaches to Grey Decision 241

10.3.3 Grey Clustering Decision

Grey cluster decision is useful for synthetic evaluations of objects with respect to
several different criteria so that decisions can be made about whether or not an
object meets the given standards for inclusion in or exclusion from a set. This
method has often been employed for classification decision-making regarding
objects or people. For instance, school students can be classified based on their
individual capabilities to receive information, to comprehend what is provided, and
to grow so that different teaching methods can be applied and different students can
be enrolled in different programs. As a second example, based on different sets of
criteria, comprehensive evaluations can be done for general employees, technicians,
and administrators respectively so that decisions can be made regarding who is
qualified for his/her job, who is ready for a promotion, and so on.
Definition 10.3.4 Assume that there are n objects to make decisions on, m criteria,
s different grey classes, the quantified evaluation value of object i with respect to
criterion j is xij , fjk ðÞ are the possibility functions of the kth grey class with respect
to the jth criterion, and wj is the synthetic decision-making weight of criterion j such
P
that m j¼1 wj ¼ 1; i ¼ 1; 2; …, n, j = 1, 2, …,m, k ¼ 1; 2;    ; s: Then

X
m
rki ¼ fjk ðxij Þwj
j¼1

is known as the decision coefficient for the object i to belong to grey class k;
ri ¼ ðr1i ; r2i ;    ; rsi Þ is known as the decision coefficient vector of object i, i ¼
P

1; 2;    ; n; and ¼ ðrki Þns the decision coefficient matrix. If max1  k  s rki ¼

rki , then the decision is that the object i belongs to grey class k .
In practical applications, it is quite often the case that many objects belong to the
same decision grey class at the same time, while there is a constraint on how many
objects are allowed in the grey class. When this occurs, we can further determine
individual objects’ precedence in grey class k on the basis of the size of integrate
clustering coefficients.

10.4 Multi-attribute Intelligent Grey Target Decision

Model

In this section, we will study a new decision model which is constructed on the
basis of four new functions of uniform effect measures (Liu et al. 2014). This new
decision model sufficiently considers the two different scenarios of whether or not
the effect values of the objectives actually hit the targets with very clear physics
significance. First, a grey target is defined as a satisfying region, which a decision
242 10 Grey Models for Decision-Making

maker wants to reach, with an inside ideal point across multiple objectives. To
facilitate the uniform distance measure of a decision strategy to the pre-defined grey
target, four kinds of measure procedures are designed including the effect measures
for benefit-type objectives and cost-type objectives, the lower effect measure for
moderate-type objectives, and the upper effect measure for moderate-type according
to three types of decision objective including benefit objective, cost objective, and
non-monotonic objective with a most preferred middle value. Then, a matrix of
synthetic effect measures can be easily obtained based on the uniform distance
measure of a decision strategy to the grey target over different objectives. Based
upon the obtained matrix information, different decision strategies can be evaluated
easily and comprehensively The proposed method has a clear physical meaning as
missing target, hitting target as well as hitting performance.

10.4.1 The Uniform Effect Measure

Definition 10.4.1 (1) Let k be a benefit type objective, that is, for k the larger the
ðk Þ
effect value is the better, and the decision grey target of objective k is uij 2
n o
ðk Þ ðkÞ ðkÞ
½ui0 j0 ; maxi maxj uij ; that is, ui0 j0 stands for the threshold effect value of
objective k: Then

ðkÞ ðkÞ
ðkÞ uij  ui0 j0
rij ¼ n o ð10:2Þ
ðkÞ ðkÞ
max max uij  ui0 j0
i j

is referred to as the effect measure of a benefit-type objective.

(2) Let k be a cost-type objective, that is, for k the smaller the effectnvalue
o is the
ðk Þ ðkÞ ðk Þ
better, and the decision grey target of objective k is uij 2 ½min min uij ; ui0 j0 ;
i j
ðkÞ
that is, ui0 j0 stands for the threshold effect value of objective k: Then

ðkÞ ðkÞ
ðkÞ ui0 j0  uij
rij ¼ n o ð10:3Þ
ðkÞ ðkÞ
ui0 j0  min min uij
i j

is referred to as the lower effect measure of moderate-value type objective.

(3) Let k be a moderate-value type objective, that is, for k the closer to a
moderate value A the effect value is the better, and the decision grey target of
ðk Þ ðk Þ ðk Þ ðkÞ ðkÞ
objective k is uij 2 ½A  ui0 j0 ; A þ ui0 j0 ; that is, both A  ui0 j0 and A + ui0 j0 are
respectively the lower and upper threshold effect values of objective k: Then,
10.4 Multi-attribute Intelligent Grey Target Decision Model 243

ðkÞ ðkÞ
(i) When uij 2 ½A  ui0 j0 ; A;

ðkÞ ðkÞ
ðkÞ uij  A þ ui0 j0
rij ¼ ðkÞ
ð10:4Þ
ui0 j0

is referred to as the lower effect measure of moderate-value type objective.

ðk Þ ðk Þ
(ii) When uij 2 ½A; A þ ui0 j0 ;

ðkÞ ðkÞ
ðkÞ A þ ui0 j0  uij
rij ¼ ðkÞ
ð10:5Þ
ui0 j0

is referred to as the upper effect measure of moderate-value type objective.

The effect measures of benefit-type objectives reflect the degrees of both how
close the effect sample values are to the maximum sample values and how far away
they are from the threshold effect values of the objectives. Similarly, the effect
measures of cost-type objectives represent how close the effect sample values are to
the minimum effect sample values and how far away the effect sample values are
from the threshold effect values of the objectives; the lower effect measures of
moderate-value type objectives indicate how far away the effect sample values that
are smaller than the moderate value A are from the lower threshold effect value, and
the upper effect measures indicate how far away the effect sample values that are
greater than the moderate value A are from the upper threshold effect values of the
objectives.
For situations of missing targets, there are the following four different
possibilities:
(1) The effect value of a benefit-type objective is smaller than the threshold value
ðkÞ ðk Þ ðk Þ
ui0 j0 , that is, uij \ui0 j0 ;
(2) The effect value of a cost-type objective is greater than the threshold value
ðkÞ ðk Þ ðk Þ
ui0 j0 , that is, uij [ ui0 j0 ;
(3) The effect value of a moderate-value type objective is smaller than the lower
ðkÞ ðk Þ ðk Þ
threshold effect value A  ui0 j0 , that is, uij \A  ui0 j0 ; and
(4) The effect value of a moderate-value type objective is greater than the upper
ðkÞ ðk Þ ðk Þ
threshold effect value A + ui0 j0 , that is, uij [ A þ ui0 j0 .
In order for the effect measures of each type of objective to satisfy the condition
ðkÞ
of normality, that is, rij 2 ½1; 1; without loss of generality, we can assume that:
n o
ðk Þ ðkÞ ðk Þ
For a benefit-type objective, uij  maxi maxj uij þ 2ui0 j0 ;
n o
ðk Þ ðkÞ ðk Þ
For a benefit-type objective, uij   mini minj uij þ 2ui0 j0 ;
244 10 Grey Models for Decision-Making

For cases where the effect value of a moderate-value type objective is smaller
ðkÞ ðk Þ ðk Þ
than the lower threshold effect value A  ui0 j0 , uij A  2ui0 j0 ; and
For cases where the effect value of a moderate-value type objective is greater
ðkÞ ðk Þ ðk Þ
than the upper threshold effect value A þ ui0 j0 , uij  A þ 2ui0 j0 .
With these assumptions, we have the proposition below.
ðkÞ
Proposition 10.4.1 The effect measures rij ði ¼ 1; 2;    ; n; j ¼ 1; 2;    ; m; k ¼ 1;
2;    ; sÞ; as defined in Definition 10.4.1, satisfy the following properties:
ðkÞ ðkÞ
(1) rij is non-dimensional; (2) the more ideal the effect, the larger rij is; and
ðkÞ
(3) rij 2 ½1; 1:
ðkÞ
Definition 10.4.2 rij ði ¼ 1; 2;    ; n; j ¼ 1; 2;    ; m; k ¼ 1; 2;    ; sÞ; as defined
in Definition 10.4.1, is called uniform effect measure of decision scheme sij .
ðkÞ
For decision scheme sij of hitting the target, rij 2 ½0; 1; and for decision scheme
ðkÞ
sij of missing the target, rij 2 ½1; 0:
 
ðkÞ
Definition 10.4.3 For a given set S; define RðkÞ ¼ rij as the matrix of uniform
nm
ð1Þ ð2Þ ðsÞ
effect measure of S with respect to objective k. For sij 2 S; rij ¼ ðrij ; rij ;    ; rij Þ is
known as the vector of uniform effect measure of the decision scheme sij .

10.4.2 The Weighted Synthetic Effect Measure

Definition 10.4.4 Assume that gk stands for the decision weight of objective k,
Ps Ps ðkÞ
k ¼ 1; 2;    ; s; satisfying k¼1 gk ¼ 1, then k¼1 gk  rij is called a weighted
synthetic effect measure of the decision scheme sij , which is still denoted as
P ðkÞ
rij ¼ sk¼1 gk  rij ; and R ¼ ðrij Þnm is known as the matrix of weighted synthetic
effect measures
In the case of weighted synthetic effect measures, rij 2 ½1; 0 belongs to the
decision scheme sij of missing the target, while rij 2 ½0; 1 belongs to the decision
scheme sij of hitting the target. For the decision scheme of hitting the target, we can
further compare the superiority of events ai , countermeasures bj , and decision
schemes sij respectively by using the magnitudes of the weighted synthetic effect
measures, i ¼ 1; 2; . . .; n; j ¼ 1; 2; . . .; m:


Definition 10.4.5 (1) If max1  j  m rij ¼ rij0 , then bj0 is known as the optimum


countermeasure of event ai ; (2) If max1  i  n rij ¼ ri0 j , then ai0 is known as the


optimum event corresponding to countermeasure bj ; (3) If max1  i  n;1  j  m rij ¼
ri0 j0 , then si0 j0 is known as the optimum decision scheme.
10.4 Multi-attribute Intelligent Grey Target Decision Model 245

The weighted multi-attribute grey target decision can be made by following the
steps below:

Step 1: Based on the set A ¼ fa1 ; a2 ;    ; an g of events and the set B ¼

fb1 ; b2 ;    ; bm g of countermeasures,
 construct the set of decision
schemes S ¼ fsij ¼ ðai ; bj Þai 2 A; bj 2 Bg;
Step 2: Determine the decision objectives k ¼ 1; 2;    ; s;
Step 3: Determine the decision weights g1 ; g2 ;    ; gs of the objectives;
Step 4: For each objective k ¼ 1; 2;    ; s; compute the corresponding observed
ðkÞ
effect matrix U ðkÞ ¼ ðuij Þnm ;
Step 5: Determine the threshold effect value of objective k ¼ 1; 2;    ; s;
Step 6: Calculate the matrix RðkÞ ¼ ðr ðkÞ Þ
ij nm of uniform effect measures of
objective k ¼ 1; 2;    ; s;
Step 7: From r ¼ Ps g  r ðkÞ , compute the matrix of synthetic effect measures
ij k¼1 k ij
R ¼ ðrij Þnm ; and
Step 8: Determine the optimum decision scheme si0 j0 .
The proposed model here has a unique feature of clear physical meaning pre-
sented as missing target, hitting target and hitting performance of different decision
strategies with a pre-defined grey target. The distance of a strategy to the grey target
over different objectives is calculated through effect measure functions as follows:
the concept of upper effect measure reflects the distance of the observed effect value
from the maximum observed effect value; the concept of lower effect measure
indicates the distance between the observed effect value from the minimum
observed effect value; and the concept of moderate effect measure tells the distance
of the observed effect value from the pre-defined most preferred effect value in the
middle.
To aggregate the performance of a strategy over different objectives, one can
make use of the concept of upper effect measure for benefit objectives where the
larger or the more the effect sample values are the better; for cost objectives where
the smaller or the fewer the effect sample values are the better, one can utilize the
concept of lower effect measure. As for non-monotonic objectives that require
“neither too large nor too small” and/or “neither too many nor too few,” one can
apply the concept of moderate effect measure. The effect measure for benefit and
cost type objectives, the lower effect measure for moderate type, and the upper
effect measure for moderate type can be further integrated as uniform effect mea-
sures by incorporating weight information over different objectives. The value of
uniform effect measures is located in the interval of [−1, 1] and has a crystal
physical meaning: if a strategy hits the target, the value will be positive and the
larger the closer to the ideal point in the grey target; if a strategy misses the target,
the value will be negative. The new model has been applied to the selection of the
supplier of a key component used in the production of large commercial aircrafts
and this application confirmed its feasibility.
246 10 Grey Models for Decision-Making

Example 10.4.1 Let us look at the selection of the supplier of a key component
used in the production of large commercial aircrafts.
In China, the production of large commercial aircrafts is managed using the
model of main manufacturers—suppliers, where a great amount of key components
comes from international suppliers. So, the scientific approach to decision-making
regarding the selection of relevant suppliers is a key determinant of the success or
failure of the operation. As a typical decision-making problem involved in the
production process of sophisticated products, the selection of suppliers is generally
accomplished through public bidding. Usually the main manufacturer first lists his
demands, then each potential supplier puts together their proposal to outline how
they meet the needs of the manufacturer. After collecting the proposals, the man-
ufacturer comprehensively evaluates all the suppliers’ submissions to select the
optimum proposal and sign the purchase agreement. As for what factors actually
affect the manufacturer’s decision, it is an extremely complicated matter. In order to
arrive at educated and scientifically sound decisions, there is a need to analyze all
the involved factors closely and holistically.
During the selection of international suppliers for a specific key component of
the production of large commercial aircrafts, there were there suppliers accepted
into the second round of the tender. To decide on the eventual supplier, let us go
through the following steps.
Step 1: Establish the sets of events, countermeasures, and situations. Let us
define event a1 as the selection of a supplier for the said component for the pro-
duction of large commercial aircrafts. So, the set of events is A ¼ fa1 g: Define the
selection of supplier 1, supplier 2, or supplier 3 to be our countermeasures b1 , b2 ,
and b3 , respectively, so that the set of countermeasures is B ¼ fb1 ; b2 ; b3 g:
Therefore, our set of situations in this case is S ¼ fsij ¼ ðai ; bj Þjai 2 A; bj 2 B;
i ¼ 1; j ¼ 1; 2; 3g ¼ fs11 ; s12 ; s13 g.
Step 2: Determine the decision objectives. Through three rounds of surveys with
relevant experts, the following 5 objectives are considered: quality, price, time of
delivery, design proposal, and competitiveness.
Among these objectives, competitiveness, quality, and design proposal are
qualitative. They are scored by relevant experts’ evaluations, and the higher the
evaluation scores the better. That is, they are benefit-type objectives. Let us take the
ðkÞ
threshold value ui0 j0 ¼ 9; k ¼ 1; 4; 5: For the objective of cost, the lower the cost the
ð2Þ
better. So, it is a cost-type objective. Let us take the threshold value ui0 j0 ¼ 15: The
objective of time of delivery is one of moderate-value type. The main manufacturer
desires the delivery at the end of the 16th month with 2 months’ deviation allowed.
ð3Þ
That is, ui0 j0 ¼ 2; the lower threshold effect value is 16 − 2 = 14, and the upper
threshold effect value is 16 + 2 = 18.
Step 3: Determine the decision weights of the objectives. To this end, we apply
the Analytic Hierarchy Process (AHP) method (see Table 10.2 for details).
10.4 Multi-attribute Intelligent Grey Target Decision Model 247

Table 10.2 The objectives’ evaluation system

Objective Quality Price Delivery Design Competitiveness
Unit Qualitative Million US$ Month Qualitative Qualitative
Order # 1 2 3 4 5
Weight 0.25 0.22 0.18 0.18 0.17

Step 4: Determine the effect sample vectors of each of the objectives:

U ð1Þ ¼ ð9:5; 9:4; 9Þ; U ð2Þ ¼ ð14:2; 15:1; 13:9Þ; U ð3Þ ¼ ð15:5; 17:5; 19Þ;
U ð4Þ ¼ ð9:6; 9:3; 9:4Þ; U ð5Þ ¼ ð9:5; 9:7; 9:2Þ:

Step 5: Assign the threshold effect values for the objectives. Because compet-
itiveness, quality, and design proposal are all benefit-type objectives, let us take the
ðkÞ
threshold values ui0 j0 ¼ 9; k ¼ 1; 4; 5: Because price is a cost-type objective, let us
ð2Þ
take the threshold value ui0 j0 ¼ 15: Because time of delivery is a moderate
value-type objective and the main manufacturer desires the delivery at the end of
ð3Þ
the 16th month with a tolerance of ±2 months, we set ui0 j0 ¼ 2; the lower threshold
effect value 16 − 2 = 14, and the upper threshold effect value 16 + 2 = 18.
Step 6: Calculate the vectors of uniform effect measures. For the three quali-
tative objectives, competitiveness, quality, and design proposal, we employ the
effect measures of benefit-type. For the objective of price, we utilize the effect
measures of cost-type. For the objective of time of delivery, we apply the lower and
upper effect measures. Thus, we obtain the following vectors of uniform effect
measures:

Rð1Þ ¼ ½1; 0:8; 0; Rð2Þ ¼ ½0:73; 0:09; 1; Rð3Þ ¼ ½0:75; 0:25; 0:5;
Rð4Þ ¼ ½1; 0:5; 0:67; and Rð5Þ ¼ ½0:71; 1; 0:29:
P5 ðkÞ
Step 7: From rij ¼ k¼1 gk  rij , we compute the following vector of synthetic
effect measures;

R ¼ ½1; 0:5; 0:67 ¼ ½0:8463; 0:4852; 0:2999:

Step 8: Make the final decision. Because r11 [ 0; r12 [ 0; r13 [ 0; it means that
all these three suppliers have hit the target. This result implies that it is reasonable
for these suppliers

 to enter the second round of the tender. However, based on
max1  j  3 r1j ¼ r11 ¼ 0:08463; it follows that the main manufacturer should
sign the agreement with supplier 1.
248 10 Grey Models for Decision-Making

10.5 The Paradox of Rule of Maximum Value

and Its Solution

The thought and methodology of statistical decision emerged in Britain in the late
18th century (Bayes 1763). L.J. Savage built the system of Bayesian Decision
Theory in the book titled. The foundations of statistics in 1954. According to
Bayesian Decision Theory (Savage 1954), the reasonable action subject abide by
the principles of maximum subjective expected utility in their decision-making. In
1969, R. Nozick published his article on the Newcomb Paradox (Nozick 1969),
which led to a major divergence of views in Bayesian Decision Theory and gave
rise to Causal Decision Theory (Lewis 1973) and Evidential Decision Theory, also
known as D-S Theory (Dempster 1968; Shafer 1976). A. Gibbard and W. Harper
tried to solve the Newcomb Paradox by defining two different expected utilities
named as U-utility and V-utility in 1978 (Gibbard and Harper 1978). Then, in 1981
E. Eells thought that people could solve the Newcomb Paradox by revising the
principles of maximum utility (Eells 1981). J.M. Joyce put forward a general theory
of conditional beliefs to illuminate the nature of causal beliefs and their role in
rational choice (Joyce 1999). It was concluded that there is less difference than is
usually thought between causal decision theory and evidential decision theory
(Joyce 1999). S. Burgess thought that the resolution of the Newcomb problem was
unqualified (Burgess 2004), and in 2010 D.H. Wolpert and G. Benford proved that
two Bayes nets are incompatible based on game theory (Wolpert and Benford
2010).
In the course of decision-making, people need divide their decision objects into
different classes or clusters, then compare and sort the objects in the same class or
cluster to help them choose the right object or objects. All types of cluster evalu-
ation models, such as statistical clustering analysis (Tryon 1939), fuzzy clustering
(Bezdek 1981), and grey clustering evaluation models such as grey variable weight
clustering model (Deng 1986), grey fixed weight clustering evaluation model (Liu
1993), grey cluster evaluation model using end-point triangular possibility func-
tions (Liu et al. 1993, 2006), grey cluster evaluation model using center-point
triangular possibility functions (Liu et al. 2011, 2012), and grey cluster evaluation
model using mixed triangular possibility functions (Liu et al. 2014, 2015, 2015) use
the rule of maximum value of component of cluster coefficient vector ri ¼
ðr1i ; r2i ;    ; rsi Þ as a basis for determining ascription of decision objects.
In cases where more than one object belongs to a class or cluster, people may be
confronted with a decision paradox. For example, assume that d1 ¼
ð0:4; 0:35; 0:25Þ and d2 ¼ ð0:41; 0:2; 0:39Þ are the clustering coefficient vectors of
objects 1 and 2, respectively. It is demonstrably the case that objects 1 and 2 both
belong to class 1, according to the principles of maximum value of clustering
coefficient. Also, object 2 is better than object 1 given that 0:41 [ 0:4: However, if
we were to consider the values of all the components of d1 ; d2 in an integrated
manner, object 1 could be perceived as being superior to object 2. This is a paradox.
10.5 The Paradox of Rule of Maximum Value and Its Solution 249

In this section, we try and find a solution for the decision paradox by using
weight vector group of kernel clustering, weighted coefficient vector of kernel
clustering for decision-making and a two-stages decision model.

10.5.1 The Weight Vector Group of Kernel Clustering

Clustering coefficient vectors cannot be compared with each other because usually
they are not unit vectors. Therefore, firstly all clustering coefficient vectors need to
be unitized.
Definition 10.5.1 Assume that ri ¼ ðr1i ; r2i ;    ; rsi Þ; i ¼ 1; 2;    ; n are n cluster-
rk
ing coefficient vectors, dki ¼ Ps i k ; dki is called unitized clustering coefficient for
k¼1
ri

decision-making of object i belonging to class k. Clearly, dki ðk ¼ 1; 2;    sÞ satisfy

Ps
i¼1 di ¼ 1:
k

Definition 10.5.2 di ¼ ðd1i ; d2i ;    ; dsi Þ; ði ¼ 1; 2;    ; nÞ is called unitized cluster-

ing coefficient vectors for decision-making of object i: The following conclusion
about unitized clustering coefficient vector di is also suitable for non-unitized
clustering coefficient vector ri . Therefore, the “unitized” can be omitted.
Sort the components of di according to their values, that is,
di dki 2    dki l    dki s .
k1

 
Definition 10.5.3 Assume that max1  k  s fdki g ¼ dki , then dki is called the maxi-
mum component of clustering coefficient vector di . Given that all the corresponding
coefficients of two decision coefficient vectors di , dj are equal, then there is no
difference between di , dj . When two objects i, j belong to a class k* and the
 
maximum component dki [ dkj , it means that di is better than dj by the rule of
maximum value; but it is possible to think that dj is better than di if we consider the
values of all the components of d1 ; d2 in an integrated manner. This is a decision
paradox of rule of maximum value.
To solve the decision paradox of rule of maximum value, firstly the weight
vector group of kernel clustering is defined. The basic step to solve the paradox is to
cluster the information which is included in other components around dki , and
supporting objects i come under class k into component k. Then it is necessary to
obtain a new decision coefficient vector which contains factors included in other
components around dki .
Definition 10.5.4 Assume that there are s classes of decision-making, and real
numbers wk 0; k ¼ 1; 2;    ; s; let
250 10 Grey Models for Decision-Making

1
g1 ¼ Ps ðws ; ws1 ; ws2 ;    ; w1 Þ;
wk
k¼1
1
g2 ¼ P ðws1 ; ws ; ws1 ; ws2 ;    ; w2 Þ;
ws1 þ sk¼2 wk
1
g3 ¼ P ðws2 ; ws1 ; ws ; ws1 ;    ; w3 Þ;
ws1 þ ws2 þ sk¼3 wk
;
1
gk ¼ Ps1 Ps ðwsk þ 1 ; wsk þ 2 ;    ; ws1 ; ws ; ws1 ;    ; wk Þ;
i¼sk þ 1 wi þ i¼k wi
;
1
gs1 ¼ Ps ðw2 ; w3 ;    ws1 ; ws ; ws1 Þ;
ws1 þ k¼2 wk
1
gs ¼ Ps ðw1 ; w2 ; w3 ;    ; ws1 ; ws Þ;
k¼1 wk

then gk ðk ¼ 1; 2;    ; sÞ is called a weight vector group of kernel clustering about

class k (Liu et al 2014).
Note: s-dimensional vector gk ¼ ðg1k ; g2k ;    ; gsk Þðk ¼ 1; 2;    ; sÞ is the multi-
plication of scalar ak ¼ Ps1 1 P
s with vector fk , where the function of
i¼sk þ 1
wi þ i¼k
wi
scalar factor ak is to ensure gk ðk ¼ 1; 2;    ; sÞ is a unit vector. Also, the k-th
component of vector factor fk ðk ¼ 1; 2;    ; sÞ is ws , which is the maximum
component of fk . Then the k-th component ws can be taken as a center, and the
other components on both sides of the k-th component ws descend step by step. The
k-th component with the largest contribution for the decision-making object
belongs to grey class k, so the k-th component of fk should take the maximum
weight ws . The values of other components are set by the principle which states that
“the component which is closest to the k-th component has the largest contribution
for object I belonging to class k, so it is given the largest weight; the component
which is farthest from the k-th component has the smallest contribution for object
I belonging to class k, so it is given the smallest weight”.

10.5.2 The Weighted Coefficient Vector of Kernel Clustering

for Decision-Making

Definition 10.5.5 Assume there are n decision objects and s different grey classes,
then xki ¼ gk  dTi is called the weightedcoefficient of kernel clustering for
decision-making of object i about grey class k. And
10.5 The Paradox of Rule of Maximum Value and Its Solution 251

xi ¼ ðx1i ; x2i ;    ; xsi Þ; i ¼ 1; 2;    ; n

is called the weighted coefficient vector of kernel clustering for decision-making of

object i.
n o 
n o 
Definition 10.5.6 Let max1  k  s xki1 ¼ xki1 , max1  k  s xki2 ¼ xki2 , when
xi1 [ xi2 , then decision object i1 is better than decision object i2 in grey class k  .
n o 
n o 
Definition 10.5.7 Let max1  k  s xki1 ¼ xki1 , max1  k  s xki2 ¼ xki2 ,    ;
n o 
max1  k  s xkil ¼ xkil , in other words, objects i1 ; i2 ;    ; il all belong to grey class
k . Also, xi1 [ xi2 [    [ xil , and if the number of objects contained in the
decision grey class k is l1 , then objects i1 ; i2 ;    ; il1 are called the taken object of
grey class k , and the rest of the objects are called the candidates of grey class k  .
The two stages decision model to solve the decision paradox with the weight
vector group of kernel clustering and weighted coefficient vector of kernel clus-
tering for decision-making can be constructed step by step as outlined below
(Liu et al. 2014).
Stage 1
Step 1: Compute unitized clustering coefficient vector di

di ¼ ðd1i ; d2i ;    ; dsi Þ; ði ¼ 1; 2;    ; nÞ

Step 2: Estimate the distinguishability of the clustering coefficient vectors of

objects belonging to class k*. If the order of priority of the objects i belonging to
class k* is easy to identify, turn to step 6; in cases where the order of priority of the
objects belonging to class k* is difficult to identify, turn to step 3;
Stage 2
Step 3: Set the weight vector group of kernel clustering ðg1 ; g2 ;    ; gs Þ;
Step 4: Calculate the weighted coefficient vector of kernel clustering for
decision-making of decision object i

xi ¼ ðx1i ; x2i ;    ; xsi Þ; i ¼ 1; 2;    ; n ;

 
Step 5: Determine object i belonging to grey class k by max1  k  s xki ¼ xki ;
Step 6: Sort the decision objects which belong to class k* according to the values
  
of dki1 ; dki2 ;    ; dkil for case where there are l objects belonging to class k .
252 10 Grey Models for Decision-Making

10.5.3 Several Functional Weight Vector Groups of Kernel

Clustering

Proposition 10.5.1 Assume that

2
g1 ¼ ðs; s  1; s  2;    ; 1Þ
sðs þ 1Þ
1
g2 ¼ sðs þ 1Þ ðs  1; s; s  1; s  2;    ; 2Þ
2 þ ðs  2Þ
1
g3 ¼ sðs þ 1Þ ðs  2; s  1; s; s  1;    ; 3Þ
2 þ ð2s  6Þ
;
1
gk ¼ fsðs þ 1Þ ðs  k þ 1; s  k þ 2;    ; s  1; s; s  1;    ; kÞ;
2 þ ½ðk  1Þs  kðk1Þ
2 
;
2
gs1 ¼ sðs þ 1Þ ð2; 3;    ; s  1; s; s  1Þ
2 þ ðs  2Þ
2
gs ¼ ð1; 2; 3;    s  1; sÞ
sðs þ 1Þ

Then gk ðk ¼ 1; 2;    ; sÞ is a weight vector group of kernel clustering (Liu et al.

2014).
Proposition 10.5.2 Assume that

1 1 1 1 1 1
g1 ¼ Ps ð ; 2 ; 3 ;    s1 ; s Þ
1
2 2 2
k¼1 2k 2 2
1 1 1 1 1 1
g2 ¼ ð 1 Ps1 1 Þð 2 ; ; 2 ; 3 ;    ; s1 Þ
22 þ k¼1 2k
2 2 2 2 2
1 1 1 1 1 1 1
g3 ¼ ð 1 Ps2 1 Þð 3 ; 2 ; ; 2 ; 3    ; s2 Þ
23 þ 1
22 þ k¼1 k
2 2 2 2
2
2 2
;
1 1 1 1 1 1 1
gk ¼ fPk Psk þ 1 1 gð k ; k1 ;    ; 2 ; ; 2 ;    ; sk þ 1 Þ
1
i¼2 2i þ i¼1 2i 2 2 2 2 2 2
;
1 1 1 1 1 1
gs1 ¼ Ps1 ð ; ;    2 ; ; 2Þ
þ 1
22
1
k¼1 2k
2s1 2s2 2 2 2
1 1 1 1 1 1
gs ¼ Ps 1
ð s ; s1 ;    3 ; 2 ; Þ
k¼1 2k 2 2 2 2 2
10.5 The Paradox of Rule of Maximum Value and Its Solution 253

Then gk ðk ¼ 1; 2;    ; sÞ is a weight vector group of kernel clustering (Liu et al.

2015).
Proposition 10.5.3 For case s = 10, assume that

1
g1 ¼ ð1; 0: 9; 0: 8; 0: 7    ; 0:1Þ
5: 5
1
g2 ¼ ð0: 9; 1; 0: 9; 0: 8;    ; 0: 2Þ
6: 3
1
g3 ¼ ð0: 8; 0: 9; 1; 0: 9;    ; 0: 3Þ
6: 9
;
1
gk ¼ Pk P ð0: ð10  kÞ; 0: 8; 0: 9; 1; 0: 9    ; 0:kÞ
1 þ i¼1 0: ð10  iÞ þ 9i¼k 0:i
;
1
g9 ¼ ð0: 2; ;    0: 8; 0: 9; 1; 0: 9Þ
6: 3
1
g10 ¼ ð0: 1; ;    0: 7; 0: 8; 0: 9; 1Þ
5: 5

Then gk ðk ¼ 1; 2;    ; sÞ is a weight vector group of kernel clustering (Liu et al.

2015).

10.6 Practical Applications

Example 10.6.1 Strategic vendor selection for the C919 collaboration. C919 is the
first large commercial aircraft developed by Commercial Aircraft Corporation of
China Ltd (COMAC). Many domestic and overseas vendors joined the develop-
ment program. Vendors A and B took part in the development task of the C919
program for a specific key component. One of the two vendors, either A or B,
should be chosen and confirmed as a strategic vendor according to COMAC’s
criteria.
The consulting group collected all data according the evaluation index system,
which was determined in advance. Then the clustering coefficient vectors of A and
B are defined as follows:

dA ¼ ðd1A ; d2A ; d3A ; d4A ; d5A Þ ¼ ð0:246; 0:338; 0:292; 0:124; 0Þ

dB ¼ ðd1B ; d2B ; d3B ; d4B ; d5B Þ ¼ ð0:089; 0:352; 0:312; 0:197; 0Þ
254 10 Grey Models for Decision-Making

Here, classes 1, 2, 3, 4, 5 correspond to ‘especially excellent’, ‘excellent’,

‘good’, ‘moderate’, and ‘poor’, respectively.



From max1  k  5 dkA ¼ 0:338 ¼ d2A , max1  k  5 dkB ¼ 0:352 ¼ d2B , it is
known that the two vendors A and B both belong to class ‘excellent’. It seems that
B should be selected and confirmed as the strategic vendor if we compare the
clustering coefficients d2A of A belonging to class excellent with d2B of B belonging
to class excellent, because d2A ¼ 0:338\d2B ¼ 0:352: But we found that the clus-
tering coefficients d1A ¼ 0:246 of A belonging to class ‘especially excellent’ is
greater than the clustering coefficients d1B ¼ 0:089 of B belonging to class ‘espe-
cially excellent’ if we compare dA and dB in an integrated way. Therefore, the
values of each component of the clustering coefficient vectors dA and dB should be
integrated by a should be integrated by a weight vector group of kernel clustering.
The weight vector group of kernel clustering presented in Proposition 10.5.2 is
used to integrate the values of each component of the clustering coefficient vectors
dA and dB . Notice that s = 5. We obtain:

32 1 1 1 1 1 16 1 1 1 1 1
g1 ¼ ð ; ; ; ; Þ; g2 ¼ ð 2 ; ; 2 ; 3 ; 4 Þ;
31 2 22 23 24 25 19 2 2 2 2 2
4 1 1 1 1 1 16 1 1 1 1 1
g3 ¼ ð 3 ; 2 ; ; 2 ; 3 Þ; g4 ¼ ð 4 ; 3 ; 2 ; ; 2 Þ;
5 2 2 2 2 2 19 2 2 2 2 2
32 1 1 1 1 1
g5 ¼ ð 5 ; 4 ; 3 ; 2 ; Þ
31 2 2 2 2 2

Then, from xkj ¼ gk  dTj ; j ¼ A; B, we have

32 1 1 1 1 1
x1A ¼ g1  dTA ¼ ð ; ; ; ; Þ  ð0:246; 0:338; 0:292; 0:124; 0ÞT ¼ 0:26
31 2 22 23 24 25
16 1 1 1 1 1
x2A ¼ g2  dTA ¼ ð 2 ; ; 2 ; 3 ; 4 Þ  ðð0:246; 0:338; 0:292; 0:124; 0ÞT ¼ 0:27
19 2 2 2 2 2
xA ¼ g3  dA ¼ 0:23; x4A ¼ g4  dTA ¼ 0:16; x5A ¼ g5  dTA ¼ 0:10
3 T

xA ¼ ðx1A ; x2A ; x3A ; x4A ; x5A Þ ¼ ð0:26; 0:27; 0:23; 0:16; 0:10Þ
x1B ¼ g1  dTB ¼ 0:19; x2B ¼ g2  dTB ¼ 0:25; x3B ¼ g3  dTB ¼ 0:24;
x4B ¼ g4  dTB ¼ 0:19; x5B ¼ g5  dTB ¼ 0:12
xB ¼ ðx1B ; x2B ; x3B ; x4B ; x5B Þ ¼ ð0:19; 0:25; 0:24; 0:19; 0:12Þ

When comparing the weighted coefficient vector of kernel clustering decision-

making of xA and xB , we found that x1A ¼ 0:26 [ x1B ¼ 0:19; x2A ¼ 0:27 [
x2B ¼ 0:25; at the same time, x4A ¼ 0:16\x4B ¼ 0:19, x5A ¼ 0:10\x5B ¼ 0:12:
10.6 Practical Applications 255

So, we can use the judgment that the vendor A is better than vendor B. Vendor A
should be selected and confirmed as the strategic vendor.
We can obtain the same conclusion if the weight vector group of kernel clus-
tering presented in Proposition 10.5.1 or Proposition 10.5.3 is used to integrate the
values of each component of the clustering coefficient vectors dA and dB .
It is directed against the decision paradox that the conclusion we arrive at by
comparing the maximum components dki and dkj of di and dj is in conflict with the
conclusion we arrive at by comparing di and dj , in an integrated way. The weight
vector group of kernel clustering and the weighted coefficient vector of kernel
clustering for decision-making are defined, and a novel two-stage decision model
using weight vector group of kernel clustering and weighted coefficient vector of
kernel clustering for decision-making is put forward here. The decision paradox that
the value of the maximum component dki of di is close to the maximum component
dkj of dj is solved effectively. Indeed, the new model presented in this chapter is
applied to a practical problem, which concerns the strategic partner selection of a
collaboration vendor who can take part in the development of COMAC’s C919
program. The outcome can provide a basis for COMAC’s strategic vendor
selection.
Chapter 11
Grey Control Systems

11.1 Introduction

As a scientific concept, the so-called control stands for a special effect a controlling
device exerts on controlled equipment. It is a purposeful, selective and dynamic
activity. A control system contains at least three parts, including a controlling
device, controlled equipment, and an information path. A control system made up
of these three parts is known as an open loop control system, as shown in Fig. 11.1.
Each open loop control system is quite elementary in that the input directly controls
the output, with no resistance against disturbances.
A control system with a feedback return is known as a closed loop control
system, as shown in Fig. 11.2. The closed loop control system materializes its
control through the combined effect of the input and the feedback of the output. One
of the outstanding characteristics of closed loop systems is their strong ability to
assist disturbances, with their outputs constantly vibrating around pre-determined
objectives. Therefore, closed loop control systems possess a degree of stability.
A grey control system stands for such a system whose control information is
only partially known, and is known as a grey system for short. The control of grey
systems is different to that of general white systems, mainly due to the existence of
grey elements in such systems. Under such conditions, one first needs to understand
the possible connection between the systems’ behaviors and the parametric matrices
of the grey elements, how the systems’ dynamics differ from one moment to the
next and, in particular, how to obtain a white control function to alter the charac-
teristics of the systems and to materialize control of the process of change of the
systems. Grey control contains not only the general situation of systems involving
grey parameters, but also the construction of controls based on grey systems
analysis, modeling, prediction, and decision-making. Grey control thinking can
reveal the essence of the problems at hand and help materialize the purpose of
control.

© Springer Science+Business Media Singapore 2017 257

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_11
258 11 Grey Control Systems

Fig. 11.1 The open loop control system

Fig. 11.2 The closed loop control system

11.2 Controllability and Observability of Grey System

The concepts of controllability and observability are two fundamental structural

characteristics of systems seen from the angle of control and observation. This
section focuses on the problems of controllability and observability of grey linear
systems.
Definition 11.2.1 Assume that U ¼ ½u1 ; u2 ;    ; us T is a control vector, X ¼
½x1 ; x2 ;    ; xn T a state vector, and Y ¼ ½y1 ; y2 ;    ; ym T the output vector. Then
(
X_ ¼ AðÞX þ BðÞU
ð11:1Þ
Y ¼ CðÞX

is known as the mathematical model of a grey linear control system, where

AðÞ 2 Gnn ; BðÞ 2 Gns ; CðÞ 2 Gmn . Correspondingly, AðÞ is known as
the grey state matrix, BðÞ the grey control matrix, and CðÞ the grey output
matrix.
In some studies, to emphasize the fact that U, X, and Y change the dynamic
characteristics of the system over time, we also respectively write the control
vector, state vector, and the output vector as UðtÞ; XðtÞ, and YðtÞ.
The first group of equations

_
XðtÞ ¼ AðÞXðtÞ þ BðÞUðtÞ ð11:2Þ
11.2 Controllability and Observability of Grey System 259

in the mathematical model of grey linear control systems in Eq. (11.1) is known as
the state equation, while the second group of equations

YðtÞ ¼ CðÞXðtÞ ð11:3Þ

is known as the output equation.

Definition 11.2.2 For a given precision and an objective vector J ¼ ½j1 ; j2 ;    ; jm T ,
with a controlling device and a control vector UðtÞ such that the output of the system
can reach objective J while satisfying the required precision through controlling the
input, then the system is said to be controllable.
Definition 11.2.3 For a given time moment t0 and a pre-determined precision, if
there is t1 2 ðt0 ; 1Þ such that based on the system’s output YðtÞ, t 2 ½t0 ; t1 , one can
measure the system’s state XðtÞ within the required precision, then the system is
said to be observable within the time interval ½t0 ; t1 . If for any t0 ; t1 , the system is
observable within the interval ½t0 ; t1 , then the system is said to be observable.
According to control theory, it follows that whether or not a grey system is
controllable or observable is determined by whether or not the controllability matrix
and the observability matrix, made up of AðÞ; BðÞ, are of full rank. That is, the
following result holds true.
Theorem 11.2.1 For the system in Eq. (11.1), define

LðÞ ¼ ½BðÞ AðÞBðÞ A2 ðÞBðÞ  An1 ðÞBðÞT

DðÞ ¼ ½CðÞ CðÞAðÞ CðÞA2 ðÞ  CðÞAn1 ðÞT

Then the following hold true:

(1) When rankðLðÞÞ ¼ n, the system is controllable; and
(2) When rankðDðÞÞ ¼ n, the system is observable (Su and Liu 2008).
Based on this result, the following four theorems can be established.
Theorem 11.2.2 For the system in Eq. (11.1), if the grey control matrix BðÞ 2
Gnn satisfies BðÞ = diag[ 11 , 22 , …, nn ], where each grey entry along the
diagonal is non-zero, then the system is controllable.
Theorem 11.2.3 For the system in Eq. (11.1), if the grey output matrix CðÞ 2
Gnn satisfies CðÞ = diag[ 11 , 22 , …, nn ], where each grey entry along the
diagonal is non-zero, then the system is observable.
Theorem 11.2.4 For the system in Eq. (11.1), if the control matrix BðÞ 2 Gnn
satisfies BðÞ = diag[ 11 , 22 , …, mm , 0, …, 0] with rankBðÞ ¼ m\n, and the
grey state matrix AðÞnn = diag[0, …, 0, m þ 1;1 ,m þ 2;2 ,…, n;nm ] with
rankAðÞ ¼ n  m\n, then the system is controllable.
260 11 Grey Control Systems

Theorem 11.2.5 For the system in Eq. (11.1), if the grey output matrix CðÞ 2
Gmn satisfies CðÞ = diag[ 11 ,22 , …, mm ] with rankCðÞ ¼ m\n and the
grey state matrix
0 1
0  0 1;m þ 1 0  0
B C
B0  0 02;m þ 2  0 C
B C
B. .. .. .. .. C
B ..  . . .  . C
B C
B C
AðÞ ¼B 0  0 0 0  nm;n C;
B C
B0   0 C
B 0 0 0 C
B. .. .. .. .. C
B. C
@.  . . .  . A
0 cdots 0 0 0  0
rankAðÞ ¼ n  m\n;

then the system is observable.

11.3 Transfer Functions of Grey System

The concept of transfer functions stands for a fundamental relationship between the
input and output of time invariant, linear grey control systems. Its rich connection
with the expressions of the systems’ state spaces can be described by using the
concepts of controllability and observability.

11.3.1 Grey Transfer Function

Definition 11.3.1 Assume that the mathematical model of an nth order

linear system with grey parameters is given as follows:

dn x d n1 x
n n
þ n1 n1 þ    þ 0 x ¼   uðtÞ ð11:4Þ
dt dt

After applying Laplace transform to both sides of this equation, we obtain

XðsÞ 
GðsÞ ¼ ¼ ð11:5Þ
UðsÞ n sn þ n1 sn1 þ    1 s þ 0

where LðxðtÞÞ ¼ XðsÞ and LðuðtÞÞ ¼ UðsÞ. Equation (11.5) is known as a grey
transfer function, which is the ratio of the Laplace transform of the response xðtÞ of
the nth order grey linear control system and the Laplace transform of the driving
11.3 Transfer Functions of Grey System 261

term uðtÞ. In fact, the transfer function represents a fundamental relationship

between the input and output of a first order grey linear control system. From the
following theorem, it follows that each nth order grey linear system can be reduced
to an equivalent first order grey linear system.
Theorem 11.3.1 For an nth order grey linear system as shown in Eq. (11.4), there
is an equivalent first order grey linear system.
Proof Assume that the given nth order grey linear system is

dn x d n1 x
n þ  n1 þ    þ 0 x ¼   uðtÞ
dtn dtn1

Let

dx dx1 d 2 x dx2 d n1 x dxn1

x ¼ x1 ; ¼ ¼ x2 ; 2 ¼ ¼ x3 ;    ; n1 ¼ ¼ xn
dt dt dt dt dt dt

Therefore, we have

dxn 0 1 2 n1 
¼ x1  x2  x3      xn þ uðtÞ
dt n n n n n

and the nth order system is reduced to the following first order system

_
XðtÞ ¼ AðÞXðtÞ þ BðÞUðtÞ

where XðtÞ ¼ ½x1 ; x2 ;    ; xn T , UðtÞ ¼ uðtÞ,

2 3
0
2 3 6 7
0 1 0  0 6 07
60  7 6 7
6 0 1 0 7 6 .7
AðÞ ¼ 6 7; and BðÞ ¼ 6 . 7
6 .
6    0 7 6
7:
7
40 0   1 5 6 07
  1   n1
 n 6 7
0
n n 45
n

This ends the proof.

11.3.2 Transfer Functions of Typical Links

A grey control system that is symbolically written in an equation is also known as a

grey link. When the transfer function of a link is known, from the relationship
XðsÞ ¼ GðsÞ  UðsÞ and the Laplace transform of the driving term, one can obtain
262 11 Grey Control Systems

Fig. 11.3 The driving and

response terms

the Laplace transform of the response. Then, by using the inverse Laplace trans-
form, one can produce the response xðtÞ. The relationship between the driving and
response terms is depicted in Fig. 11.3.
In the following definition, let us look at the transfer functions of several typical
links.
Definition 11.3.2 The link between driving term uðtÞ and response term xðtÞ
satisfying

xðtÞ ¼ KðÞuðtÞ ð11:6Þ

is known as a grey proportional link, where KðÞ is the grey magnifying coefficient
of the link.
Proposition 11.3.1 The transfer function of a grey proportional link is

GðsÞ ¼ KðÞ ð11:7Þ

The characteristics of a grey proportional link are that when a jump occurs in the
driving quantity, the response value changes proportionally. This kind of change
and relationship between the drive and response are depicted in Fig. 11.4.
Definition 11.3.3 When driven by a unit jump, if the response is given by

xðtÞ ¼ KðÞð1  etT Þ ð11:8Þ

then the link is known as a grey inertia link, where T stands for a time constant of
the link.

Fig. 11.4 The grey proportional link

11.3 Transfer Functions of Grey System 263

Fig. 11.5 The grey inertia link

Proposition 11.2.2 The transfer function of a grey inertial link is given by

KðÞ
GðsÞ ¼ ð11:9Þ
T  sþ1

The characteristics of a grey inertia link are that when a jump occurs in the
driving quantity, the response can reach a new state of balance only after a period
of time. Figure 11.5 provides a block diagram and the curve of change of the
~
response of a grey inertia link when KðÞ ¼ 1.
Definition 11.3.4 When the drive and response are related as follows, the link is
known as grey integral link:
Z
xðtÞ ¼ KðÞuðtÞdt ð11:10Þ

Proposition 11.3.3 The transfer function of a grey integral link is given below:

KðÞ
GðsÞ ¼ ð11:11Þ
s

For a grey integral link, when the drive is a jump function, its response is
xðtÞ ¼ KðÞut, as shown in Fig. 11.6.
Definition 11.3.5 If the response and the drive are related as follows, the link is
known as a grey differential link:

duðtÞ
xðtÞ ¼ KðÞ ð11:12Þ
dt
Proposition 11.3.4 The transfer function of a grey differential link is given as
follows:
264 11 Grey Control Systems

Fig. 11.6 The grey integral link

GðsÞ ¼ KðÞs ð11:13Þ

The characteristics of a grey differential link are that when the drive stands for a
jump, the response becomes an impulse with an infinite amplitude.
Definition 11.3.6 If the drive and response are related as follows, the link is known
as a grey postponing link, where sðÞ is a grey constant:

xðtÞ ¼ uðt  sðÞÞ ð11:14Þ

Proposition 11.3.5 The transfer function of a grey postponing link is given below:

GðsÞ ¼ esðÞs ð11:15Þ

For a grey postponing link, when the drive is a jump function, it takes some time
for the response to react accordingly. For details, see Fig. 11.7.
The figure above represents some typical links met in practical applications.
Many complicated devices and systems can be treated as combinations of these
typical links. For instance, when the grey proportional link is combined with a grey
differential link, one can obtain a grey proportional differential link. When a grey
integral link is connected with grey postponing link, one establishes a grey integral
postponing link. Along the same lines, multi-layered combinations can be

Fig. 11.7 The grey postponing link

11.3 Transfer Functions of Grey System 265

developed for practical purposes. One of the purposes of studying grey transfer
functions is that we can investigate the stabilities and other properties of systems by
looking at the extreme values of relevant transfer functions.

11.3.3 Matrices of Grey Transfer Functions

Matrices of grey transfer functions can be employed to express a fundamental

relationship between the multi-inputs and multi-outputs of grey linear control
systems. In particular, for the following grey linear control system
(
_
XðtÞ ¼ AðÞXðtÞ þ BðÞUðtÞ
YðtÞ ¼ CðÞXðtÞ

Employing Laplace transforms produces

(
sXðsÞ ¼ AðÞXðsÞ þ BðÞUðsÞ
YðsÞ ¼ CðÞXðsÞ

and
(
ðsE  AðÞÞXðsÞ ¼ BðÞUðsÞ
YðsÞ ¼ CðÞXðsÞ

If ðsE  AðÞÞ is invertible, then we can further obtain

(
XðsÞ ¼ ðsE  AðÞÞ1 BðÞUðsÞ
YðsÞ ¼ CðÞXðsÞ

That is, we have YðsÞ ¼ CðÞðsE  AðÞÞ1 BðÞUðsÞ.

Definition 11.3.7 The m n matrix below is known as the matrix of grey transfer
functions:

GðsÞ ¼ CðÞðsE  AðÞÞ1 BðÞ ð11:16Þ

Definition 11.3.8 For an nth order grey linear system, if the state grey matrix AðÞ
of the corresponding equivalent first order system is non-singular, then

lim GðsÞ ¼ CðÞAðÞ1 BðÞ ð11:17Þ

s!0
266 11 Grey Control Systems

is known as a grey gain matrix. If the grey gain matrix CðÞAðÞ1 BðÞ is used
to replace the transfer function GðsÞ, then the system is reduced into a proportional
link. Because YðsÞ ¼ GðsÞUðsÞ, when m ¼ s ¼ n, if GðsÞ is non-singular, we have
the following:
UðsÞ ¼ GðsÞ1 YðsÞ ð11:18Þ

Definition 11.3.9 The following matrix is known as a grey structure matrix:

GðsÞ1 ¼ BðÞ1 ðsE  AðÞÞCðÞ1 ð11:19Þ

When the grey structure matrix is known, to make the output vector YðsÞ meet or
close to meet a certain expected objective JðsÞ, one can determine the system’s control
vector UðsÞ through G1 ðsÞ  JðsÞ. Additionally, we can also discuss the controlla-
bility and observability of systems by using matrices of grey transfer functions.

11.4 Robust Stability of Grey System

Stability is a fundamental structural characteristic of systems. It stands for an

important mechanism for a system to sustain itself and is a prerequisite for the
system to operate smoothly. This is why stability is studied in systems control
theory and it is a key objective in relevant engineering designs. Each physical
system has to be stable before it can be employed in practical applications.
The stability of grey systems focuses on the investigations of informational
changes. It also focuses on whether or not the grey system of concern stays stable or
can recover to its stability when the whitenization value of a grey parameter moves
within the field of discourse. The existence of grey parameters complicates the
study of grey systems stability, and puts them at the center of attention of control
theory and control engineering.
In grey systems modeling, there is a distinction between having a postponing
term and not having such a term; there is also a difference between having a random
term and not having such a term. Ordinarily, grey systems without involving any
random and postponing term are known as grey systems; those involving post-
poning terms without any random terms are grey postponing systems, and those
involving random terms are known as grey stochastic systems. In this section, we
will study the problem of robust stability of these three kinds of systems.

11.4.1 Robust Stability of Grey Linear Systems

The study of systems’ stability is often limited to systems without the effect of any
external input. This kind of system is known as an autonomous system. A simple
grey linear autonomous system can be written as follows:
11.4 Robust Stability of Grey System 267

(
x_ ðtÞ ¼ AðÞxðtÞ
ð11:20Þ
xðt0 Þ ¼ x0 ; 8t  t0

where x 2 Rn stands for the state vector, and AðÞ 2 Gnn is the matrix of grey
coefficients.
~ is a whitenization matrix of the grey matrix AðÞ, then
Definition 11.4.1 If AðÞ
(
~
x_ ðtÞ ¼ AðÞxðtÞ
ð11:21Þ
xðt0 Þ ¼ x0

is referred to as a whitenization system of the system in Eq. (11.20).

Ordinarily, we assume that the matrix AðÞ of grey coefficients of the system in
Eq. (11.20) has a continuous matrix cover:

~ : aij  
AðDÞ ¼ ½La ; Ua  ¼ fAðÞ ~  aij ; i; j ¼ 1; 2;    ; ng;

where Ua ¼ ð
aij Þ; La ¼ ðaij Þ.
Definition 11.4.2 If any whitenization system of the system in Eq. (11.20) is
stable, then the system in Eq. (11.20) is referred to as robust stable.
The ordinary concept of a system’s (robust) stability represents the (robust)
asymptotic stability of the system.
Theorem 11.4.1 If there is positive definite matrix P such that

PLa þ LTa P þ 2kmax ðPÞjjUa  La jjIn \0

then the system in Eq. (11.20) is robust stable (Su and Liu 2009).
Proof Let us take the Lyapunov function VðxÞ ¼ xT Px. For any whitenization
matrix AðÞ~ 2 AðDÞ, let us compute the derivative of VðxÞ with respect to t along
the trajectory of the whitenization system and obtain
:  
~ ¼ xT PLa þ LTa P x þ 2xT PDAx
V ðxÞ ¼ 2xT PAðÞx
 
 xT PLa þ LTa P þ 2kmax ðPÞkUa  La kIn x\0; 8x 6¼ 0

This implies that the system in Eq. (11.20) is robust stable. QED. If in
Theorem 11.4.1 we let P ¼ In , then we have the result shown below.
Corollary 11.4.1 If

La þ LTa
jjUa  La jj\  kmax ð Þ ð11:22Þ
2
268 11 Grey Control Systems

holds true, then the system in Eq. (11.20) is robust stable. If we employ another
form of decomposition AðÞ ~ ¼ Ua  DA of the whitenization matrix AðÞ ~ to study
the robust stability of the system in Eq. (11.20), then much like in Theorem 11.4.1
and Corollary 11.4.1 we can obtain the following results.
Theorem 11.4.2 If there is a positive definite matrix P such that

PUa þ UaT P þ 2kmax ðPÞjjUa  La jjIn \0

then the system in Eq. (11.20) is robust stable (Su and Liu 2009).
Corollary 11.4.2 If

Ua þ UaT
jjUa  La jj   kmax ð Þ ð11:23Þ
2

holds true, then the system in Eq. (11.20) is robust stable. Both Corollaries 11.4.1
and 11.4.2 respectively provide us a meaning result, because Ua  La in fact stands
for the matrix of disturbance errors of the system in Eq. (11.20); Eqs. (11.22) and
(11.23) indicate that when the norm of the disturbance error matrix varies within
the range of ð0; kÞ, the system in Eq. (11.20) will always be stable, where
La þ LTa Ua þ UaT
k ¼ maxfkmax ð 2 Þ; kmax ð 2 Þg.

Theorem 11.4.3 If La þ LTa þ kmax ½ðUa  La Þ þ ðUa  La ÞT In \0, then the system
in Eq. (11.20) is robust stable; if Ua þ UaT  kmax ½ðUa  La Þ þ ðUa  La ÞT In [ 0,
then the system is instable (Su and Liu 2009).
Example 11.4.1 Let us consider the robust stability problem of the following
2-dimensional grey linear system:
 
½2:3; 1:8 ½0:6; 0:9
x_ ðtÞ ¼ xðtÞ
½0:8; 1:0 ½2:5; 1:9

Solution: Through computations, we have

La þ LTa
jjUa  La jj ¼ 0:8072\  kmax ð Þ ¼ 1:3000
2

La þ LTa þ kmax ½ðUa  La Þ þ ðUa  La ÞT In ¼ 1:7759In \0

These inequalities indicate that, by using Corollary 11.4.1 or Theorem 11.4.3,

we can conclude that this given system is robust stable.
11.4 Robust Stability of Grey System 269

11.4.2 Robust Stability of Grey Linear Time-Delay Systems

The phenomena of timely postponing are very common. They are often the main
reason for causing instability, vibration, and poor performance in systems.
Therefore, it is very important to investigate the stability problem of postponing
systems. In particular, let us look at the following n-dimensional linear postponing
autonomous system:
(
x_ ðtÞ ¼ AxðtÞ þ Bxðt  sÞ; 8t  0;
ð11:24Þ
xðtÞ ¼ /ðtÞ; 8t 2 ½s; 0

where xðtÞ 2 Rn stands for the system’s state vector, A; B 2 Rnn the known con-
stant matrices, s [ 0 the amount of time of postponing, and uðtÞ 2 C n ½s; 0 the
nth dimensional space of continuous functions.
Definition 11.4.3 If at least one of the matrices A; B of constants in the linear
postponing system in Eq. (11.24) is grey, then this system is referred to as a grey
linear postponing autonomous system, denoted as
(
x_ ðtÞ ¼ AðÞxðtÞ þ BðÞxðt  sÞ; 8t  0;
ð11:25Þ
xðtÞ ¼ uðtÞ; 8t 2 ½s; 0:

In the following equation, we assume that the constant matrices in the system in
Eq. (11.25) are all grey and have continuous matrix covers; that is, AðÞ; BðÞ
respectively have the following form of matrix covers:

~ : aij  
AðDÞ ¼ ½La ; Ua  ¼ fAðÞ ~  aij ; i; j ¼ 1; 2;    ; ng;

BðDÞ ¼ ½Lb ; Ub  ¼ fBðÞ ~ 

~ : bij   bij ; i; j ¼ 1; 2;    ; ng

aij Þ; La ¼ ðaij Þ; Ub ¼ ðbij Þ; Lb ¼ ðbij Þ.

where Ua ¼ ð
~ BðÞ
Definition 11.4.4 If AðÞ; ~ are respectively whitenization matrices of
AðÞ; BðÞ, then
(
~
x_ ðtÞ ¼ AðÞxðtÞ ~
þ BðÞxðt  sÞ; 8t  0;
ð11:26Þ
xðtÞ ¼ uðtÞ; 8t 2 ½s; 0:

is referred to as a whitenization system of the system in Eq. (11.25).

Definition 11.4.5 If any whitenization system of the system in Eq. (11.25) is
stable, the system in Eq. (11.25) is referred to as robust stable.
270 11 Grey Control Systems

Based on whether or not the robust stability condition of a grey postponing

system depends on the amount of postponing, the robust stability condition can be
divided into two classes: postponing independent and postponing dependent. In
particular, the condition for a robust stable system to be postponing independent is
that for any time postponing s [ 0, the system is robustly asymptotic stable.
Because this condition does not require the amount of postponing, it is appropriate
for the study of the stability problem of postponing systems whose amounts of
postponing are uncertain or unknown.
The condition for a robust stable system to be postponing dependent is that for
some values of postponing s [ 0, the system is robust stable, while for some other
values of the postponing s [ 0, the system is not stable. That is why the system’s
stability is dependent on the amount of postponing.
Theorem 11.4.4 If there are positive definite matrices P; Q and positive constants
e1 ; e2 such that the symmetric matrix
0 1
N PLb P P
B LT P Q þ e2 jjUb  Lb jj2 In 0 0 C
B b C\0
@ P 0 e1 In 0 A
P 0 0 e2 In

where N ¼ LTa P þ PLa þ Q þ e1 jjUa  La jj2 In , and In stands for the identity matrix
(the same symbol will be used for the rest of this chapter), then the system in
Eq. (11.25) is robust stable (Su and Liu 2012).
Theorem 11.4.5 If there are positive definite matrices P; Q; N and positive con-
stants e1 ; e2 such that the symmetric matrix
0 1
C PLb q In P P
B LT P Q þ e2 jjUb  Lb jj2 In 0 0 0 C
B b C
B qIn N  0 C
B 0 0 C\0
@ P 0 0 e1 In 0 A
P 0 0 0 e2 In
pffiffiffi
 ¼ N 1 ; q ¼
where C ¼ LTa P þ PLa þ Q þ e1 jjUa  La jj2 In and N s, then the
system in Eq. (11.25) is robust stable.
Example 11.4.2 Let us look at the following 2-dimensional grey linear postponing
system
(
x_ ðtÞ ¼ AðÞxðtÞ þ BðÞxðt  sÞ; 8t  0;
xðtÞ ¼ uðtÞ; 8t 2 ½s; 0:
11.4 Robust Stability of Grey System 271

Assume that the upper and lower bound matrices of the continuous matrix covers
of the grey constant matrices AðÞ; BðÞ are respectively give as follows:
   
4:38 0:20 4:26 0:29
La ¼ ; Ua ¼ ;
0:19 4:33 0:27 4:22
   
0:93 0:21 0:88 0:24
Lb ¼ ; Ub ¼ :
0:23 0:86 0:26 0:82

According to Theorem 11.4.4, by using the solver in the LMI (linear matrix
inequality) control toolbox, we obtain the behavioral solution as follows:
   
9:4642 0:6983 28:0088 0:0340
P¼ ; Q¼ ;
0:6983 9:8228 0:0340 27:8605
e1 ¼ 30:0826; e2 ¼ 30:2461:

Now from Theorem 11.4.5, by using the solver in the LMI (linear matrix
inequality) control toolbox, we obtain the behavioral solution below:
   
6:8592 0:5061 20:2294 0:0246
P¼ ;Q¼ ;
0:5061 7:1191 0:0246 20:1920
 
0:0456 0
N¼ ;
0 0:0456
e1 ¼ 21:8024; e2 ¼ 21:9209; s ¼ 2:7035:

These results indicate that the system considered in this example is robust stable.
And the maximum allowed postponing length of time as obtained from
Theorem 11.4.5 is 2.7035.

11.4.3 Robust Stability of Grey Stochastic Linear

Time-Delay System

The mathematical model that describes a stochastic system is generally the Itto
stochastic differential equation, where the often seen n-dimensional Itto stochastic
differential postponing equation is
(
dxðtÞ ¼ AxðtÞ þ Bxðt  sÞ þ ½CxðtÞ þ Dxðt  sÞdwðtÞ; 8t  0;
ð11:27Þ
xðtÞ ¼ nðtÞ; nðtÞ 2 L2F0 ð½s; 0; Rn Þ; 8t 2 ½s; 0:
272 11 Grey Control Systems

where xðtÞ 2 Rn stands for the system’s state vector, A; B; C; D 2 Rnn known
constant matrices, s [ 0 the time of postponing, and wðtÞ a 1-dimensional
Brownian motion defined on a complete probability space ðX ; F ; fFt gt  0 ; PÞ.
L2F0 ð½s ; 0; Rn Þ stands for the totality of all F0 -measurable stochastic variables
n ¼ fnðtÞ :  s  t  0g that take values from Cð½s ; 0 ; Rn Þ satisfying
sups  t  0 EjnðhÞj2 \1, while Cð½s ; 0 ; Rn Þ stands for the totality of continuous
functions u : ½s ; 0 ! Rn . Under the initial condition xðtÞ ¼ nðtÞ 2
L2F0 ð½s ; 0 ; Rn Þ, the system in Eq. (11.27) has an equilibrium point xðt ; nÞ, and
corresponds to the initial value nðtÞ ¼ 0, xðt ; 0Þ 0.
There are several different concepts of stability for stochastic systems. In the
following, we list four of the important stabilities.
Definition 11.4.6 The equilibrium point xðtÞ 0 of the system in Eq. (11.27) is
referred to as stochastically stable, if for each e [ 0, lim Pð sup jxðt; t0 ;
x0 !0 t [ t0
x0 Þj [ eÞ ¼ 0.
Definition 11.4.7 The equilibrium point xðtÞ 0 of the system in Eq. (11.27) is
referred to as stochastically asymptotically stable, if it is stochastically stable and
lim Pð lim xðt; t0 ; x0 Þ ¼ 0Þ ¼ 1.
x0 !0 t! þ 1

Definition 11.4.8 The equilibrium point xðtÞ 0 of the system in Eq. (11.27) is
referred to as large-scale stochastically asymptotically stable, if it is stochastically
stable and for any t0 ; x0 , Pð lim xðt; t0 ; x0 Þ ¼ 0Þ=1.
t! þ 1

Definition 11.4.9 The equilibrium point xðtÞ 0 of the system in Eq. (11.27) is
referred to as mean square exponential stable, if there are positive constants
a [ 0; b [ 0 such that Ejxðt; t0 ; x0 Þj2  ajx0 j2 expðbtÞ, t [ t0 .
A grey system is stochastic if it involves grey parameters. Concepts related to
grey stochastic systems are generally introduced based on relevant concepts of
conventional stochastic systems. Considering the problems we will study, let us
provide the following definitions.
Definition 11.4.10 If at least one of the matrices A; B; C; D of the stochastic linear
postponing system in Eq. (11.27) is grey, then the system is referred to as a grey
stochastic linear postponing system, written as follows:
(
dxðtÞ ¼ AðÞxðtÞ þ BðÞxðt  sÞ þ ½CðÞxðtÞ þ DðÞxðt  sÞdwðtÞ; 8t  0;
xðtÞ ¼ nðtÞ; nðtÞ 2 L2F0 ð½s; 0; Rn Þ; 8t 2 ½s; 0:
ð11:28Þ

In this section, we assume that all the coefficient matrices of the system in
Eq. (11.28) are grey with continuous matrix covers. That is, the matrix covers of the
grey matrices AðÞ, BðÞ, CðÞ, and DðÞ are respectively given as follows:
11.4 Robust Stability of Grey System 273

~ ¼ ð
AðDÞ ¼ ½La ; Ua  ¼ fAðÞ ~ aij Þnn : aij  
~ aij  aij g;

~ ¼ ð
BðDÞ ¼ ½Lb ; Ub  ¼ fBðÞ ~ bij  bij g;
~ bij Þnn : bij  

~ ¼ ð
CðDÞ ¼ ½Lc ; Uc  ¼ fCðÞ ~ cij Þnm : cij  
~ cij  cij g;

and

~ ¼ ð
DðDÞ ¼ ½Ld ; Ud  ¼ fDðÞ ~ dij  dij g;
~ dij Þnn : d ij  

where La ¼ ðaij Þnn , Ua ¼ ðaij Þnn , Lb ¼ ðbij Þnn , Ub ¼ ðbij Þnn , Lc ¼ ðcij Þnn ,
Uc ¼ ðcij Þnn , Ld ¼ ðd ij Þnn , and Ua ¼ ðdij Þnn .
Definition 11.4.11 If AðÞ;~ BðÞ;
~ CðÞ,
~ and DðÞ ~ are arbitrary whitenization
matrices of the grey matrices AðÞ; BðÞ; CðÞ, and DðÞ, respectively, then
(
~
dxðtÞ ¼ AðÞxðtÞ ~
þ BðÞxðt ~
 sÞ þ ½CðÞxðtÞ ~
þ DðÞxðt  sÞdwðtÞ; 8t  0;
xðtÞ ¼ nðtÞ; nðtÞ 2 L2F0 ð½s; 0; Rn Þ; 8t 2 ½s; 0:
ð11:29Þ

is referred to as a whitenization system of the system in Eq. (11.28).

Definition 11.4.12 If any whitenization system of the system in Eq. (11.28) is
large-scale stochastic asymptotic stable, that is,

lim xðt; nÞ ¼ 0 a:s:

t!1

then the system in Eq. (11.28) is said to be large scale stochastic robust asymptotic
stable.
Definition 11.4.13 If any whitenization system of the system in Eq. (11.28) is
mean square exponential stable, that is, there are positive constants r0 and K such
that the equilibrium points of whitenization systems of the system in Eq. (11.28)
satisfy

Ejxðt; nÞj2  Ker0 t sup EjnðhÞj2 ; t  0;

s  h  0

or equivalently

1
lim sup log E jxðt; nÞj2   r0 ;
t!1 t

then the system in Eq. (11.28) is said to be mean square exponential robust stable.
274 11 Grey Control Systems

Theorem 11.4.6 For the system in Eq. (11.28), if there is a positive definite
symmetric matrix Q and there are positive constants ei , i ¼ 1 ;    ; 6, satisfying
M þ N\0, then for any initial condition n 2 CFp 0 ð½s ; 0 ; Rn Þ the following holds
true:

lim xðt ; nÞ ¼ 0 a : s:
t!1

That is, according to Su (2012), the system in Eq. (11.28) is large-scale

stochastic robust asymptotic stable, where

M ¼ QLa þ LTa Q þ ðe1 þ e2 ÞQ þ e1 2

1 kmax ðQÞ  jjUa  La jj In þ ð1 þ e4 Þð1 þ e5 ÞLc QLc
T

þ ð1 þ e1 2
4 Þð1 þ e5 Þkmax ðQÞjjUc  Lc jj In

and

N ¼ e1 1 2 1
2 ð1 þ e3 Þkmax ðQÞjjUb  Lb jj In þ e2  ð1 þ e3 ÞLTb QLb þ ð1 þ e1
5 Þð1 þ e6 ÞLd QLd
T

þ ð1 þ e1 1 2
5 Þð1 þ e6 Þkmax ðQÞjjUd  Ld jj In :

Theorem 11.4.7 For the system in Eq. (11.28), if there are positive definite sym-
metric matrix Q and positive constants ei , i ¼ 1 ;    ; 6, satisfying K þ L\0, then
for any initial condition n 2 CFp 0 ð½s ; 0 ; Rn Þ, the following holds true:

lim xðt ; nÞ ¼ 0 a : s:
t!1

That is, the system in Eq. (11.28) is large-scale stochastic asymptotic stable,
where

K ¼ QLa þ LTa Q þ ðe1 þ e2 ÞQ þ ½e1

1 kmax ðQÞtraceðGa Ga Þ þ ð1 þ e4 Þð1 þ e5 ÞtraceðLc Lc Þ
T T

þ ð1 þ e1
4 Þð1 þ e5 Þkmax ðQÞtraceðGc Gc ÞIn ;
T

and

L ¼ ½e1 1 1
2 ð1 þ e3 Þkmax ðQÞtraceðGb Gb Þ þ e2 ð1 þ e3 ÞtraceðLb Lb Þ
T T

þ ð1 þ e1 1 1
5 Þð1 þ e6 ÞtraceðLd Ld Þ þ ð1 þ e5 Þð1 þ e6 Þkmax ðQÞtraceðGd Gd ÞIn :
T T

If we let the matrix and constants in Theorems 11.4.6 and 11.4.7 be e1 ¼    ¼

e6 ¼ 1 and Q ¼ In , then we can obtain the following corollaries, respectively.
Corollary 11.4.3 If the upper and lower bound matrices of the continuous matrix
covers of the coefficient matrices of the system in Eq. (11.28) satisfy
11.4 Robust Stability of Grey System 275

La þ LTa þ 2LTb Lb þ 4LTc Lc þ 4LTd Ld

\  ð2jjUb  Lb jj2 þ jjUa  La jj2 þ 4jjUd  Ld jj2 þ 4jjUc  Lc jj2 þ 2ÞIn

then the system in Eq. (11.28) is large-scale stochastic asymptotic stable.

Corollary 11.4.4 If the upper and lower bound matrices of the continuous matrix
covers of the coefficient matrices of the system in Eq. (11.28) satisfy

La þ LTa þ ½2traceðLTb Lb Þ þ 4traceðLTc Lc Þ þ 4traceðLTd Ld ÞIn

\  ðtraceðGTa Ga Þ þ 2traceðGTb Gb Þ þ 4traceðGTc Gc Þ þ 4traceðGTd Gd Þ þ 2ÞIn

then the system in Eq. (11.28) is large-scale stochastic asymptotic stable.

Theorem 11.4.8 For the system in Eq. (11.28), if there are positive definite sym-
metric matrix Q and positive constants ei , i ¼ 1 ;    ; 3, satisfying

QLa þ LTa Q þ ðe1 þ e2 ÞQ þ e1 2

1 kmax ðQÞjjUa  La jj In
\  ½ð1 þ e3 Þkmax ðQÞtraceðMcT Mc Þ þ e1
2 kmax ðQÞtraceðMb Mb Þ
T

þ ð1 þ e1
3 Þkmax ðQÞtraceðMd Md ÞIn
T

then the system in Eq. (11.28) is large-scale stochastic robust asymptotic stable. If
in Theorem 11.4.8 we let e1 ¼ e2 ¼ e3 ¼ 1 and Q ¼ In , then we have the corollary
below.
Corollary 11.4.5 If the upper and lower bound matrices of the matrix covers of the
grey coefficient matrices in the system in Eq. (11.28) satisfy

La þ LTa þ 2In þ jjUa  La jj2 In þ 2traceðMcT Mc ÞIn

\  ½traceðMbT Mb Þ þ 2traceðMdT Md ÞIn

then the system in Eq. (11.28) is large-scale stochastic robust asymptotic stable.
Theorem 11.4.9 For the system in Eq. (11.28), if there are positive definite sym-
metric matrix Q and positive constants ei , i ¼ 1 ;    ; 6, satisfying
kmax ðMÞ þ kmax ðNÞ\0, then for any initial condition n 2 CFp 0 ð½s ; 0 ; Rn Þ, the
following holds true:

Ejxðt; nÞj2  Ker0 t sup EjnðhÞj2 ; t  0;

s  h  0

or equivalently,
276 11 Grey Control Systems

1
lim sup log Ejxðt; nÞj2   r0 :
t!1 t

Where the matrices M; N are the same as in Theorem 11.4.6,

r0 s
ðNÞ þ kmax ðQÞ
K ¼ s e kmaxkmin ðQÞ , and r0 is the unique real root of the following equation
r0 kmax ðQÞ þ kmax ðMÞ þ er0 s kmax ðNÞ ¼ 0, then the system in Eq. (11.28) is mean
square exponential robust stable.

11.5 Typical Grey Controls

Grey control stands for the control of essential grey systems, including the situation
of general control systems involving grey numbers, by constructing controls
through employing the thinking methods of grey systems analysis, modeling,
prediction, and decision-making.

11.5.1 Control with Abandonment

The dynamic characteristics of grey systems are mainly determined by the matrices
GðsÞ of grey transfer functions. So, to realize effect control over the systems’
dynamic characteristics, one of the effective methods is to modify and correct the
matrices of transfer functions and the structure matrices.
Definition 11.5.1 Assume that G1 ðsÞ is a system’s structure matrix, and G1

ðsÞ
an objective structure matrix, then

D1 ¼ G1 1

ðsÞ  G ðsÞ ð11:30Þ

is known as a structural deviation matrix. From G1 ðsÞYðsÞ ¼ UðsÞ and

1 1
G1

ðsÞ ¼ D þ G1 ðsÞ, we obtain ðG1

ðsÞ  D ÞYðsÞ ¼ UðsÞ. That is,

1
G1

ðsÞYðsÞ  D YðsÞ ¼ UðsÞ ð11:31Þ

Definition 11.5.2 D1 YðsÞ is referred to as a superfluous term. The control

through a feedback of D1 YðsÞ to cancel the superfluous term is known as a control
with abandonment (Deng 1965). Through the effect of the feedback of D1 YðsÞ, the
system G1 ðsÞYðsÞ ¼ UðsÞ is reduced to

G1 ðsÞYðsÞ þ D1 YðsÞ ¼ UðsÞðG1 ðsÞ þ D1 ÞYðsÞ ¼ UðsÞ

11.5 Typical Grey Controls 277

Fig. 11.8 The control with abandonment

That is, G1

ðsÞYðsÞ ¼ UðsÞ has already processed the desired objective
structure.
The number of entries in the structural deviation matrix D1 , used in a control
with abandonment, directly affects the number of components in the controlling
equipment. So, when considering the economics, reliability, and ease of application
of a dynamic system, one must keep the number of elements in the deviation matrix
D1 as low as possible. That is to say, in the objective structural matrix, one should
try to keep the corresponding entries of the original structure matrix. The idea of
control with abandonment is depicted in Fig. 11.8.

11.5.2 Control of Grey Incidence

Definition 11.5.3 Assume that Y ¼ ½y1 ; y2 ;    ; ym T stands for the output vector,
and J ¼ ½j1 ; j2 ;    ; jm T the objective vector. If the components of the control
vector U ¼ ½u1 ; u2 ;    ; us T satisfy

uk ¼ fk ðcðJ; YÞÞ; k ¼ 1; 2;    ; s ð11:32Þ

278 11 Grey Control Systems

Fig. 11.9 The control system of grey incidence

where cðJ; YÞ is the degree of grey incidence between the output vector Y and the
objective vector J, then the system control is known as a control of grey incidence.
A grey incidence control system is obtained by attaching a grey incidence
controller to the general control system. It determines the control vector U through
the degree of grey incidence cðJ; YÞ so that the degree of incidence between the
output vector and the objective vector does not go beyond a pre-determined range.
The idea of the control system of grey incidence is depicted in Fig. 11.9.

11.5.3 Control of Grey Prediction

All the various kinds of controls studied earlier are about applying controls after
first checking whether or not the system’s behavioral sequence satisfies some
pre-determined requirements. Such post-event controls evidently suffer from the
following weaknesses:
(1) Expected future disasters cannot be prevented;
(2) Instantaneous controls cannot be done; and
(3) Adaptability is weak.
The so-called grey predictive control is designed based on the system’s future
behavioral tendency, which is predicted using the system’s behavioral sequences
and the patterns discovered from the sequences. This kind of control can be
employed to avoid future adverse events from happening; it can be implemented in
a timely fashion, and possesses a wide range of applicability.
The idea of a grey predictive control system is graphically shown in Fig. 11.10.
Its working principle is that first one must collect and organize the device’s
behavioral sequence of the output vector Y; secondly, one must use a prediction
device to compute the predicted values for the future steps; and lastly, one must
compare the predicted values with the objective and determine the control vector U
so that the future output vector Y will be as close to the objective J as possible.
11.5 Typical Grey Controls 279

Fig. 11.10 Grey predictive control

Definition 11.5.4 Assume that ji ðkÞ; yi ðkÞ; ui ðkÞ ði ¼ 1; 2;    ; mÞ are respec-

tively the values of the objective component, output component, and control
component at time moment k. For i ¼ 1; 2;    ; m, let

ji ¼ ðji ð1Þ; ji ð2Þ;    ; ji ðnÞÞ

yi ¼ ðyi ð1Þ; yi ð2Þ;    ; yi ðnÞÞ
ui ¼ ðui ð1Þ; ui ð2Þ;    ; ui ðnÞÞ

For the control operator f : ðji ðkÞ; yi ðkÞÞ ! ui ðkÞ,

ui ðkÞ ¼ f ðji ðkÞ; yi ðkÞÞ ð11:33Þ

when k [ k, the system is known as a post-event (or after-event) control; when

k ¼ k, the system is known as an on-time control; and when k\k, the system is
known as a predictive control.
Definition 11.5.5 If the control operator f satisfies

f ðji ðkÞ; yi ðkÞÞ ¼ ji ðkÞ  yi ðkÞ ð11:34Þ

That is,

ui ðkÞ ¼ ji ðkÞ  yi ðkÞ ð11:35Þ

then when k [ k, the system is known as an error-afterward control; when k ¼ k,

the system is known as an error-on-time control; and when k\k, the system is
known as an error-predictive control.
Definition 11.5.6 Let yi ¼ ðyi ð1Þ; yi ð2Þ;    ; yi ðnÞÞði ¼ 1; 2;    ; mÞ stand for a
sample of output components and its GM (1, 1) response formula be given as
follows:
8
>
< ^yð1Þ ðk þ 1Þ ¼ ðyi ð1Þ  bi Þeai k þ bi
i
ai ai
>
: ^yð0Þ ðk þ 1Þ ¼ ^yð1Þ ðk þ 1Þ  ^yð1Þ ðkÞ
i i i
280 11 Grey Control Systems

If the control operator f satisfies

ð0Þ
ui ðn þ k0 Þ ¼ f ðji ðkÞ; yi ðkÞÞ; n þ k0 \ki ; i ¼ 1; 2;    ; m ð11:36Þ

then the system control is known as a grey predictive control.

In a grey predictive control system, predictions are often done using metabolic
models. So, the parameters of the prediction device vary with time. When a new
data value output is produced and accepted by the sampling device, an old data
value is removed so that a new model is developed. Accordingly, a series of new
predicted values are provided. Doing so guarantees the strong adaptability of the
system.
Example 11.5.3 Let us look at the EDM (electric discharge machining) grey control
system.(Yang and Zheng 1996). The investigation on the control systems of EDM
machines has been an important effort in the field of electric discharge machining.
Each EDM can be seen as a stochastic time-dependent nonlinear system involving
many parameters. Applications mainly include those situations when the conven-
tional controls of linear, constant coefficient systems cannot produce adequate
outcomes. The current commonly employed EDM control systems are established
based on modern control theory. The frequently applied self-adaptive control sys-
tems generally employ mathematical models of approximation with accompanied
high costs without actually realizing optimal results. Grey control is not like precise
mathematical models based on complete knowledge of a system as addressed in
modern control theory. It is also unlike fuzzy control, where the system is treated as
a black box as all the information about the internal working of the system is
disregarded, which leads to low accuracy controls. The parameters, structures, and
other aspects of grey models vary with time. Such dynamic modeling can be highly
appropriate for the study of EDM machines with high degrees of uncertainty and
produce relatively more satisfactory control effects.
For EDM control systems, the objects of control are EDM machine tools, where
outputs need signals from the testing of EDM machine tools as well as the control
quantity U, that is, the signals about the control of the EDM machine tools. EDM
control systems, in general, mean the control over systems that serve EDM machine
tools. For instance, let us look at the traditional gap-voltage feedback servo control
system. Due to a lack of linear relationship between the gas voltage, gap size,
discharge strength, discharge state, and servo reference voltage, the effect of
employing only one gap-voltage feedback servo control system is not very good.
In order to make up for the insufficiency of single loop controls, one can employ
double-loop controls with the inner loop being the traditional gap-voltage feedback
control and the outer loop being an impulse discharge rate feedback control that
instantaneously adjusts the inner loop. The block-design chart of this control system
is depicted in Fig. 11.11. Figure 11.11 shows that this control design represents a
system of two loops. Based on the collected sequence of gap voltage readings
Ug ðKÞ, the inner loop employs the GM model to predict the next moment
U^ g ðK þ i þ 1Þ. Here, i stands for the prediction steps, which are then fed into the
11.5 Typical Grey Controls 281

Fig. 11.11 EDM control system

input end to determine the servo reference voltage value Us , which is a propor-
tionality coefficient. The outer loop establishes a GM model based on a sequence
^ þ i þ 1Þ. When these predicted
YðKÞ of output values to predict the next steps YðK
values are compared with requirements Y , a sequence eðKÞ ¼ Y^  Y
of errors is

found. These error values are then fed back into the system to adjust the propor-
tionality coefficient K1 and the servo reference voltage Us , in order to adjust the
inner loop. That is,

^ Us ¼ K2 U
DU ¼ K1 ðY
 YÞ; ^ g  DU

^ g  K1 ðY
 YÞ,
Therefore, Us ¼ K2 U ^ where parameters K1 ; K2 are determined
by experiments.
Example 11.5.4 Let us now look at the grey predictive control for the vibration of a
rotor system (Zhu and Zhi 2002). The theory and methods for active vibration
control of rotors have caught more attention in recent years. Many new control
theories, such as neural network theory, time-delay theory, self-learning theory,
fuzzy theory, and H∞ theory have gradually been employed in research on active
control theory of rotors, leading to some good outcomes. For a Jeffcott symmetric
rotor follower system with an electromagnetic damper as its executor, we employ
control theory and methods of grey predictions to investigate an active amplitude
control of vibration. We first establish a grey predictive control module with the
GM (1, 1) as its main component. In the vibration control system of the rotor, let
I 0 ðkÞ and x0 ðkÞ, k ¼ 1; 2; . . . ; n, respectively be electric current inputting into the
electromagnetic damper and the corresponding maximum output amplitude of the
rotor vibration. By employing the available experimental measurement results from
the literature, we obtain a set of data of I 0 ðkÞ and the relevant x0 ðkÞ, as shown in
Table 11.1, when the sensitivity of the transducer is 104 V/m.
282 11 Grey Control Systems

Table 11.1 Sampled data of I 0 ðkÞ and x0 ðkÞ when the transducer’s sensitivity is 104 V/m
I 0 ðkÞ (A) 0.1 0.125 0.175 0.225 0.325
x0 ðkÞ (dm m) 1.4 1.35 1.2 0.9 0.65

Based on the mechanism of the GM (1, 1) model, we establish the following

modification model of the system based on the errors of the grey predictions:
0
að0Þ ðk þ 1Þ ¼ a½xð0Þ ð1Þ  beak þ dðk  iÞða0 Þ½qð0Þ ð1Þ  b0 ea k
^

where a ¼ 0:1862; xð0Þ ð1Þ ¼ 1:4; b ¼ 9:3298; a0 ¼ 0:14; qð0Þ ð1Þ ¼ 0:36; b0 ¼ 3:78,
and
(
1 ki
dðk  iÞ ¼
0 k\i

The design of our grey predictive control of the rotor system is shown in
Fig. 11.12.
In this control system, the displacement signal of the rotor system is measured by
the eddy current transducer. The sampling equipment collects the data from the
amplitude recorder, and through the effect of the grey prediction controller, con-
trolling voltage is produced. This voltage is transformed into a controlling electric
current through the current amplifier. Then, when this electric current flows through
the stator coil of the electromagnetic damper, an electromagnetic force is created,
which in turn controls the amplitude of vibration of the rotor within the expected
range so that the system’s stability is achieved.
For this grey predictive control system developed for the vibration of the said
Jeffcott symmetric rotor follower, our computer simulation, when compared to the

Fig. 11.12 Grey predictive control of the rotor system

11.5 Typical Grey Controls 283

Table 11.2 Simulations and actual measurements

I(A) X1m (m) X2m (m) X3m (m) e12 (%) e13 (%) e23 (%)
0.1 1.41  10−4 1.4  10−4 1.4  10−4 0.71 0.71 0
0.125 1.27  10−4 1.227  10−4 1.35  10−4 3.5 −5.93 −9.11
0.175 1.03  10−4 0.949  10−4 1.2  10−4 9 −13.8 −20.92
0.225 0.83  10−4 0.745  10−4 0.9  10−4 11.4 −7.78 −17.22
0.325 0.55  10−4 0.46  10−4 0.65  10−4 19.57 −13.38 −29.23

physical measurements of the amplitudes, indicates that the maximum amplitudes

under the control are only about 7 % of those physically observed without the
control imposed on the rotor system. Table 11.2 respectively provides the results of
the maximum amplitudes of two separate computational simulations and the actual
measurements X1m , X2m , and X3m , along with the change in the static electricity i of
the electromagnetic damper, when the sensitivity of the transducer is k1 = 104 V/m,
and the corresponding errors e12, e13, and e23 between X1m and X2m , X1m and X3m ,
and X2m and X3m .
Chapter 12
Introduction to Grey Systems Modeling
Software

12.1 Introduction

In 1982, Julong Deng initiated and established grey systems theory. Currently, grey
systems theory is widely applied in areas such as social sciences, economics,
agriculture, meteorology, military and science, providing solutions to a large
number of practical problems and challenges met in everyday life. Various versions
of grey systems modeling software have played a very important role in such large
scale practical applications of grey systems theory. Along with the rapid devel-
opment of information technology, high level programming languages have grad-
ually matured, applications of computing packages have been routinized, and grey
systems modeling programs have also become sophisticated.
In 1986, Xuemeng Wang and Jiangjun Luo created their grey systems modeling
software using BASIC language and published Programs of Grey Systems’
Prediction, Decision-Making, and Modeling. In 1991, Xiuli Li and Ling Yang
respectively developed grey modeling software using GWBASIC and Turbo C. In
2001, Xuemeng Wang, Jizhong Zhang, and Rong Wang published the book entitled
“Computer Procedures for Grey Systems Analysis and Applications,” in which they
listed the structure and procedure codes established for grey modeling. All of these
computer software packages were developed on the DOS platform and have
become obsolete in the more user-friendly Windows framework.
In 2003, Dr. Bing Liu developed the first grey systems modeling software for
Windows using VisualBasic6.0. As soon as this package was available, it was most
welcomed in the community of scholars and practitioners of the grey systems
research, and became the first choice of application in the field of grey systems
modeling. With the evolution of software development technology, changes in
computer operation, and continued grey theory research progress, some of the
weaknesses of such software packages came to the fore, including the following:

© Springer Science+Business Media Singapore 2017 285

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1_12
286 12 Introduction to Grey Systems Modeling Software

• Data entry was tedious.

The single worksheet frame limited the operational flexibility on data sequences.
It was especially inconvenient when large amounts of data were dealt with in
clustering analysis and grey decision-making. Additionally, the only available way
for data entry made users feet tired so that the efficiency and accuracy of data entry
were greatly affected.
• The classification of the modules was not scientifically sound.
This software system divided modules according to the number of data partic-
ipating in the modeling process. However, the modules should have been designed
according to functions.
• The software system could not show relevant computational procedures.
Most of the users of grey systems theory are scientists and practitioners. The
purpose of their use of the software system was mainly their scholarly works. This
means that, other than the computational outcomes, such professionals are also very
interested in knowing specific procedural details. However, the original software
package could only provide the final results of the computation and was unable to
reveal relevant computational details.
• The system’s capability was disconnected from the most recent research.
Grey systems theory has been an extremely active area of the broader systems
science. In particular, in recent years some works with high practical value have
appeared. However, the software system was not upgraded along with research pro-
gress, causing the software to become obsolete in terms of its functions and capability.
• Problems with the choice of package development.
VisualBasic6.0 is a graphics-based software development tool created by
Microsoft Company. Due to its simplicity, functionality, versatility, and other
strengths, as soon as it was introduced it was welcomed by many software devel-
opers. However, VisualBasic6.0 is an IDE (integrated development environment)
based on BASIC, a typical programming language with many known weaknesses
that greatly limit its applicability in scientific computing. This is because scientific
computing requires high levels of accuracy. Thus, grey systems modeling software
developed using VisualBasic6.0 inherently suffers from many weaknesses.

12.2 Software Features and Functions

On one hand, an ideal grey systems modeling software package needs to have the
computational power to handle practical models, and on the other hand it has to deal
with user confirmation, registration, and other functionalities. The software system
accompanying this book sufficiently combines the capabilities of the C/S
12.2 Software Features and Functions 287

(client/server) and B/S (Browser/Server) modules, where the C/S part completes
computational functions, while the B/S part handles the relevant operations that
serve the user and his communication with the server. With a view to improve
existing systems, the design of this package focuses more on the reliability, prac-
ticality, compatibility, upgradability, accuracy, operational convenience, visual
appeal and user friendliness. This package has the following characteristics:
• Data entry is convenient and fast.
For data sequences of the same kind, the package provides a rectangular window
into which the user can simply copy the sequences with one operation. For grey
clustering and grey decision-making modules that involves large amounts of data, it
is evidently inconvenient to employ the traditional way of entering data values. In
such instances, the user can enter data in an Excel document and then open the data
file into this package system. This software system makes use of the powerful data
entry ability of Excel while making data entry convenient for the user.
• Modules are designed according to functionalities.
In software engineering, a module is a relatively independent system unit of
procedures. Each such unit of procedures handles and materializes a relatively
independent task. In other words, it contains a group of independent procedures.
Each program module has its own external characteristics, such as its own name,
label, and interfaces. During the design of this software package, the developer
scientifically organized the contents of grey systems theory, defined the relevant
functions and related modules.
• This system provides operational details as well as periodic results.
For modules with complicated computational procedures where intermediate
results are also important, the system provides a textbox that can store and show
multi-line operational details. The user can monitor data changes in each compu-
tational step so that he can further understand how the model operates. Also, the
software interface provides relevant information to remind the user of the relevant
formulas employed in the model.
• The functionalities of the modules are greatly expanded.
Based on current practical applications of grey systems theory, combined with
the most recent research results, this software system is the most up-to-date system
available in the market. It includes: weakening operators (mean weakening buffer
operators, geometric mean weakening buffer operators, weighted mean weakening
buffer operators), strengthening operators (mean strengthening buffer operators,
geometric mean strengthening buffer operators, weighted mean strengthening buffer
operators), grey incidence analysis (relative degree of incidence, closeness degree
of incidence), clustering analysis (based on center-point triangular whitenization
weight functions), grey prediction (GM (1, n) and DGM (1, 1) models), grey
decision analysis (intelligent grey target decision making), among other contents.
288 12 Introduction to Grey Systems Modeling Software

• The degree of accuracy of the computational results can be adjusted.

The computation precision of different systems is different. In this software
system there is a ComboBox, which can select of computational precision.
Therefore, the user can choose the desired degree of accuracy for his work.
• The operation of the software system is convenient and easy to learn.
This software system is based on the Windows interface using pull-down menus,
where the commonly employed modeling techniques of grey systems theory are
effectively gathered together. The user only needs to have an elementary under-
standing of how a desktop PC works to successfully use this software system. At
the same time, this system has a relatively strong ability to locate and correct
mistakes. When an illegal operation is performed, the system will provide an
accurate and detailed hint.
• This system is developed using Visual C#.
C# is an object-oriented programming language created by Microsoft and an
important part of Microsoft’s .NET development environment. Also, Microsoft
Visual C# is an integrated development environment (IDE) constructed on C# by
Microsoft. It is designed for the operation of many application software packages
created on the .NET framework. C# possesses powerful capabilities, type safety,
object orientation, and other superb functions. It is currently the main development
tool of C/S software architecture.

12.3 Main Components

The new edition of the grey system modeling software consists of five modules
including grey sequence operators, grey incidence analysis models, grey clustering
evaluation models, grey forecasting models and grey models for decision-making,
given the currently available research on grey systems theory and its practical
applications. The basic constitution of the grey system modeling software is shown
in Table 12.1, and the software system modules are shown in Tables 12.2, 12.3,
12.4, 12.5 and 12.6.

Table 12.1 The basic constitution of the grey system modeling software
Grey system modeling software B/S part User info
Statistics
C/S part Grey sequence operators
Grey incidence analysis
Grey clustering evaluation
Grey forecasting
Grey decision-making
12.3 Main Components 289

Table 12.2 Grey sequence operators

Grey sequence Weakening operators Average weakening buffer operator (AWBO)
operators Weighted average weakening buffer operator
(WAWBO)
Geometric average weakening buffer operator
(GAWBO)
Weighted geometric average weakening buffer
operator (WGAWBO)
Strengthening Even strengthening buffer operator (ESBO)
operators Average strengthening buffer operator (ASBO)
Weighted average strengthening buffer operator
(WASBO)
Information mining Accumulating generation operator
operators Inverse accumulating generation operator
Even operator by adjacent neighbor
Operator of stepwise ratio

Table 12.3 Grey incidence analysis models

Grey incidence analysis models Deng’s model of degree of grey incidence
Absolute degree of grey incidence
Relative degree of grey incidence
Synthetic degree of grey incidence
Closeness degree of grey incidence
Similitude degree of grey incidence

Table 12.4 Grey clustering evaluation models

Grey clustering evaluation Grey clustering model of variable weight
models Grey clustering model of fixed weight
Grey clustering model using center-point mixed triangular
possibility function
Grey clustering model using end-point mixed triangular
possibility function

Table 12.5 Grey forecasting models

Grey forecasting models Singular variable models Even GM (1, 1)
Original difference GM (1, 1)
Even difference GM (1, 1)
Discrete grey model
Grey Verhulst model
Multi-variable models Model GM (0, N)
Model GM (1, N)
290 12 Introduction to Grey Systems Modeling Software

Table 12.6 Grey models for decision-making

Grey models for decision-making Weighted multi-attribute grey target decision
Two stages model for decision-making

12.4 Operation Guide

12.4.1 The Confirmation System

To verify legal ownership, the user needs to enter his account number and password
before he can actually start using the system. However, if every time the user uses
the system he has to confirm his legal ownership of the software package, it will
become an annoyance. So, to guarantee the legal ownership of the user and
maintain the operational simplicity of the system, the system applies the
XML-based client programming technique. When the user attempts to run the
program for the first time, the system will prompt him to provide the needed
account number and password. The provided data will then be delivered through the
internet to the database located at the server to verify their legality. When the user
attempts to use the program on different, subsequent occasions, he can directly enter
the main interface window without having to enter his account number and pass-
word again.
On the first time of confirmation, if the user does not have an account number or
password, he needs to click on the “User registration” button (see Fig. 12.1) to
register for a free user account (B/S). If the user forgets his password, he can click
on the “Recall password” button to retrieve his password. Figure 12.2 is the flow
chart of confirmation.

Fig. 12.1 The confirmation window

12.4 Operation Guide 291

Fig. 12.2 The confirmation flow chart

12.4.2 Using the Software Package

After successful confirmation of legal ownership, the user will enter the system’s
main interface window, as shown in Fig. 12.3. Various grey systems theory
modules (and their sub-modules) are administrated through menus.
Figure 12.4 provides the flow chart of various sub-modules of the system.
Data Entering Before running the program, one needs to first enter data into the
software package and specify the system parameters. As mentioned earlier, there
are two ways to input data. One can directly enter data into the provided text box or
import data from an external Excel document. For those modules that require large
amounts of data, the only way provided for entering data is through importing data
292 12 Introduction to Grey Systems Modeling Software

Fig. 12.3 The main interface window

Fig. 12.4 Sub-modules

from Excel documents. The following sections look at the details of these two data
entering methods.
• Enter data directly into the provided text box.
With VisualC#, there are two kinds of controllers available for direct data entry.
One is the TextBox controller, and the other the ComboBox controller. The former
controller is used to develop the standard Windows’ editing controller of the
text-box, which is used to acquire the user’s input or show what is already stored in
the storage space. When entering data into the text-box, right click the mouse inside
the text box. When the cursor blinks inside the text box, one can start entering data.
12.4 Operation Guide 293

The ComboBox of the Windows’ window group is mainly used to show data in a
down-drop list box. As a default, ComboBox consists of two parts: the top is a text
box in which the user is allowed to enter data, and the bottom is a list box where the
user can make selections. It is because the ComboBox consists of the text box on
the top and the list box at the bottom that it is named a ComboBox. When using the
ComboBox to enter data, the user needs to first check whether or not the list box
contains the data he wants. If so, he can simply use the mouse to directly make the
selection; otherwise, he needs to enter data into the text box on the top. The detailed
procedure for entering data in the ComboBox is similar to that of operating the
TextBox and is therefore omitted here.
Note: When entering data using either the TextBox or the ComboBox, the state
of entry method needs to be adjusted to half-angle. Data values entered in the state
of full angle will be treated by the program as illegal data, which will directly affect
the normal operation of the program and potentially lead to unexpected outcomes.
• Import data from an Excel document.
Both the TextBox and ComboBox can only accept small amounts of data values.
For entering large sums of information, the use of either the Textbox or the
ComboBox is inefficient, and can also lead to errors. To resolve the problem of
entering large sums of data values when dealing with grey clustering and grey
decision-making, for instance, it is very often the case that large amounts of
information are involved, and this software system makes use of the powerful
Excel. First enter and edit the needed data in Excel, and then use the provided
interface to import the Excel data into the software system. Excel is one of the
components of Microsoft Office. It is a tabulated testing and computing software
developed for Microsoft Windows and Apple’s Macintosh. Its straightforward
interface, excellent capabilities of computation and graphics make it the most
widely employed PC software used for data analysis. Through Excel, our software
package system can conveniently acquire data.
Each Excel document generally contains three tables, respectively labeled as
Sheet 1, Sheet 2, and Sheet 3. When an Excel document opens, it generally shows
Sheet 1. When entering data according to the system’s requirements, one can directly
type in the corresponding values in the relevant rows and columns. Upon finishing
data entry into the Excel document, one can employ our system’s input function to
import the Excel data. When importing an Excel data document, first select the path
from which the Excel file is located. As soon as the importing path and location of
the file is confirmed, the data will be successfully imported. In fact, the process of
importing data connects the Excel file to our system through a specific path so that
the data in the Excel file can be mapped into the database controller DataGridView.
DataGridView is a database controller of VisualC#, which can exactly and
entirely reveal data from a source file. Through the DataGridView controller of
VisualC#, data can be acquired from an Excel file. However, our system does not
provide any of the editing capabilities of DataGridView. In other words, if it is
found in DataGridView that there is an error in the data, this error cannot be
294 12 Introduction to Grey Systems Modeling Software

corrected directly within DataGridView. Instead, one has to return to the original
Excel file to make the correction and then reimport the entire corrected file back
into the grey systems modeling package.
Notes:
• The DataGridView controller does not have any editing capability. To make
changes in the data, one has to do it in the original Excel file.
• When entering data into an Excel document, one needs to do so in the mode of
“half-angle.” All data entered in the mode of “full-angle” will be treated as
illegal entries, which will directly affect the normal operation of the grey sys-
tems modeling package, and potentially lead to unexpected outcomes.
• The table names of the Excel file have to be the default Sheet 1, Sheet 2 and/or
Sheet 3 without any modification, otherwise the import of data will be affected.
• The data entry field of Excel is very large. However, one often needs only a few
rows and columns. Make sure that there are no symbols or blank cells accidently
entered into other area of the field. Otherwise, the data transfer will be affected.

Model Computations
(1) Grey sequence operators
Click on “Sequence generation.” From the pull-down menu that appears, select the
module according to the practical modeling need. The corresponding detailed
modeling interface appears. Let us use the “average weakening buffer operator” as
an example to illustrate how to apply grey sequence generations. What is shown in
Fig. 12.5 is the interface of the mean weakening buffer operator.

Fig. 12.5 The interface of the mean weakening buffer operator

12.4 Operation Guide 295

Fig. 12.6 The required Excel file format

This interface window contains three main areas: the first shows the original data
sequence, which is the area for data entry or importing data; the second area shows
the “order and outcome precision,” in which it is possible to adjust the order of the
operator being applied and the corresponding precision of the computational out-
puts based on one’s modeling needs; and the third the area is where computational
results are shown. After the data entry is completed, click on the “mean weakening
buffer operator (AWBO)” button. Immediately, the generated sequence will appear
in the generated sequence window. Figure 12.6 shows a work sheet of an Excel
document. When applying this Excel capability and importing data from Excel, the
user has to follow this shown format exactly.
(2) Grey incidence analysis models
Similar to the generation of grey sequences, there are two ways to input data for all
parts of incidence analysis, so such data entry details are omitted here. However,
this is not valid for Deng’s degree of grey incidence due to the need for a large
amount of data. For Deng’s degree of grey incidence, this software system allows
only data entry through Excel documents without the option of direct data entry.
Figure 12.7 shows the editing format of an Excel document, while Fig. 12.8 shows
the complete work interface.
(3) Grey clustering evaluation models
Similar to Deng’s degree of grey incidence, grey clustering evaluation also requires
a large amount of original data. However, in grey clustering evaluation it is possible

Fig. 12.7 The required Excel file format

296 12 Introduction to Grey Systems Modeling Software

Fig. 12.8 The complete work interface of Deng’s degree of grey incidence

to have several different types of data, including objective-criterion data, possibility

functions, and criteria weights. Therefore, for grey clustering evaluation, the system
again provides only one way to enter data, which is by importing Excel files. The
key to using this part of the functions is to correctly edit the different types of data
in the Excel documents. Sheet 1 contains the objective-criteria data (Fig. 12.9),
Sheet 2 the corresponding possibility functions (Fig. 12.10), and Sheet 3 the
weights of the criteria (Fig. 12.11).
Figure 12.12 shows the operating interface of a grey clustering analysis. As for
how to apply grey variable weight clustering and analysis based on center-point
mixed triangular possibility functions, the operational details are similar and
therefore omitted.
(4) Grey prediction models
Grey prediction models stand for an important part of grey systems theory. The
operation of each individual prediction model is roughly the same. So, let us use the
EGM (1, 1) model to illustrate how to use the software system. The main steps
include: enter or import data; click the “computation, simulation, prediction” button
to compute the model parameters and the simulated values, and select the simu-
lation accuracy; enter the desired number of predicted values, then click “prediction
results.” Figure 12.13 shows the operational interface of the EGM (1, 1) model.
(5) Grey decision-making
This part of the software package contains two modules, namely the multi-attribute
grey target decision-making model and the two-stage model for decision-making.
12.4 Operation Guide 297

Fig. 12.9 The objective-criteria data

Fig. 12.10 The corresponding possibility functions

Fig. 12.11 The weights of the criteria

298 12 Introduction to Grey Systems Modeling Software

Fig. 12.12 The operating interface of a grey clustering evaluation

Fig. 12.13 The operational interface of the GM (1, 1) model

12.4 Operation Guide 299

Fig. 12.14 The exact layout of data page for multi-attribute grey target decision-making

Fig. 12.15 The entire operating page of multi-attribute grey target decision-making

The data layout of an intelligent grey target decision-making model is the same
as that of any synthesized objective decision-making model, except that there is an
additional column of threshold value, as shown in Fig. 12.14, where the interval
[14, 18] means that the lower effect threshold value is 14 and the upper effect value
is 18. Figure 12.15 shows the entire operating page of an intelligent grey target
decision-making model.
Farewell to Our Tutor

The memorial speech by Sifeng Liu on behalf of students of Prof. Deng

At 12:15 afternoon on June 22, 2013, our most beloved tutor, Professor Deng
Julong, saw the end of his 80-year life journey and left us forever. Professor Deng
Julong was a tireless lifelong pioneer and founder of Grey System Theory, so the
world has lost a visionary. In recent days, dark clouds gathered and there was a long
period of wet weather: it was God’s crying for the death of a great scholar.
Prof. Deng graduated from the Department of Electrical Engineering of the
Huazhong Institute of Technology in 1955 and then taught at the Department of
Automatic Control. In the 1960s, he put forward the idea of control with aban-
donment. In the 1970s, the method of control with abandonment became a typical
control method internationally. In 1982, he pioneered Grey System Theory and
created a brand new subject area in the history of science. His academic achieve-
ments were highly respected by many in the scientific circles.
In Prof. Deng’s academic career of 60 years, there existed neither holidays,
weekends nor a line between service and retirement. Prof. Deng was the editor of
The Journal of Grey System, an international journal, for 24 years. In his capacity
as editor he screened articles, checked the contents of their experiments and edited
them in English; he was dedicated and tireless. Until the last moment of his life, he
was still working on publishing a scholarly book.
I still remember 1983, when I first participated in a course on Grey System
Theory. The mimeographed teaching materials had a blue cover and were presented
as a book. It was like finding a treasure, as the book attracted me deeply. The book
really inspired me, as I was a young scholar who was going through a period of
confusion and lack of direction for academic study. The book shone with sparkles
of wisdom and built a lighthouse for a knowledge seeker who was in the mist of
trying to find his way in academic research. This book became the light in my life’s
journey. From then onwards I forged an indissoluble bound with Grey System
Theory.
My most unforgettable memory is the first time I joined Prof. Deng’s compre-
hensive course on Grey System Theory in Yu Jiashan’s air-raid shelter in 1986. The
course deepened my understanding and awareness of many scientific problems in
Grey System Theory. In 1995, when I was about 40 years old, I formally became a

© Springer Science+Business Media Singapore 2017 301

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1
302 Farewell to Our Tutor

disciple of Prof. Deng, and from then on I took on the mission of disseminating and
developing Grey System Theory.
Today, Grey System Theory is accepted by academics worldwide. A variety of
academic works on grey system theory have been published in different languages,
including English, Japanese, Korean and Romanian. In 1989, the British journal
entitled The Journal of Grey System was launched. In 1997, a Chinese publication
named Journal of Grey System was launched in Taiwan. Later in 2004, this same
publication started to be published in English. In 2011, Emerald launched a new
journal entitled Grey system: Theory and application.
Since November 2007, the biennial IEEE International Conference on Grey
System and Intelligent Services has been successfully held in Nanjing, China,
attracting scholars from different parts of the world. The fourth IEEE International
Conference on Grey System and Intelligent Services will be held at the University
of Macau, in November 2013.
Currently, a significant number of scholars from China, United States, England,
Germany, Japan, Australia, Canada, Austria, Russia, Turkey, the Netherlands, Iran,
and others, have been involved in the research and application of Grey System
Theory. Many countries have begun to recruit and cultivate doctoral and other
postgraduate students in the area of Grey System Theory. To date, more than 100
students have graduated and received PhDs in this area, and tens of thousands of
graduate students have carried out their research on by the methods of Grey System
Theory.
At the Nanjing University of Aeronautics and Astronautics, Grey System Theory
has become an important course for undergraduate, masters and doctoral students
from many different colleges.
In 2008, Grey System Theory was selected as the national extractive course. In
2013, it was selected as an open learning course, free of charge to all lovers of Grey
System Theory.
Currently, there are more than 70 projects on Grey System Theory research and
applications funded by the National Natural Science Foundation of China. Many
projects have been supported by the European Union, the United Kingdom, USA,
Canada, Spain, Romania and other countries.
In 2012, De Montfort University in the UK funded and organized the first
European collaboration network for Grey System research, and representatives from
14 European Member States attended the session.
Grey System Theory, as an emerging discipline, has carved its place in science
and demonstrated strong vitality.
Farewell to our tutor!
God bless you!
(Prof. Sifeng Liu is a former PhD Student of Prof. Julong Deng. Currently Prof.
Liu works at the Nanjing University of Aeronautics and Astronautics. He also
serves as the founding director of the Institute for Grey Systems Studies, the
founding chair of the IEEE SMC Technical Committee on Grey Systems, and the
president of Grey Systems Society of China.)
References

Amanna, A., Price, M. J., et al. (2011). Grey systems theory applications to wireless
communications. Analog Integrated Circuits and Signal Processing, 69(2–3), SI, 259–269.
Andrew, A. M. (2011). Why the world is grey. Grey Systems: Theory and Application, 1(2), 112–
116.
Arasteh, A., Aliahmadi, A., & Omran, M. M. (2014). Application of gray systems and fuzzy sets in
combination with real options theory in project portfolio management. Arabian Journal for
Science and Engineering, 39(8), 6489–6506.
Aydemir, E., Bedir, F., & Ozdemir, G. (2015). Degree of greyness approach for an EPQ model
with imperfect items in copper wire industry. The Journal of Grey System, 27(2), 13–26.
Bahrami, S., Hooshmand, R.-A., & Parastegari, M. (2014). Short term electric load forecasting by
wavelet transform and grey model improved by PSO (particle swarm optimization) algorithm.
Energy, 72, 434–442.
Bristow, M., Fang, L., & Hipel, K. W. (2012). System of systems engineering and risk
management of extreme events: Concepts and case study. Risk Analysis, 32(11), 1935–1955.
Camelia, D., Scarlat, E., & Maracine, V. (2012). Grey relational analysis between firm’s current
situation and its possible causes: A bankruptcy syndrome approach. Grey Systems: Theory and
Application, 2(2), 229–239.
Camelia, D., Bradea, I., & Emil, S. (2013). A computational grey based model for companies risk
forecasting. The Journal of Grey System, 25(3), 70–83.
Carmona Benitez, R. B., Carmona Paredes, R. B., Lodewijks, G., et al. (2013). Damp trend grey
model forecasting method for airline industry. Expert Systems with Applications, 40(12), 4915–
4921.
Cengiz, K., Mesut, Y., & İhsan, K. (2010). Fuzzy and grey forecasting techniques and their
applications in production systems. Production Engineering and Management, 252, 1–24.
Chan, J. W. K., & Tong, T. K. L. (2007). Multi-criteria material selections and end-of-life product
strategy: Grey relational analysis approach. Materials & Design, 28(5), 1539–1546.
Chang, B. R., & Tsai, H. F. (2008). Forecast approach using neural network adaptation to support
vector regression grey model and generalized auto-regressive conditional heteroscedasticity.
Expert Systems with Applications, 34(2), 925–934.
Chang, S. C., Lai, H. C., & Yu, H. C. (2005). A variable P value rolling grey forecasting model for
Taiwan semiconductor industry production. Technological Forecasting and Social Change, 72
(5), 623–640.
Che-Jung, C., Der-Chiang, L., Chien-Chih, C., et al. (2014). A forecasting model for small
non-equigap data sets considering data weights and occurrence possibilities. Computers &
Industrial Engineering, 67, 139–145.
Chen, C. I., Chen, H. L., & Chen, S. P. (2008). Forecasting of foreign exchange rates of Taiwan’s
major trading partners by novel nonlinear Grey Bernoulli model NGBM (1,1).
Communications in Nonlinear Science and Numerical Simulation, 13(6), 1194–1204.

© Springer Science+Business Media Singapore 2017 303

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1
304 References

Chen, C. L., Dong, D. Y., Chen, Z. H., et al. (2008). Grey systems for intelligent sensors and
information processing. Journal of Systems Engineering and Electronics, 19(4), 659–665.
Chen, D. J., Zhang, Y. L., & Chen, M. Y. (2005). Grey macroscopic control prediction model of
systems clouds and applications. Control and Decision Making, 20(5), 553–557.
Chen, J., Shufang, Z., & Liu, Y. (2012). On multi-agents decision and simulation of catastrophe
insurance based on grey game model. Soft Science, 26(7), 131–136.
Chen, L., Yan, Q., & Ruru, D. (2011). Comparative analysis of the two common methods for
skylight Measurement based on the ASD Spectroradiometer. Tropical Geography, 31(2),
182–186.
Chen, M. Y. (1982). Grey dynamic control of boring machine. Journal of Huazhong University of
Science and Technology, 10(2), 7–11.
Chen, M. Y. (1985). Stability of grey systems and problem of calmness. Fuzzy Mathematics, 5(2),
54–58.
Chen, R., Song, Z., & Kang, L. (2005). Application of grey system to reservoir evaluation in the
seven district of Xinjiang Karamay Oilfield. Inner Mongolia Petroleum Chemical Industry,
7:110–113.
Chen, X., Xia, J., & Xu, Q. (2009). Self-memory prediction mode with grey differential model.
China Science E: Technological Science, 39(2), 341–350.
Chen, X. X., & Liu, S. F. (2008). On grey multi-attribute group decision-making with completely
unknown weight information. Management science of China, 16(5), 146–152.
Chen, Y. J., Fu, Y. H., & Sun, H. (2007). Grey theory applied in the construction control of
steel-pipe-concrete arch bridges. Journal of Harbing University of Technology, 39(4),
546–548.
Chen, Y. K., & Tan, X. R. (1995). Grey relational analysis on serum markers of liver fibrosis. The
Journal of Grey System, 7(1), 63–68.
Cheng, Q. Y., & Qiu, W. H. (2001). Grey system forecast for firing accuracy of gun. Journal of
Systems Science and Systems Engineering, 10(2), 205–211.
Chiang, J. Y., & Chen, C.-K. (2008). Application of grey prediction to inverse nonlinear heat
conduction problem. International Journal of Heat and Mass Transfer, 51(3), 576–585.
Chiang, Y. M., & Hsieh, H. H. (2009). The use of the Taguchi method with grey relational
analysis to optimize the thin-film sputtering process with multiple quality characteristic in color
filter manufacturing. Computers & Industrial Engineering, 56(2), 648–661.
Chiao, C. H., & Wang, W. Y. (2002). Reliability improvement of fluorescent lamp using grey
forecasting model. Microelectronics Reliability, 42(1), 127–134.
Chirwa, E. C., & Mao, M. (2006).Application of grey model GM (1, 1) to vehicle fatality risk
estimation. Technological Forecasting and Social Change, 73(5), 588–605.
Chiu, N. H. (2009). An early software-quality classification based on improved grey relational
classifier. Expert Systems with Applications, 36(7), 10727–10734.
Chou, J. H., Chen, S. H., & Li, J. J. (2000). Application of the Taguchi-genetic method to design
an optimal grey-fuzzy controller of a constant turning force system. Journal of Materials
Processing Technology, 105(3), 333–343.
Chu, Y. F. (2008). Predictive research on the development scale of Chinese logistics based on grey
systems theory. Management Review, 20(3), 58–64.
Churchman, G. W. (1979). The systems approach and its enemies. New York: Basic Books.
Cobb, C. W., & Douglas, P. H. (1928). A theory of production. American Economic Review, 18
(Supplement), 139–165.
Cristóbal, S., & Ramón, J. (2015). A cost forecasting model for a vessel dry docking. Journal of
Ship Production and Design, 31(1), 58–62(5).
Cui, J., Xin, Y., & Liu,-X. (2012). Research on surface to air missile type selection based on
multi-criteria gray decision-making. Tactical Missile Technology, 1, 7–10.
Cui, J., & Dang, Y. G. (2009). On the GM (1, 1) prediction accuracy based on another class of
strengthening buffer operators. Control and Decision-Making, 24(1), 44–48.
References 305

Cui, J., Dang, Y. G., & Liu, S. F. (2008). An improvement method for solving criteria weights
based on degrees of grey incidence. Management Science of China, 16(5), 141–145.
Cui, J., Dang, Y. G., & Liu, S. F. (2009). Precision analysis of the GM (1, 1) model based on a
new weakening operator. Systems Engineering: Theory and Practice, 29(7), 132–138.
Cui, L. Z., Liu, S. F., & Wu, Z. P. (2009). Some new weakening buffer operators and their
applications. Control and Decision-Making, 24(8), 1252–1256.
Cui, L., Liu, S., & Wu, Z. (2010). On the construction of a new strengthening buffer operator and
its application. Systems Engineering Theory and Practice, 30(3), 484–489.
Czeslaw, C. (2008). Decomposition of the symptom observation matrix and grey forecasting in
vibration condition monitoring of machines. International Journal of Applied Mathematics and
Computer Science, 18(4), 569–579.
Dai, W. Z., & Li, J. F. (2005). Modeling with non-equal-distant GM (1, 1). Systems Engineering:
Theory and Practice, 25(9), 89–93.
Dai, W., & Su, Y. (2012). On the construction of a strengthening buffer operator of new
information priority and its application. Systems Engineering Theory and Practice, 38(8),
1329–1334.
Dang, Y. G. (1995). Valuation of grey clustering with human experience. The Journal of Grey
System, 7(2), 179–184.
Dang, Y. G., Liu, B., & Guang, Y. Q. (2005). On strengthening buffer operators. Control and
Decision-Making, 20(12), 1332–1336.
Dang, Y. G., & Liu, S. F. (2004). The GM models that x(n) is taken as initial value. Kybernetes.
The International Journal of Systems & Cybernetics, 33(2), 247–254.
Dang, Y. G., & Liu, S. F. (2006). Multi-attribute grey incidence decision model for interval
number. Kybernetes, 35(7), 1265–1272.
Dang, Y. G., Liu, S. F., & Fang, Z. G. (2003). The matrix method for computing the cross-section
set of maximum flows in networks. Systems Engineering: Theory and Practice, 23(9),
125–128.
Dang, Y. G., Liu, S. F., Li, B. J., et al. (2000). Predictive analysis on the contribution of
technology in the dominant industries. Systems Engineering: Theory and Practice, 20(8),
95–99.
Dang, Y. G., Liu, S. F., Liu, B., et al. (2004). An improvement on the slope degree of grey
incidence. Engineering Science of China, 6(3), 41–44.
Dang, Y. G., Liu, S. F., Liu, B., et al. (2004). About weakening buffer operators. Management
Science of China, 12(2), 108–111.
Dang, Y. G., Liu, S. F., Liu, B., et al. (2004). Study on multi-criteria decision-making based on
interval number incidence. Journal of Nanjing University of Aeronautics and Astronautics,
36(3), 403–406.
Dao, T. L., Dinh, Q. T., & Kyoung, K. A. (2015). A torque estimator using online tuning grey
fuzzy PID for applications to torque-sensorless control of DC motors. Mechatronics,
26, 45–63.
Delgado, A., & Romero, I. (2016). Environmental conflict analysis using an integrated grey
clustering and entropy-weight method: A case study of a mining project in Peru.
Environmental Modelling & Software, 77, 108–121.
Deng, J. L. (1982a). Control problems of grey systems. Systems & Control Letters, 1(5), 288–294.
Deng, J. L. (1982b). Grey Control system. Journal of Huazhong University of Science and
Technology, 10(3), 9–18.
Deng, J. L. (1984). Grey dynamic models and their application in the long-term prediction of food
production. Exploration of Nature, 3(3), 7–43.
Deng, J. L. (1985). Grey control systems. Wuhan: Press of Huazhong University of Science and
Technology.
Deng, J. L. (1985a). The incidence space of grey systems. Fuzzy Mathematics, 5(2), 1–10.
Deng, J. L. (1985b). Generation functions of grey systems. Fuzzy Mathematics, 5(2), 11–22.
Deng, J. L. (1985c). The GM model of grey systems theory. Fuzzy Mathematics, 5(2), 23–32.
306 References

Deng, J. L. (1985d). Five kinds of grey predictions. Fuzzy Mathematics, 5(2), 33–42.
Deng, J. L. (1985e). Grey situational decision making. Fuzzy Mathematics, 5(2), 43–50.
Deng, J. L. (1986). Grey prediction and decision-making. Wuhan: Press of Huazhong College of
Science and Technology.
Deng, J. L. (1987). Grey exponential law of accumulating generation: the optimization information
problem of grey control systems. Journal of Huazhong University of Science and Technology,
12(5), 11–16.
Deng, J. L. (1988). Grey multi-dimensional planning. Wuhan: Press of Huazhong University of
Science and Technology.
Deng, J. L. (1989). Grey information space. The Journal of Grey System, 1(2), 103–118.
Deng, J. L. (1990). A Course in grey systems theory. Wuhan: Press of Huazhong University of
Science and Technology.
Deng, J. L. (1991). Representational forms of information in grey systems theory. Grey Systems
Theory and Practice, 1(1), 1–12.
Deng, J. L. (1992). Grey causality and information cover. The Journal of Grey System, 4(2),
99–119.
Deng, J. L. (1993). Grey differential equation. The Journal of Grey System, 5(1), 1–14.
Deng, J. L. (1995). Extent information cover in grey system theory. The Journal of Grey System, 7
(2), 131–138.
Deng, J. L. (1996). Several problems in grey systems theory and development. In S. F. Liu & Z.
X. Xu (Eds.), New advances in grey systems research (pp. 1–10). Wuhan: Press of Huazhong
University of Science and Technology.
Deng, J. L., & Ng, D. K. W. (1996). Chaos in grey model GM (1, N). The Journal of Grey System,
8(1), 1–10.
Deng, J. L., & Zhou, C. S. (1986). Sufficient conditions for the stability of a class of interconnected
dynamic systems. Systems & Control Letters, 5(2), 105–108.
Dong, F. Y. (2007). Development prediction on the finance of Chinese firms based on newly
improved GM (1, 1) model. Management Science of China, 15(4), 93–97.
Dounis, A.I., Tiropanis, P., Argiriou, A., et al. (2011). Intelligent control system for reconciliation
of the energy savings with comfort in buildings using soft computing techniques. Energy and
Buildings, 43, 66–74.
Dymova, L., Sevastjanov, P., & Pilarek, M. (2013). A method for solving systems of linear
interval equations applied to the Leontief input-output model of economics. Expert Systems
With Applications, 40(1), 222–230.
Ejnioui, A., Otero, C. E., & Otero, L. D. (2013). Prioritisation of software requirements using grey
relational analysis. International Journal of Computer Applications in Technology, 47(2–3),
100–109.
Emil, S., & Delcea, C. (2011). Complete analysis of bankruptcy syndrome using grey systems
theory. Grey Systems: Theory and Application, 1(1), 19–32.
Erdal, K., Baris, U., & Okyak, K. (2010). Grey system theory-based models in time series
prediction. Expert Systems with Applications, 37(4), 1784–1789.
Esme, U., Kulekci, M. K., Ustun, D., et al. (2015). Grey-based fuzzy algorithm for the
optimization of the ball burnishing process. Materials Testing, 57(7–8), 666–673.
Evans, M. (2014). An alternative approach to estimating the parameters of a generalised grey
Verhulst model: An application to steel intensity of use in the UK. Expert Systems with
Applications, 41(4), 1236–1244.
Fan, T. G., Zhang, Y. C., & Quan, X. Z. (2001). Grey model GM (1, 1, t) and its chaos feature.
Journal of Grey System, 13(2), 117–120.
Fang, X., Chen, Y., & Shaoquan, L. (2012). Application of multidimensional grey evaluation
methods in coal and gas outburst prediction. Industrial Safety and Environmental Protection,
38(12), 81–83.
References 307

Fang, Z. G., & Liu, S. F. (2003). Grey matrix model based on pure strategy. In M. I. Dessouky &
C. Heavey (Eds.), Proceedings of the 32nd international conference on computers & industrial
engineering[C] (pp. 520–525). Ireland: Gemini International Limited Dublin.
Fang, Z. G., & Liu, S. F. (2003). Grey matrix game models based on pure strategies (1):
Construction of standard grey matrix game models. Journal of Southeast University (Natural
Science Edition), 33(6), 796–800.
Fang, Z. G., & Liu, S. F. (2003). Two-player zero-sum game models based on the grey matrices of
pure strategies. Journal of Nanjing Aeronautics and Astronautics, 35(4), 441–445.
Fang, Z. G., & Liu, S. F. (2005). A study on the GM (1, 1) (GMBIGN (1, 1)) model based on
interval grey sequences. Engineering Science of China, 7(2), 57–61.
Fang, Z. G., Liu, S. F., & Dang, Y. G. (2003). Analysis of the concealed allocation model with the
optimal grey information entropy of military traffic flow. Management Science of China, 11(3),
56–61.
Fang, Z. G., Liu, S. F., & Dang, Y. G. (2004). Algorithm model research of the logical cutting tree
on the network maximum flow. Kybernetes: The International Journal of Systems &
Cybernetics, 33(2), 255–262.
Fang, Z. G., Liu, S. F., Zhang, Y. J., et al. (2002). Consistency analysis on international free trades
and the development of Chinese national economic interests. Management Science of China,
10(special issue), 221–225.
Fang, Z., Liu, S., & Shi, S. (2010). Grey game theory and its applications in economic
decision-making. New York: Taylor&Francis Group.
Fang, Z., Wang, C., Zhang, N., et al. (2014). Analysis of functional game model based on grey
information variable. Chinese Journal of Management Science, 22(2), 112–118.
Farooq, U., Siddiqui, M. A., Gao, L., & Jennifer, L. H. (2012). Intelligent transportation systems:
an impact analysis for Michigan. Journal of Advanced Transportation, 46(1), 12–25.
Feng, Y. C., Li, D., Wang, Y. Z., et al. (2001). A grey model for response time in database system
DM3. Journal of Grey System, 13(3).
Feng, Z. Y. (1992). Direct grey modeling. Journal of Applied Mathematics, 15(3), 345–354.
Gao, F., Zhang, Y., & Gao, P. (2012). Research on speed controller model for high-speed train
based on grey genetic algorithm. Computer Measurement & Control, 20(5), 1272–1275.
Gini, C. (1921). Measurement of inequality of incomes. The Economic Journal (Blackwell
Publishing), 31(121), 124–126.
Goel, B., Singh, S., & Sarepaka, R. V. (2015). Optimizing single point diamond turning for
mono-crystalline germanium using grey relational analysis. Materials and Manufacturing
Processes, 30(8), 1018–1025.
Golinska, P., Kosacka, M., Mierzwiak, R., et al. (2015). Grey decision making as a tool for the
classification of the sustainability level of remanufacturing companies. Journal of Cleaner
Production, 105, 28–40.
Golmohammadi, D., & Mellat-Parast, M. (2012). Developing a grey-based decision-making model
for supplier selection. International Journal of Production Economics, 137(2), 191–200.
Guan, Y. Q., & Liu, S. F. (2007). Strengthening buffer operator sequences constructed on fixed
points and applications. Control and Decision Making, 22(10), 1189–1193.
Guan, Y., & Liu, S. (2008). On matrix of linear buffer operator and its application. Journal of
Applied Mathematics of Universities, 23(3), 357–362.
Guo, H. (1991). Grey exponential law attributed to mutual complement of series. The Journal of
Grey System, 3(2), 153–162.
Guo, H., Xiao, X., & Jeffrey, F. (2014). Problem of grey bilevel multi-objective linear
programming and its algorithm. Control and Decision, 29(7), 1193–1198.
Guo, P. (2007). MAC control on grey error corrections and application in workloads of machine
groups. Journal of System Simulation, 19(18), 4326–4330.
Guo, R. K. (2007). Modeling imperfectly repaired system data via grey differential equations with
unequal-gapped times. Reliability Engineering & System Safety, 92(3), 378–391.
308 References

Guo, X., Liu, S., & Wu, L. (2013). Modeling and algorithm of prediction about pollutants’
emission reduction by combining gray theory and Markov chain. Research on Computer
Application, 30(12), 3670–3673.
Guo, X., Liu, S., & Fang, Z. (2014). Self-memory prediction model of interval grey number based
on grey degree of compound grey number. Systems Engineering and Electronics, 36(6),
1124–1129.
Guo, X., Liu, S., Wu, L., et al. (2015). A multi-variable grey model with a self-memory component
and its application on engineering prediction. Engineering Applications of Artificial
Intelligence, 42, 82–93.
Gurden, S. P., Westerhuis, J. A., Bijlsma, S., et al. (2001). Modelling of spectroscopic batch
process data using grey models to incorporate external information. Journal of Chemometrics,
15(2), 101–121.
Hamzacebi, C., & Es, H. A. (2014). Forecasting the annual electricity consumption of Turkey
using an optimized grey model. Energy, 70, 165–171.
Han, X., Nan, H., Chen, J., et al. (2014). Grey cluster model used to evaluate the development
scheme of warhead in antimissile missile of air defense. Journal of Air Force Engineering
University, 1(1), 29–33.
Haken, H. (2011). Book reviews: Grey information: theory and practical applications. Grey
Systems: Theory and Application, 1(1), 105–106.
Hao, Q., Zhu, M. L., & Li, B. (1996). Methods of ordering fuzzy and grey incidences and
applications. Systems Engineering: Theory and Practice, 16(6), 44–49.
Hao, Y. H., & Wang, X. M. (2000). Period residual modification of GM (1, 1) modeling. Journal
of Grey System, 12(2), 181–183.
Hao, Y., Zhao, J., Li, H., et al. (2012). Karst hydrological processes and grey system model.
Journal of the American Water Resources Association, 48(4), 656–666.
Hao, Y., Cao, B., Chen, X., et al. (2013). A piecewise grey system model for study the effects of
anthropogenic activities on karst hydrological processes. Water Resources Management, 27(5),
1207–1220.
He, R. S., & Hang, S. F. (2007). Damage detection by a hybrid real-parameter genetic algorithm
under the assistance of grey relation analysis. Engineering Applications of Artificial
Intelligence, 20(7), 980–992.
He, W. Z., Song, G. X., & Wu, A. D. (2005). A class of schemes for computing the parameters of
the GM (1, 1) model. Systems Engineering: Theory and Practice, 25(1), 69–75.
He, W. Z., & Wu, A. D. (2006). Linear programming methods for estimating the parameters of
Verhulst model and applications. Systems Engineering: Theory and Practice, 26(8), 141–144.
He, Y., & Bao, Y. D. (1992). Grey Markov prediction model and applications. Systems
Engineering: Theory and Practice, 12(4), 59–63.
Hipel, K. W. (2011). Book reviews: Grey systems: theory and applications. Grey Systems: Theory
and Application, 1(3), 274–275.
Hodzic, M., & Tai, L.-C. (2016). Grey Predictor reference model for assisting particle swarm
optimization for wind turbine control. Renewable Energy, 86, 251–256.
Hsieh, C. H. (2001). Grey data fitting model and its application to image coding. Journal of Grey
System, 13(1), 245–254.
Hsu, C. C., & Chen, C. Y. (2003). A modified grey forecasting model for long-term prediction.
Journal of the Chinese institute of engineers, 26(3), 301–308.
Hsu, C. C., & Chen, C. Y. (2003). Applications of improved grey prediction model for power
demand forecasting. Energy conversion and management, 44(14), 2241–2249.
Hsu, L. C., & Wang, C. H. (2007). Forecasting the output of integrated circuit industry using a
grey model improved by the Bayesian analysis. Technological Forecasting and Social Change,
74(6), 843–853.
Hsu, Y. T., & Yeh, J. (2000). A novel image compression using grey models on a dynamic
window. International Journal of Systems SCIENCE, 31(9), 1125–1141.
References 309

Hu, F., Huang, J. G., & Zhang, Q. F. (2007). Research on effect evaluation of underwater
navigation devices based on grey systems theory. Journal of Northwest University of
Technology, 25(3), 411–415.
Hu, Q. Z. (2007). Decision-making model of grey incidences that optimizes urban traffic networks.
Journal of Systems Engineering, 22(6), 607–612.
Hu, W. B., Hua, B., & Yang, C. Z. (2002). Building thermal process analysis with grey system
method. Building and Environment, 37(6), 599–605.
Hu, X.-L., Wu, Z.-P., & Han, R. (2013). Analysis on the strengthening buffer operator based on
the strictly monotone function [J]. International Journal of Applied Physics and Mathematics,
3(2), 132–136.
Hu, Y. C. (2007). Grey relational analysis and radial basis function network for determining costs
in learning sequences. Applied Mathematics and Computation, 184(2), 291–299.
Huang, F. Y. (1994). Transformation method of grey systems modeling. Systems Engineering:
Theory and Practice, 14(6), 35–38.
Huang, S. J., & Huang, C. L. (2000). Control of an inverted pendulum using grey prediction
model. IEEE Transactions on Industry Applications, 36(2), 452–458.
Huang, J., & Zhong, X. L. (2009). A general accumulation grey predictive control model and its
optimal computational scheme. Systems Engineering: Theory and Practice, 29(6), 147–156.
Huang, K. Y., & Jane, C. J. (2009). A hybrid model for stock market forecasting and portfolio
selection based on ARX, grey system and RS theories. Expert Systems with Applications,
36(3), 5387–5392.
Huang, S. J., Chiu, N. H., & Chen, L. W. (2008). Integration of the grey relational analysis with
genetic algorithm for software effort estimation. European Journal of Operational Research,
188(3), 898–909.
Huang, Y. L., & Fang, B. (2007). Systems’ quasi-quantitative incidence matrices based on grey
models. Journal of System Simulation, 19(3), 474–477.
Huang, Y. P., & Chu, H. C. (1996). Practical consideration for grey modeling and its application to
image processing. The Journal of Grey System, 8(3), 217–233.
Huang, Y. P., & Chu, H. C. (1999). Simplifying fuzzy modeling by both gray relational analysis
and data transformation methods. Fuzzy Sets and Systems, 104(2), 183–197.
Ji, P. R., Huang, W. S., & Hu, X. Y. (2001). On characteristics of grey prediction models. Systems
Engineering: Theory and Practice, 21(9), 105–108.
Jia, H. F., & Zheng, Y. Q. (1998). Combined prediction models of grey time series and its
application in predicting annual precipitations. Systems engineering: Theory and Practice,
18(8), 42–45.
Jiang, C. W., Hou, Z. J., & Zhang, Y. C. (2000). Chaos analysis of the grey forecasting model.
Journal of Grey System, 12(4), 319–322.
Jiang, W. (2012). An intelligent diagnosis method based on grey rough set theory for wind turbine
driving chain. Power System and Clean Energy, 28(12), 79–83.
Jiang, Y. Q., Yao, Y., Deng, S. M., et al. (2004). Applying grey forecasting to predicting the
operating energy performance of air cooled water chillers. International Journal of
Refrigeration, 27(4), 385–392.
Jie, J. X., Song, B. F., & Liu, D. X. (2004). A grey incidence method designed for selecting the
optimum design of the top layer of aircrafts. Journal of Systems Engineering, 19(4), 350–354.
Jie, C., Sifeng, L., & Naiming, X. (2012). Novel grey decision making model and its numerical
simulation. Transactions of Nanjing University of Aeronautics & Astronautics, 29(2),
112–117.
Jin, Y. F., Zhu, Q. S., & Xing, Y. K. (2006). The grey logical method for inserting the missing
values in sequential data. Control and Decision Making, 21(2), 236–240.
Kadier, A., Abdeshahian, P., Simayi, Y., et al. (2015). Grey relational analysis for comparative
assessment of different cathode materials in microbial electrolysis cells. Energy, 90,
1556–1562.
310 References

Karmakar, S., & Mujumdar, P. P. (2007). A two-phase grey fuzzy optimization approach for water
quality management of a river system. Advances in Water Resources, 30(5), 1218–1235.
Kaya, Y. (2015). Hidden pattern discovery on epileptic EEG with 1-D local binary patterns and
epileptic seizures detection by grey relational analysis. Australasian Physical & Engineering
Sciences in Medicine, 38(3), 435–446.
Khuman, A. S., Yang, Y., Liu, S. (2016). Grey relational analysis and natural language processing
to: grey language processing. The Journal of Grey System, 28(1), 88–97.
Kim, D., Goh, T., Park, M., et al. (2015). Fuzzy sliding mode observer with grey prediction for the
estimation of the state-of-charge of a lithium-ion battery. Energies, 8(11), 12409–12428.
Ko, A. S., & Chang, N. B. (2008). Optimal planning of co-firing alternative fuels with coal in a
power plant by grey nonlinear mixed integer programming model. Journal of Environmental
Management, 88(1), 11–27.
Ko, J. Z. (1996). Data sequence transformation and GM (1, 1) modeling precision. In S. F. Liu &
Z. X. Xu (Eds.), New Advances in Grey systems Research (pp. 233–235). Wuhan: Press of
Huazhong University of Science and Technology.
Kose, E., & Forrest, J. Y. L. (2015). N-person grey game. Kybernetes, 44(2), 271–282.
Kose, E., & Tasci, L. (2015). Prediction of the vertical displacement on the crest of Keban Dam.
The Journal of Grey System, 27(1), 12–20.
Ku, L. L., & Huang, T. C. (2006). Sequential monitoring of manufacturing processes: an
application of grey forecasting models. International Journal of Advanced Manufacturing
Technology, 27(5), 543–546.
Kumar, S. S., Uthayakumar, M., Kumaran, S., Thirumalai, et al. (2015). Parametric optimization
of wire electrical discharge machining on aluminium based composites through grey relational
analysis. Journal of Manufacturing Processes, 20, 33–39.
Kung, C. Y., & Wen, K. L. (2007). Applying grey relational analysis and grey decision-making to
evaluate the relationship between company attributes and its financial performance: A case
study of venture capital enterprises in Taiwan. Decision Support Systems, 43(3), 842–852.
Kuo, Y. Y., Yang, T. H., & Huang, G. W. (2008). The use of grey relational analysis in solving
multiple attribute decision-making problems. Computers & Industrial Engineering, 55(1),
80–93.
Kuzu, A., Bogosyan, S., & Gokasan, M. (2016). Predictive input delay compensation with grey
predictor for networked control system. International Journal of Computers Communications
& Control, 11(1), 67–76.
Lai, H. H., Lin, Y. C., & Yeh, C. H. (2005). Form design of product image using grey relational
analysis and neural network models. Computers & Operations Research, 32(10), 2689–2711.
Lai, Y., Chen, X., Qin, Q., et al. (2004). Corrosion analysis and corrosion rate s prediction of
NiFe2O4 cermet inert anodes. Journal of Center South University, 35(6), 896–901.
Lee, W. M., & Liao, Y. S. (2003). Self-tuning fuzzy control with a grey prediction for wire rupture
prevention in WEDM. International Journal of Advanced Manufacturing Technology, 22(7-8),
481–490.
Li, B., & Wei, Y. (2009). A new GM (1, 1) model established with optimized grey derivatives.
Systems Engineering: Theory and Practice, 29(2), 100–105.
Li, B. Z., & Zhu, X. X. (2007). On the RIS evolutionary mechanism based on dissipation and grey
incidence entropy. Studies of Science of Sciences, 25(6), 1239–1244.
Li, B. L., & Deng, J. L. (1984). A grey model for the biological prevention and treatment system of
cotton aphids. Exploration of Nature, 3(3), 44–49.
Li, B. J., Liu, S. F., & Zhu, Y. D. (2000). A method to determine the reliability of grey intervals.
Systems Engineering: Theory and Practice, 20(4), 104–106.
Li, C., Chen, K., & Xiang, X. (2015). An integrated framework for effective safety management
evaluation: Application of an improved grey clustering measurement. Expert Systems with
Applications, 42(13), 5541–5553.
References 311

Li, C. H., & Tsai, M. J. (2009). Multi-objective optimization of laser cutting for flash memory
modules with special shapes using grey relational analysis. Optics & Laser Technology, 41(5),
634–642.
Li, D.-C., Chang, C.-J., & Chen, W.-C. (2011). An extended grey forecasting model for
omnidirectional forecasting considering data gap difference. Applied Mathematical Modelling,
35, 5051–5058.
Li, G. D., Yamaguchi, D., & Nagai, M. (2007). A grey-based decision-making approach to the
supplier selection problem. Mathematical and Computer Modelling, 46(3), 573–581.
Li, G. D., Yamaguchi, D., & Nagai, M. (2007). A GM (1, 1)–Markov chain combined model with
an application to predict the number of Chinese international airlines. Technological
Forecasting and Social Change, 74(8), 1465–1481.
Li, G. D., Yamaguchi, D., & Nagai, M. (2007). A grey-based decision-making approach to the
supplier selection problem. Mathematical and Computer Modelling, 46(3), 573–581.
Li, G. D., Masuda, S., & Nagai, M. (2015). Predictor design using an improved grey model in
control systems. International Journal of Computer Integrated Manufacturing, 28(3), 297–
306.
Li, J., Sun, C. X., & Cheng, W. G. (2002). Study on fault diagnosis of insulation of oil-immersed
transformer based on grey cluster theory. Transactions of China Electrotechnical Society,
17(4), 80–83.
Li, J. B., Zhao, J. Y., Zheng, X. X., et al. (2008). A new evaluation method of the state of electric
transformers based on grey target theory. Journal of Jilin University (Industry Edition), 38(1),
201–205.
Li, J. F., & Dai, W. Z. (2004). Multi-period grey modeling method based on stepwise ratios and its
application in Chinese GDP modeling. Systems engineering: Theory and Practice, 24(9),
98–102.
Li, P., Yang, H., Sun, L., et al. (2011). Application of gray prediction and time series model in
spacecraft prognostic. Computer Measurement & Control, 19(1), 111–113.
Li, Q. X. (2009). Grey dynamic input–output analysis. Journal of Mathematical Analysis and
Applications, 359(2), 514–526.
Li, Q. X., & Liu, S. F. (2008). The foundation of the grey matrix and the grey input–output
analysis. Applied Mathematical Modelling, 32(3), 267–291.
Li, Q.-X., Liu, S.-F., & Lin, Y. (2012). Grey enterprise input-output analysis. Journal of
Computational and Applied Mathematics, 236(7), 1862–1875.
Li, S. B. (2007). Grey incidence projection model and application for evaluating the friendliness of
regional environment. Science and Sciences and Management of Science and Technology, 28
(8), 103–107.
Li, T. J., Xiong, W., Wang, P., et al. (1990). Grey prediction methods of experiments without
structural damages. Journal of Huazhong University of Science and Technology, 23(5), 75–78.
Li, W. X. (1991). Degrees of incidence of continuous processes. Journal of Huazhong University
of Science and Technology, 19(3), 113–116.
Li, X. B., Sun, H. Y., & Wu, Y. X. (2009). Dynamic tracking multi-variable fuzzy predictive
functional control of fuel oil feeding temperature. Computer Engineering and applications,
45(9), 200–203.
Li, X. C. (1999). Expansion of the validity range of the GM (1, 1) model of the grey systems.
Systems engineering: Theory and Practice, 19(1), 97–101.
Li, X., Yuan, Z., Zhang, G., et al. (2014). Some properties of grey differential equation GM (1, 1,
b). Systems Engineering Theory and Practice, 34(5), 1249–1255.
Li, X. F., Wang, J. J., Ma, J., et al. (2007). Grey incidence analysis for the quality of cotton fiber
materials and the resultant yarns and single yarn strength. Journal of Applied Science, 25(1),
100–102.
Li, X., Dang, Y., & Wang, Z. (2012). Harmonic buffer operators with variable weight and effect
strength comparison. Systems Engineering Theory and Practice, 32(11), 2486–2492.
312 References

Li, X. Q. (1995). A further investigation of quantified grey incidences. Systems Engineering, 13

(6), 58–61.
Li, X. Q., Li, S. R., & Han, X. L. (1997). A generalized application of grey systems GM (n, h)
model. Systems Engineering: Theory and Practice, 17(8), 82–85.
Li, X., Tang, B., & Tang, B. (2007). The optimal collocation model of missile nuke based on the
gray decision theory. Fire Control and Command Control, 32(2), 42–47.
Liang, B., Dai, Y., Chen, T., et al. (2014).Grey correlation optimization for shale gas exploration
and development areas of complicated geological parameter features. Journal of China Coal
Society, 39(3), 524–530.
Liang, C., Gu, D., & Bichindaritz, I. (2012). Integrating gray system theory and logistic regression
into case-based reasoning for safety assessment of thermal power plants. Expert Systems with
Applications, 39(5), 5154–5167.
Liang, Q. W., Song, B. W., & Jia, Y. The grey Verhulst model of the costs on the research of
torpedoes. Journal of System Simulation, 17(2), 257–258.
Liao, R. J., Yang, J. P., Grzybowski, S., et al. (2012). Forecasting dissolved gases content in power
transformer oil based on weakening buffer operator and least square support vector
machine-Markov. IET Generation, Transmission & Distribution, 6(2), 142–151.
Lin, C. T., & Yang, S. Y. (2003). Forecast of the output value of Taiwan’s opto-electronics
industry using the Grey forecasting model. Technological Forecasting and Social Change,
70(2), 177–186.
Lin, J., Ren, H., & Shen, Z. (2009). Study on primary influence factors for application of grey
system theong to velocity of explosive forming projectile. Journal o f Projectiles, Rockets,
Missiles and Guidance, 29(3), 112–116.
Lin, J. L., & Lin, C. L. (2002). The use of the orthogonal array with grey relational analysis to
optimize the electrical discharge machining process with multiple performance characteristics.
International Journal of Machine Tools and Manufacture, 42(2), 237–244.
Lin, J. L., Wang, K. S., Yan, B. H., et al. (1999). Optimization of parameters in GM (1, 1) model
by Taguchi method. Journal of Grey System, 11(4), 257–277.
Lin, W., & Ho, Y. X. (1996). Research and application of grey multi-target programming. In S.
F. Liu & Z. X. Xu (Eds.), New advances in grey systems research (pp. 373–375). Wuhan: Press
of Huazhong University of Science and Technology.
Lin, Y., & Liu, S. F. (1999). Regional economic planning based on systemic analysis of small
samples (I). Problems of Nonlinear Analysis in Engineering Systems, 2(10), 24–33.
Lin, Y., & Liu, S. F. (1999). Several programming models with unascertained parameters and their
application. Journal of Multi-Criteria Decision Analysis, 8, 206–220.
Lin, Y., & Liu, S. F. (2000). Law of exponentiality and exponential curve fitting. Systems Analysis
Modelling Simulation, 38(1), 621–636.
Lin, Y., & Liu, S. F. (2000). Regional economic planning based on systemic analysis of small
samples (II). Problems of Nonlinear Analysis in Engineering Systems, 6(11), 33–49.
Lin, Y., Liu, S. F., & McNeil, D. H. (1998). The past, present, and future of systems science
research. Systems Science and Sustainable Development (pp. 245–252). Beijing: Press of
Scientific and Technological Literature.
Lin, Y. H., & Lee, P. C. (2007). Novel high-precision grey forecasting model. Automation in
Construction, 16(6), 771–777.
Lin, Y. H., Lee, P. C., & Chang, T. P. (2009). Adaptive and high-precision grey forecasting model.
Expert Systems with Applications, 36(6), 9658–9662.
Lin, Y., Wang, T., Wang, L., et al. (2005). Stability analysis of high excavated slope in three
gorges project. Journal of Tianjing University, 38(10), 936–940.
Liu, B., Liu, S. F., & Dang, Y. G. (2003). The GST-based data mining techniques of time
sequences. Proceedings of the 32nd International Conference on Computers & Industrial
Engineering (pp. 548–553).
Liu, B., Liu, S. F., & Dang, Y. G. (2003). Time series data mining techniques based on grey
systems theory. Engineering Science of China, 5(9), 32–35.
References 313

Liu, B., Liu, S. F., Cui, Z. J., et al. (2003). Optimization of time response functions of the GM (1,
1) model. Management Science of China, 11(4), 54–57.
Liu, J. P., Ji, C. S., & Li, H. (2001). Application of fixed weight grey clustering analysis in
evaluating coal mining methods. Journal of Coal Industry, 26(5), 493–495.
Liu, J., Liu, S., & Fang, Z. (2015). Fractional-order reverse accumulation generation GM (1, 1)
model and its applications. The Journal of Grey System, 27(4), 52–62.
Liu, J., Xiao, X., Guo, J., et al. (2014). Error and its upper bound estimation between the solutions
of GM (1, 1) grey forecasting models. Applied Mathematics and Computation, 246, 648–660.
Liu, P. L. (2001). Stability of continuous and discrete time-delay grey systems. International
Journal of Systems Science, 32(7), 29–39.
Liu, Q., Zhong, Z., & Ai, B. (2010). Research on frequency planning of GSM-R system based on
grey cluster theory with rough set. Journal of the China Railway Society, 32(5), 53–58.
Liu, S. A., Du, H. T., & Wang, X. L. (2001). Rough set theory and applications. Systems
Engineering: Theory and Practice, 21(10), 77–82.
Liu, S. F. (1989). On Perron-Frobenius theorem of grey nonnegative matrix. Journal of Grey
System, 1(2), 157–166.
Liu, S. F. (1991). The three axioms of buffer operator and their application. Journal of Grey
System, 3(1), 39–48.
Liu, S. F., & Guo, T. B. (1991). Grey systems theory and applications. Kaifeng: Press of Henan
University.
Liu, S. F. (1992). Generalized degree of grey incidence. In Z. Shengkai, (Eds.), Information and
systems (pp. 113–116). Dalian: DMU Publishing House.
Liu, S. F. (1993a). Fixed weight grey clustering evaluation analysis. In New methods of grey
systems (pp. 178–184). Beijing: Agriculture Press.
Liu, S. F. (1993b). Von Neumann’s economic growing turnpike of Henan Province. In Z. Weimin
(Eds.), Systems science and systems engineering (pp. 316 – 319). Beijing: International
Academic Publishers.
Liu, S. F., & Zhu, Y. D. (1993). Models based on criteria’s triangular membership functions for
the evaluation of regional economics. Journal of Agricultural Engineering, 9(2), 8–13.
Liu, S. F., & Zhu, Y. D. (1994). Study on triangular model and indexes in synthetic evaluation of
regional economy. Mianyun Chen (pp. 1274–1279). Systems Control Information
Methodologies & Applications: HUST Publishing House.
Liu, S. F., & Zhu, Y. D. (1996). Grey-econometrics combined model. The Journal of Grey System,
8(1), 103–110. Liu, S. F., Li, B. J., & Dang, Y. G. (2004). The G-C-D model and technical
advance. Kybernetes: The International Journal of Systems & Cybernetics, 33(2), 303–309.
Liu, S. F. (1994). Grey forecast of drought and inundation in Henan Province. The Journal of Grey
System, 6(4), 279–288.
Liu, S. F. (1995). On measure of grey information. The Journal of Grey System, 7(2), 97–101.
Liu, S. F. (1996). Axiom on grey degree. The Journal of Grey System, 8(4), 397–400.
Liu, S. F., & Xu, Z. X. (Eds.). (1996). New Advances in grey systems research. Wuhan: Press of
Huazhong University of Science and Technology.
Liu, S. F. (1997). Prediction trap of shock-disturbed systems and buffer operators. Journal of
Huazhong University of Science and Technology, 25(1), 28–31.
Liu, S. F., & Forrest, J. (1997). The role and position of grey system theory in science
development. The Journal of Grey System, 9(4), 351–356.
Liu, S. F., & Dang, Y. G. (1997). The degree of satisfaction of the floating and fixed positional
solutions of LPGP. Journal of Huazhong University of Science and Technology, 25(1), 24–27.
Liu, S. F. (1998). Methods of grey mathematics and systems analysis for science and technology
management. PhD Dissertation of Huazhong University of Science and Technology, Wuhan.
Liu, S. F., Li, B. J., et al. (1998). Evaluation criteria of regional leading industries and
mathematical models. Management Science of China, 6(2), 8–13.
Liu, S. F., & Lin, Y. (1998). An Introduction to grey systems: foundations, methodology and
applications. Grove City, PA: IIGSS Academic Publisher.
314 References

Liu, S. F., Dang, Y. G., Li, B. J., et al. (1998). Computational analysis on the periodic
contributions of technological advances in Henan Province. Journal of Henan Agricultural
University, 32(3), 203–207.
Liu, S. F., Guo, T. B., & Dang, Y. G. (1999). Grey systems theory and applications (2nd ed.).
Beijing: Science Press.
Liu, S. F., & Deng, J. L. (1999). GM (1, 1) coding for exponential series. The Journal of Grey
System, 11(2), 147–152.
Liu, S. F., & Deng, J. L. (2000). Range of applicability of the GM (1, 1) model. Systems
Engineering: Theory and Practice, 20(5), 121–124.
Liu, S. F., & Wang, Z. Y. (2000). Entropy of grey evaluation coefficient vector. The Journal of
Grey System, 12(3), 323–326.
Liu, S. F., Zhao, L., & Wang, Z. Y. (2001). A new method for evaluating risky investment.
Management Science of China, 9(2), 22–26.
Liu, S. F. (2002). Evaluation criteria for comprehensive scientific and technological strength and
mathematical models. Journal of Nanjing University of Aeronautics and Astronautics, 32(5),
409–412.
Liu, S. F. (2003). The appearance, development and current state of grey systems theory.
Management Science of China, 16(4), 14–17.
Liu, S. F., & Wang, R. L. (2003). Grey incidence analysis of the third industries during the ninth
fifth period in Nanjing City. Nanjing University of Science and Technology, 16(5), 55–58.
Liu, S. F., & Lin, Y. (2004). An axiomatic definition for the degree of greyness of grey numbers.
IEEE System Man and Cybernetics, 2420–2424.
Liu, S. F., Tang, X. W., Yuan, C. Q., et al. (2004). On the order of Chinese industrial structures.
Economic Advances, 5, 53–56.
Liu, S. F. (2004). Appearance and development of grey systems theory. Journal of Nanjing
University of Aeronautics and Astronautics, 36(2), 267–272.
Liu, S. F., & Dang, Y. G. (2004). Technical change and the funds for science and technology.
Kybernetes: The International Journal of Systems & Cybernetics, 33(2), 295–302.
Liu, S. F., Dang, Y. G., & Lin, Y. (2004). Synthetic utility index method and venturous capital
decision-making. Kybernetes: The International Journal of Systems & Cybernetics, 33(2),
288–294.
Liu, S. F. (2006). On index system and mathematical model for evaluation of scientific and
technical strength. Kybernetes: The International Journal of Systems & Cybernetics, 35(7-8),
1256–1264.
Liu, S. F., & Lin, Y. (2006). Grey information: theory and practical applications. London:
Springer-Verlag.
Liu, S. F., Lin, Y., et al. (2006). On measures of information content of grey numbers. Kybernetes,
35(5), 899–904.
Liu, S. F., Fang, Z. G., & Lin, Y. (2006). Study on a new definition of degree of grey incidence.
Journal of Grey System, 9(2), 115–122.
Liu, S. F., & Xie, N. M. (2009). A new grey evaluation method based on improved triangular
whitenization weight functions. Journal of Systems Engineering, in press.
Liu, S. F., Xie, N. M., & Forrest, J. (2009). A class of new models of grey incidence analysis.
Systems Engineering: Theory and Practice, in press.
Liu, S. F., Wan, S. Q., et al. (2009). Models for evaluating the importance of the rescuers’
locations in terms of accidents in complex traffic networks. Management Science of China, 17
(1), 119–124.
Liu, S. F., & Lin, Y. (2010). Advance in Grey Systems Research. Springer-Verlag.
Liu, S. F., & Yuan, W. F. (2010). Multi-attribute intelligent grey target decision model. Control
and Decision, 25(8), 1159–1163.
Liu, S., & XN, Fang Zhigeng. (2010). Algorithm rules of interval grey numbers based on the
“Kernel” and the degree of greyness of grey numbers. Systems Engineering and Electronics, 32
(2), 313–316.
References 315

Liu, S. F., & Lin, Y. (2011). Grey Systems Theory and Applications. Springer-Verlag.
Liu, S. F., & Xie, N. M. (2011). A new grey evaluation method based on reformative triangular
whitenization weight function. Journal of Systems Engineering, 26(2), 244–250.
Liu, S., FangZhigeng, Y. Y., et al. (2012). General grey numbers and its operations. Grey Systems:
Theory and Application, 2(3), 341–349.
Liu, S. F., Yang, Y. J., Cao, Y., et al. (2013). A summary on the research of GRA models. Grey
Systems: Theory and Application, 3(1), 7–15.
Liu, S. F., Yang, Y. J., Wu, L. F., et al. (2014). Grey system theory and its application (7th edn.).
Liu, S., Fang, Z., & Yang, Y. (2014). On the two stages decision model with grey synthetic
measure and a betterment of triangular whitenization weight function. Control and Decision,
29(7), 1232–1238.
Liu, S., Forrest, J., & Yang, Y. (2015). Grey system: Thinking, methods, and models with
applications. In Contemporary Issues in Systems Science and Engineering (pp. 153–224). John
Wiley & Sons, Inc.
Liu, S., Zeng, B., Liu, J., et al. (2015). Four basic models of GM (1, 1) and their suitable
sequences. Grey System: Theory and Application, 5(2), 41–156.
Liu, S., Yang, Y., Fang, Z., et al. (2015). Grey cluster evaluation models based on mixed triangular
whitenization weight functions. Grey Systems: Theory and Application, 5(3), 410–418.
Liu, S., Zhang, S., Jian, L., et al. (2015). Performance evaluation of large commercial aircraft
vendors. The Journal of Grey System, 27(1), 1–11.
Liu, S. F., et al. (2016). Grey system theory and its application (8th ed.). Beijing: The Science
Press.
Liu, S., Tao, L., & Xie, N., et al, (2016). On the new model system and framework of grey system
theory. The Journal of Grey System, 28(1), 1–15.
Liu, S., Yang, Y., Xie, N., & Forrest, J. (2016). New progress of Grey system theory in the new
millennium. Grey Systems Theory and Application, 6(1), 2–31.
Liu, W. (2013). Generalized incidence analysis model of interval grey number. Journal of
Zhengzhou University, 45(2), 41–44, 89.
Liu, Y., Chen, X., Zhang, G., et al. (2004). Application of mining algorithm based on gray
association rule in aluminium electrolysis control. The Chinese Journal of Nonferrous Metals,
14(3), 494–498.
Liu, Y., Chen, S., Zhang, M., et al. (2006). The application in target tracking of buffer operator and
data fusion technology. Journal of Applied Sciences, 24(2), 154–158.
Liu, Y., Jian, L., et al. (2013). Probabilistic decision method of hybrid grey cluster, variable
precision rough sets and fuzzy set with application. Journal of Management Engineering, 27
(3), 110–115.
Liu, Z. Y. (1995). Grey production functions with n kinds of input factors. Systems Engineering:
Theory and Practice, 15(9), 6–8.
Liu, Z. Y. (2000). Two fundamental problems of grey systems analysis. Systems Engineering:
Theory and Practice, 20(9), 123–124.
Lloret-Climent, M., & Nescolarde-Selva, J. (2014). Data analysis using circular causality in
networks. Complexity, 19(4), 15–19.
Lu, X., & Wang, C. (2013). Simulation on ATO speed controller based on grey prediction control.
City Track Traffic Development Research, 2, 62–65.
Luo, D. (2006). A characteristic vector method for grey decision making. Systems Engineering:
Theory and Practice, 24(4), 67–71.
Luo, D., & Wang, X. (2012). The multi-attribute grey target decision method for attribute value
within three-parameter interval grey number. Applied Mathematical Modelling, 36(5), 1957–
1963.
Luo, D., & Liu, S. F. (2004). Combined grey and rough decision making models. Journal of
Xiamen University, 43(1), 26–30.
Luo, D., & Liu, S. F. (2004). On grey dynamic programming. Systems Engineering: Theory and
Practice, 24(4), 56–62.
316 References

Luo, D., & Liu, S. F. (2005). Grey incidence decision-making methods for systems with
incomplete information. Journal of Applied Science, 23(4), 408–412.
Luo, D., & Liu, S. F. (2005). Study on grey incidence decision-making methods. Management
Science of China, 13(1), 101–106.
Luo, D., Liu, S. F., & Dang, Y. G. (2003). Optimization of grey GM (1, 1) model. Engineering
Science of China, 5(8), 50–53.
Luo, D., & Wang, J. (2012). Theory and methods of grey decision-making. Science Press.
Luo, M. F. (1994). Fault detection, diagnosis and prognosis using grey system theory. Monash
University, A Dissertation Submitted for the Degree of PhD in Engineering.
Luo, Q. C. (1989). Design and application of grey optimized input-output models. Systems
Engineering: Theory and Practice, 9(5), 55–59.
Luo, R. C., & Chen, T. M. (2000). Autonomous mobile target tracking system based on grey-fuzzy
control algorithm. IEEE Transactions on Industrial Electronics, 47(4), 920–931.
Luo, X. M., & Yang, H. H. (1994). Grey comprehensive evaluation models. Systems Engineering
and Electronic Technology, 16(9), 18–25.
Lv, A. L., & Shao, Y. (1996). Diagnosis models for difficult diseases. In New advances in grey
systems research. Wuhan: Press of Huazhong University of Science and Technology.
Lv, A. L. (2000). Theory and application of incidence spaces of medical grey information. Wuhan:
PhD. Dissertation of Huazhong University of Science and Technology.
Lv, D. G., Wang, L., & Zhang, P. (2007). A grey incidence method for fuzzy multi-attribute
decision-making regarding structural design. Journal of Harbin University of Technology,
39(6), 841–844.
Lv, F. (1997). Study on the distinguishing coefficients of grey systems incidence analysis. Systems
Engineering: Theory and Practice, 17(6), 49–54.
Lv, L. Z., & Wu, W. J. (2001). Optimization of the grey GM (1, 1) model. Systems Engineering:
Theory and Practice, 21(8), 92–96.
Mahdaviani, S. H., Parvari, M., & Soudbar, D. (2016). Simultaneous multi-objective optimization
of a new promoted ethylene dimerization catalyst using grey relational analysis and entropy
measurement. Korean Journal of Chemical Engineering, 33(2), 423–437.
Mao, S., Gao, M., & Xiao, X. (2015). Fractional order accumulation time-lag GM (1, N, s) model
and its application. Systems Engineering—Theory & Practice, 35(2), 430–436.
Memon, M. S., Lee, Y. H., & Mari, S. I. (2015). Group multi-criteria supplier selection using
combined grey systems theory and uncertainty theory. Expert Systems with Applications,
42(21), 7951–7959.
Meng, X., Wang, C., He, F., et al. (2012). On life prediction of gun tube based on combined model
of grey linear regression. Journal of Nanjing University of, Science and Technology, 36(4),
635–638.
Mi, C. M., Liu, S. F., & Yang, J. (2004). Grey incidence analysis of Jiangsu Province’s investment
in science and technology and economic growth. Science of Sciences and Management of
Science and Technology, 25(1), 34–36.
Mi, C. M., Liu, S. F., Wu, Z. P., et al. (2009). Study on the sequence of strengthening buffer
operators of the opposite directional accumulation. Control and Decision Making, 24(3),
352–355.
Mi, G., Yang, R., & Liang, L. (2014). Research on the method of diagnosis of complex fault in
track circuit based on combined model. Journal of the China Railway Society, 36(10), 65–69.
Modigliani, F., & Brumbergh, R. (1954). Utility analysis and the consumption function: An
interpretation of cross-section data. In K. Kurihara (Ed.), Post Keynesian Economics. London:
George Allen and Unwin.
Mohammadreza, S., Seyed Hossein Razavi, H., & Shide Sadat, H. (2013). A fuzzy grey goal
programming approach for aggregate production planning. The International Journal of
Advanced Manufacturing Technology, 64(9-12), 1715–1727.
Morán, J., Granada, E., Míguez, J. L., et al. (2006). Use of grey relational analysis to assess and
optimize small biomass boilers. Fuel Processing Technology, 87(2), 123–127.
References 317

Mousavi, S. M., Mirdamadi, S., Siadat, A., et al. (2015). An intuitionistic fuzzy grey model for
selection problems with an application to the inspection planning in manufacturing firms.
Engineering Applications of Artificial Intelligence, 39, 157–167.
Mu, R., & Zhang, J. Q. (2008). Layered comprehensive evaluation based on grey incidence
analysis. Systems Engineering: Theory and Practice, 28(10), 125–130.
Nagesh, S., Murthy, H. N., Narasimha, P. R., et al. (2015). Influence of nanofillers on the quality
of CO2 laser drilling in vinylester/glass using orthogonal array experiments and grey relational
analysis. Optics and Laser Technology, 69, 23–33.
Nirmal, S. K., Rakesh, S., & Vishal, S. S. (2013). Grey-based Taguchi analysis for optimization of
multi-objective machining process in turning. Strategic Decision Science, 4(2), 110–127.
Olson, D. L., Jijun, Z., & Desheng, W. (2005). The method of grey related analysis to multiple
attribute decision making problems with interval numbers. Mathematical and Computer
Modelling, 42(9–10), 991–998.
Olson, D. L., & Wu, D. S. (2006). Simulation of fuzzy multi-attribute models for grey
relationships. European Journal of Operational Research, 175(1), 111–120.
Ossowski, M., & Korzybski, M. (2013). Data mining based algorithm for analog circuits fault
diagnosis. Przeglad Elektrotechniczny, 89(2a), 285–287.
Oztaysi, B. (2014). A decision model for information technology selection using AHP integrated
TOPSIS-Grey: The case of content management systems. Knowledge-Based Systems, 70,
44–54.
Palanci, O., Gok, S. Z., Alparslan, Ergun, S., et al. (2015). Cooperative grey games and the grey
Shapley value. Optimization, 64(8), 1657–1668.
Pawlak, Z. (1982). Rough sets. International Journal of Computer & Information Sciences, 11(5),
341–356.
Pawlak, Z. (1991). Rough sets: theoretical aspects of reasoning about data. Dordrecht: Kluwer
Academic Publishers.
Peng Fang, W., & Guoping, F. M. (2005). Grey programming cluster and application in evaluation
of oil and gas cap layer. Journal of Hunan University of Science and Technology, 20(2), 5–10.
Phillips, A. W. H. (1958). The relation between unemployment and the rate of change of money
wage rates in the United Kingdom, 1861–1957. Economica, n.s., 25(2), 283–299.
Pitchipoo, P., Venkumar, P., & Rajakarunakaran, S. (2015). Grey decision model for supplier
evaluation and selection in process industry: a comparative perspective. International Journal
of Advanced Manufacturing Technology, 76(9–12), 2059–2069.
Prasad, K., Chalamalasetti, S., S. R., & Damera, N. R. (2015). Application of grey relational
analysis for optimizing weld bead geometry parameters of pulsed current micro plasma arc
welded in conel 625 sheets. International Journal of Advanced Manufacturing Technology,
78(1–4), 625–632.
Qi, M., Deng, L., Ge, D., et al. (2007). Decomposition method for colored medical figures
designed by using the maximum threshold entry value of the grey incidence space. Journal of
Chinese University of Science and Technology, 37(12), 1543–1545.
Qian, W. Y., & Dang, Y. G. (2009). The GM (1, 1) model based on fluctuating sequences. Systems
Engineering: Theory and Practice, 29(3), 149–154.
Qian, W., Dang, Y., & Liu, S. (2012). Grey GM (1, 1, t) model with time power and application.
Systems Engineering Theory and Practice, 32(10), 2247–2252.
Qiao, G., Zhang, W., & Xueshan (2009). Speed control based on fuzzy PID control with grey
prediction in the deep sea stepping system. Journal of China Coal Society, 34(11), 1550–1553.
Qiu, W., Li, S., Zhao, Q. L., et al. (2007). Grey evaluation and model prediction of the forest
coverage in Heilongjiang Province. Journal of Harbin University of Technology, 39(10),
1650–1652.
Rajesh, R., & Ravi, V. (2015). Supplier selection in resilient supply chains: a grey relational
analysis approach. Journal of Cleaner Production, 86, 343–359.
Rajeswari, K., Lavanya, S., & Lakshmi, P. (2015). Grey fuzzy sliding mode controller for vehicle
suspension system. Control Engineering and Applied Informatics, 17(3), 12–19.
318 References

Ramesh, S., Viswanathan, R., & Ambika, S. (2016). Measurement and optimization of surface
roughness and tool wear via grey relational analysis, TOPSIS and RSA techniques.
Measurement, 78, 63–72.
Rao, C. J., Xiao, X. P., et al. (2006). Generalized accumulated generating operation and its
generating space. Dynamics of Continuous Discrete and Impulsive Systems-series
B-Applications & Algorithms, 13, 517–521.
Rao, R., & Yadava, V. (2009). Multi-objective optimization of Nd:YAG laser cutting of thin
superalloy sheet using grey relational analysis with entropy measurement. Optics & Laser
Technology, 41(8), 922–993.
Ruan, A. Q., & Liu, S. F. (2007). Grey GERT networks and precision of grey estimates based on
customers’ demand. Systems Engineering, 25(12), 100–104.
Sadeghi, M., Rashidzadeh, M., & Soukhakian, M. (2012). Using analytic network process in a
group decision-making for supplier selection. Informatica, 23(4), 621–643.
Sahoo, S., Dhar, A., & Kar, A. (2016). Environmental vulnerability assessment using grey analytic
hierarchy process based model. Environmental Impact Assessment Review, 56, 145–154.
Salmeron, J. L. (2010). Modelling grey uncertainty with fuzzy grey cognitive maps. Expert
Systems with Applications, 37, 7581–7588.
Salmeron, J. L., & Gutierrez, E. (2012). Fuzzy grey cognitive maps in reliability engineering.
Applied Soft Computing, 12, 3818–3824.
Samet, H., & Mojallal, A. (2014). Enhancement of electric arc furnace reactive power
compensation using Grey-Markov prediction method. IET Generation Transmission &
Distribution, 8(9), 1626–1636.
Sarpkaya, C., & Sabir, E. C. (2016). Optimization of the sizing process with grey relational
analysis. Fibres & Textiles in Eastern Europe, 24(1), 49–55.
Senthilkumar, N., Sudha, J., & Muthukumar, V. (2015). A grey-fuzzy approach for optimizing
machining parameters and the approach angle in turning AISI 1045 steel. Advances in
Production Engineering & Management, 10(4), 195–208.
Sevastjanov, P., & Dymova, L. (2009). A new method for solving interval and fuzzy equations:
linear case. Information Sciences, 179, 925–937.
Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk.
Journal of Finance, 19(3), 425–442.
Shen, V., Chung, Y. F., & Chen, T. S. (2009). A novel application of grey system theory to
information security. Computer Standards & Interfaces, 31(2), 277–281.
Shui, N. X., & Qing, T. C. (1998). Some theoretical problems about grey systems’ GM (1, 1)
model. Systems Engineering: Theory and Practice, 18(4), 59–63.
Singh, T., Patnaik, A., & Chauhan, R. (2016). Optimization of tribological properties of cement
kiln dust-filled brake pad using grey relation analysis. Materials & Design, 89, 1335–1342.
Song, Z. M. (2003). Order structures of grey systems. PhD Dissertation of Huazhong University of
Science and Technology, Wuhan.
Song, Z. M., Liu, X. Q., & Wang, S. Z. (1997). Curvature simulation method of grey exponential
curves. Systems Engineering: Theory and Practice, 17(6), 55–57.
Song, Z. M., Tong, X. J., & Xiao, X. P. (2001). Grey GM (1, 1) model of the type of central
approximation. Systems Engineering: Theory and Practice, 21(5), 110–113.
Song, Z. Q., & Tan, C. Q. (1997). Analysis criteria, treatments and applications of refined
evaluation of petroleum deposits using grey systems theory. Systems Engineering: Theory and
Practice, 17(3), 74–82.
Song, Z. M., & Ning, X. X. (2008). Construction and operation of a grey evaluation system of
virtual firms. Systems Engineering, 26(1), 20–24.
Su, C., & Liu, S. (2008). On the asymptotic stability of grey stochastic linear delay systems.
Control and Decision, 23(5), 571–574.
Su, C., & Liu, S. (2009). On robust stability of p-moment index of stochastic system with
distributed delay and the interval parameters. Applied Mathematics and Mechanics, 30(7),
856–864.
References 319

Sun, C. (2005). Present situation and prospect of online monitoring and diagnosis technology of
the state of power transmission and transformation equipment. China Power, 38(2), 1–7.
Sun, C. X., Li, J., & Zheng, H. P. (2002). A new method of faulty insulation diagnosis in power
transformer based on degree of area incidence analysis. Power System Technology, 26(7),
24–29.
Sun, C. X., Bi, W. M., & Zhou, Q. (2003). New gray prediction parameter model and its
application in electrical insulation fault prediction. Control Theory & Applications, 20(5),
798–801.
Sun, X. D., Jiao, Y., & Hu, J. S. (2005). A decision-making method based on grey incidences and
ideal solutions. Management Science of China, 13(4), 63–68.
Sutanto, S., Go, A. W., Ismadji, S., et al. (2015). Taguchi method and grey relational analysis to
improve in situ production of FAME from sunflower and Jatropha curcas Kernels with
subcritical solvent mixture. Journal of the American Oil Chemists Society, 92(10), 1513–1523.
Tabaszewski, M., & Cempel, C. (2015). Using a set of GM (1, 1) models to predict values of
diagnostic symptoms. Mechanical Systems and Signal Processing, 52–53, 416–425.
Tamura, Y., Zhang, D. P., Umeda, N., & Sakeshita, K. (1992). Load forecasting using grey
dynamic model. The Journal of Grey System, 4(4), 49–58.
Tan, G., Wang, L., & Cheng, Y. (2011). Prediction method for cable tension state of cable-stayed
bridges based on grey system theory in could areas. Journal of Jilin University (Engineering
and Technology Edition), 41(2), 170–173.
Tan, J. B., & Lv, Y. J. (2008). Weak uniformity and uniformity of grey judgment matrices and
their properties. Systems engineering: Theory and Practice, 28(3), 159–165.
Tan, X. R. (1997). Grey medical incidence theory and applications. PhD Dissertation of Huazhong
University of Science and Technology, Wuhan.
Tang, K., Zhou, N., & Fan, X. (2012). Analysis on the factors influencing the gas well productivity
of S2 gas pool in Permian of Zizhou gas field. Computing Techniques for Geophysical and
Geochemical Exploration, 34(6), 723–728.
Tang, W. M. (2006). A new prediction model based on grey support vector machine. Journal of
Systems Engineering, 21(4), 410–413.
Tang, W. X. (1995). A new method for estimating the parameters of the GM (1, 1) model and
hypothesis testing. Systems Engineering: Theory and Practice, 15(3), 20–49.
Thananchai, L. (2008). Grey prediction on indoor comfort temperature for HVAC systems. Expert
Systems with Applications, 34(4), 2284–2289.
Tian Jianyan, L. Y. (2007). Research on grey prediction model of the slab temperature in heating
furnace. Journal of Northeast University, 28(S1), 6–10.
Tong, X. J., Chen, M. Y., & Zhou, L. (2002). On AGO Effect of the Grey Model[J]. Systems
Engineering-Theory and Practice, 22(11), 121–125.
Twala, B. (2014). Extracting grey relational systems from incomplete road traffic accidents data:
the case of Gauteng Province in South Africa. Expert Systems, 31(3), 220–231.
Vallee, R. (2008). Book Reviews: grey information: theory and practical applications[J].
Kybernetes, 37(1), 89.
Varun, A., & Venkaiah, N. (2015). Grey relational analysis coupled with firefly algorithm for
multi-objective optimization of wire electric discharge machining. Proceedings of the
Institution of Mechanical Engineers Part B-Journal of Engineering Manufacture, 229(8),
1385–1394.
Verma, A., Sarangi, S., & Kolekar, M. H. (2014). Stator winding fault prediction of induction
motors using multiscale entropy and grey fuzzy optimization methods. Computers & Electrical
Engineering, 40(7), 2246–2258.
Wang, H. Y., Liu, L., & Liu, L. (2007). Applied study of BP neural networks based on GRA and
PCA. Management Review, 19(10), 50–55.
Wang, J., & Liu, S. (2009). On measuring and sorting for the efficiency index of interval DEA
based 9on interval position and grey incidence model. Systems Engineering and Electronics,
31(6), 2146–2150.
320 References

Wang, J. J., & Jing, Y. Y. (2008). Integrated evaluation of distributed triple-generation systems
using improved grey incidence approach. Energy, 33(9), 1427–1437.
Wang, J. Q., Ren, S. X., & Chen, X. H. (2009). Preference ordering of grey stochastic
multi-criteria decision makings. Control and Decision Making, 24(5), 701–705.
Wang, L. (2009). Study on cooperation mechanism of one to many in perishable products system.
Journal of Huazhong University of Science and Technology, 37(8), 12–15.
Wang, W., Wu, M., Cao, W., et al. (2010). Fuzzy-expert control based on combination grey
prediction model for flue temperature in coke oven. Control and Decision, 25(2), 185–190.
Wang, W. P. (1993). Analysis of optimal values in GLP. The Journal of Grey System, 5(4),
315–317.
Wang, W. P. (1997). Study on grey linear programming. The Journal of Grey System, 9(1), 41–46.
Wang, W. P., & Deng, J. L. (1992). Intension measurement of information contained in grey
systems propositions. Grey Systems Theory and Practice, 2(1), 41–43.
Wang, W. P. (1994). Theory and methods on dealing with grey information. PhD. Dissertation of
Huazhong University of Science and Technology, Wuhan.
Wang, X. C. (2007). On relationship between rough sets and grey systems. Fuzzy Systems and
Mathematics, 20(6), 129–135.
Wang, X., & Nie, H. (2008). On a method of fatigue life prediction based on grey system model
GM (1, 1). Transaction of Nanjing University of Aeronautics and Astronautics, 40(6),
845–848.
Wang, X. Y., & Yang, X. (1997). Grey systems methods for quality predictive control and
diagnosis. Systems Engineering: Theory and Practice, 17(5), 105–108.
Wang, X. M. (1993). Grey dynamic models for analyzing economic growths and cycles. Systems
Engineering: Theory and Practice, 13(1), 42–47.
Wang, X. M., & Luo, J. J. (1986). Collected programs for grey systems prediction,
decision-making, and modeling. Beijing: Press of Science Dissemination.
Wang, Y., Cheng, Z., Wang, H., et al. (2011). Application of gray relational cluster method in
muon tomography. Nuclear Electronics & Detection Technology, 31(8), 871–873.
Wang, Y. M., Dang, Y. G., & Wang, Z. X. (2008). Optimization of the background values of
non-equal-distant GM (1, 1) model. Management Science of China, 16(4), 159–162.
Wang, Y. H., & Dang, Y. G. (2009). A comprehensive posterior evaluation method based on the
grey fixed weight clustering of the D-S evidence theory. Systems Engineering: Theory and
Practice, 29(5), 123–128.
Wang, Y., & Cao, Y. (2010). Gray neural network model of aviation safety risk. Journal of
Aerospace Power, 25(5), 1036–1042.
Wang, Y., Zhou, T., Zhang, M., et al. (2013). Application of grey relational analysis in Yaojialing
Zn-Au polymetallic deposit prediction. Journal of Hefei University of Technology, 36(10),
1236–1241.
Wang, Z. X., Dang, Y. G., & Liu, S. F. (2008). An optimized GM (1, 1) model based on discrete
exponential functions. Systems Engineering: Theory and Practice, 28(2), 61–67.
Wang, Z. X., Dang, Y. G., & Liu, S. F. (2009). Variable buffer operators and their acting strengths.
Control and Decision Making, 29(8), 1218–1222.
Wang, Z. X., Dang, Y. G., & Liu, S. F. (2008). Two-piece grey models and applications. Systems
Engineering: Theory and Practice, 28(11), 109–114.
Wang, Z. X., Dang, Y. G., & Song, C. P. (2009). Multi-objective grey situational decision-making
models based on interval numbers. Control and Decision Making, 24(3), 388–392.
Wang, Z., Dang, Y., & Liu, S. (2009). On buffer operators with variable weight and supplement to
buffer operator axioms. System Engineering, 27(1), 113–117.
Wang, Z. (2013). On derivative model of power model GM (1, 1). Systems Engineering Theory
and Practice, 33(11), 2894–2902.
Wang, Z. L., Liu, Y. L., & Shi, K. Q. (1995). On Kenning grey degree. The Journal of Grey
System, 7(2), 103–110.
References 321

Wang, Z. L. (1993). Application of grey systems theory in analysis of cultivars. In Collected

papers of grey systems research (pp. 305 – 307). Kaifeng: Press of Henan University.
Wang, Z. L. (1998). A theory for grey modeling techniques. Ph.D. Dissertation of Huazhong
University of Science and Technology, Wuhan.
Wang, Z. L., & Li, X. Z. (1996). Ordering and inequalities of grey numbers. In New Advances of
Grey Systems Research (pp. 364–366). Wuhan: Press of Huazhong University of Science and
Technology.
Wei, H., Lin, X., Zhang, Y., et al. (2013). Research on the application of grey system theory in the
pattern recognition for chromatographic fingerprints of traditional Chinese medicine. Chinese
Journal of Chromatography, 31(2), 127–132.
Wei, Y., Kong, X., & Hu, D. (2011). A kind of universal constructor method for buffer operators.
Grey Systems: Theory and Application, 1(2), 178–185.
Wei, Y., & Zeng, K. (2015). The simplified relational axioms and the axiomatic definition of
special incidence degrees. Systems Engineering - Theory & Practice, 35(6), 1528–1534.
Wong, H., Hu, B. Q., & Xia, J. (2006). Change-point analysis of hydrological time series using
grey relational method. Journal of Hydrology, 324(1), 323–338.
Wu, A. X., Xi, Y., Yang, B. H., Chen, X. S., et al. (2007). Study on grey forecasting model of
copper extraction rate with bioleaching of primary sulfide ore. Acta Metallurgica Sinica, 20(2),
117–128.
Wu, C. C., & Chang, N. B. (2003). Grey input–output analysis and its application for
environmental cost allocation. European Journal of Operational Research, 145(1), 175–201.
Wu, C. C., & Chang, N. B. (2003). Global strategy for optimizing textile dyeing manufacturing
process via GA-based grey nonlinear integer programming. Computers & Chemical
Engineering, 27(6), 833–854.
Wu, C. C., & Chang, N. B. (2004). Corporate optimal production planning with varying
environmental costs: A grey compromise programming approach. European Journal of
Operational Research, 155(1), 68–95.
Wu, D. D., & Olson, D. L. (2010). Fuzzy multiattribute grey related analysis using DEA.
Computers & Mathematics with Applications, 60(1), 166–174.
Wu, J. H., & Wen, K. L. (2001). Rolling error in GM (1, 1) modeling. Journal of Grey System, 13
(1), 77–80.
Wu, L. F., Liu, S. F., Yao, L. G., et al. (2013). Grey system model with the fractional order
accumulation. Communications in Nonlinear Science and Numerical Simulation, 18(7),
1775–1785.
Wu, L., Liu, S., Fang, Z., et al. (2015). Properties of the GM (1, 1) with fractional order
accumulation. Applied Mathematics and Computation, 252, 287–293.
Wu, L., Liu, S., & Yang, Y. (2015). A model to determine OWA Weights and its application in
energy technology evaluation. International Journal of Intelligent Systems, 30(7), 798–806.
Wu, L., Liu, S., Yao, L., et al. (2015). Using fractional order accumulation to reduce errors from
inverse accumulated generating operator of grey model. Soft Computing, 19(2), 483–488.
Wu, L., Liu, S., & Yang, Y. (2016). A gray model with a time varying weighted generating
operator. IEEE Transactions on Systems Man Cybernetics-Systems, 46(3), 427–433.
Wu, L., Liu, S., & Yang, Y. (2016). Grey double exponential smoothing model and its application
on pig price forecasting in China. Applied Soft Computing, 9, 117–123.
Wu Liyun, W., & Zhengpeng, L. (2013). Quadratic time- varying parameters discrete grey model.
Systems Engineering Theory and Practice, 33(11), 2887–2892.
Wu, Q., & Liu, Z. T. (2009). Real formal concept analysis based on grey-rough set theory.
Knowledge-Based Systems, 22(1), 38–45.
Wu, Y., Yang, S. Z., & Tao, J. H. (1988). Discussions on grey prediction and time series
prediction. Journal of Huazhong University of Science and Technology, 16(3), 27–33.
Wu, Z. P., Liu, S. F., Cui, L. Z., Mi, C. M., & Wang, J. L. (2009). Some new weakening buffer
operators designed on monotonic functions. Control and Decision Making, 24(7), 1055–1058.
322 References

Wu, Z. J., & Wang, A. M. (2007). Grey-fuzzy comprehensive evaluations for the planning of
reconstructible manufacturing systems. Mechanic Engineering of China, 18(19), 2313–2318.
Wu, Z., Xu, B., Gu, C., et al. (2012). On comprehensive evaluation methods of the service
condition of the dam. China Science: Technological Sciences, 42(11), 1243–1254.
Xia, J., & Zhao, H. Y. (1996). Grey artificial neural network models and applications in short-term
prediction of runoffs. Systems Engineering: Theory and Practice, 16(11), 82–90.
Xia, T., Jin, X., Xi, L., et al. (2015). Operating load based real-time rolling grey forecasting for
machine health prognosis in dynamic maintenance schedule. Journal of Intelligent
Manufacturing, 26(2), 269–228.
Xia, X., Wang, Z., & Changhong. (2005). Degree of grey incidence for the quality of processing
and vibration of rolling bearing. Journal of Aerospace Power, 20(2), 250–254.
Xiao Jun, Z. W. (2009). Grey incidence analysis applied to fault diagnosis of drone crash. Journal
of Sichuan Ordnance Engineering, 30(9), 112–115.
Xiao, X. P. (2000). On parameters in grey models. Journal of Grey System, 11(4), 315–324.
Xiao, X. P. (2006). Grey relational analysis and application of hybrid index sequences. Dynamics
of Continuous Discrete and Impulsive Systems-series B-Applications & Algorithms,
13, 915–919.
Xiao, X. P. (1997). Theoretical research and remarks on quantified models of grey degrees of
incidence. Systems Engineering: Theory and Practice, 17(8), 76–81.
Xiao, X. P. (2002). On methods of grey systems modeling. PhD. Dissertation of Huazhong
University of Science and Technology, Wuhan.
Xiao, X. P., Xie, L. C., & Huang, D. R. (1995). A computational improvement of degrees of grey
incidence and applications. Mathematical Statistics and Management, 14(5), 27–30.
Xiao, X. P., Song, Z. M., & Li, F. (2004). Foundations of grey technology and applications.
Beijing: Science Press.
Xiao, X., Liu, J., & Guo, H. (2013). Properties and optimization of generalized accumulation grey
model. Systems Engineering Theory and Practice, 33(1), 1–9.
Xiao, X. P., & Wang, H. H. (2014). Change of GM (1, 1, a) model background value on the
influences of relative error. Systems Engineering - Theory & Practice, 34(2), 408–415.
Xiao, Y., Shao, D. G., Deng, R., et al. (2007). Comprehensive evaluation supporting systems for
the effects of hydraulic engineering projects. Journal of Wuhan University, 40(4), 49–52.
Xie, N. M., & Liu, S. F. (2004). Grey layered analysis method and its positional solutions. Journal
of Jiangnan University, 3(1), 87–89.
Xie, N. M., & Liu, S. F. (2005). Discrete GM (1, 1) model and modeling mechanism of grey
prediction models. Systems Engineering: Theory and Practice, 28(4), 93–99.
Xie, N. M., & Liu, S. F. (2005). Research on discrete grey model and its mechanism. In IEEE
international conference on systems, man and cybernetics (Vol. 1). IEEE systems, man and
cybernetics society, proceedings - 2005 international conference on systems, man and
cybernetics (pp. 606–610).
Xie, N. M., & Liu, S. F. (2006). Generalized discrete grey models and their optimal solutions.
Systems Engineering: Theory and Practice, 26(6), 108–112.
Xie, N. M., & Liu, S. F. (2006). A class of discrete grey models and the effects of their predictions.
Journal of Systems Engineering, 21(5), 520–523.
Xie, N. M., & Liu, S. F. (2007). Similarity and consistency of several classes of incidence models.
Systems Engineering, 25(8), 98–103.
Xie, N. M., & Liu, S. F. (2007). Research on the multiple and parallel properties of several grey
relational models. International Conference on Grey Systems and Intelligence Service
183–188.
Xie, N. M., & Liu, S. F. (2008). Discrete grey models of multi-variables and applications. Systems
Engineering: Theory and Practice, 28(4), 100–107.
Xie, N. M., & Liu, S. F. (2008). Characteristics of discrete grey models of approximately
exponential sequences. Systems Engineering and Electronic Technology, 30(5), 863–867.
References 323

Xie, N. M., & Liu, S. F. (2008). Affine characteristics of discrete grey models. Control and
Decision Making, 23(2), 200–203.
Xie, N. M., & Liu, S. F. (2008). Research on the order-keeping property of several grey relational
models. Journal of Grey System, 11(3), 157–164.
Xie, N. M., & Liu, S. F. (2009). Discrete grey forecasting model and its optimization. Applied
Mathematical Modelling, 33(1), 1173–1186.
Xie, N. M., & Liu, S. F. (2009). Research on evaluations of several grey relational models adapt to
grey relational axioms. Journal of Systems Engineering and Electronics, 20(2), 304–309.
Xie, N. M., & Liu, S. F. (2009). Ordering of grey numbers with the consideration of probability
distributions. Systems Engineering: Theory and Practice, 29(6), 169–175.
Xie, Y. M., Yu, H. P., Chen, J., et al. (2007). Design for square-cup deep-drawing based on grey
systems theory. Journal of Mechanical Engineering, 27(3), 10–15.
Xie, N. M., Liu, S. F., & Yuan, C. Q. (2014). Grey number sequence forecasting approach for
interval analysis: a case of china’s gross domestic product prediction. The Journal of Grey
System, 26(1), 45–58.
Xie Yanmin, Y., & Luping, C. J. (2007). Application of grey theory in deep drawing robust design
Chinese Journal of Mechanical Engineering, 43(3), 54–59.
Xiong, H. J., Chen, M. Y., & Ju, T. (1999). Two classes of grey models of control systems.
Journal of Wuhan Jiaotong Science and Technology, 23(5), 465–468.
Xiong, H. J., Xiong, P. F., Chen, M. Y., et al. (1999). A generalization of the grey systems SCGM
(1, h)_c model. Journal of Huazhong University of Science and Technology, 27(1), 1–3.
Xu, C., X., Xiaoyan, S., & Hailong, W. (2010). Application grey-econometrics model in traffic
volume prediction. Highway Engineering, 35(5), 34–38.
Xu, J. P. (1993). Optimizing theory on working procedure with grey numbers. The Journal of Grey
System, 5(3), 189–196.
Xu, J. P. (1995). On a kind of information grey number. The Journal of Grey System, 7(2), 111–
130.
Xu, W. X., & Zhang, Q. S. (2001). An integrated computational scheme based on on grey theory
and fuzzy mathematics. Systems Engineering: Theory and Practice, 21(4), 114–119.
Xu, Y. D., & Wu, Z. Y. (2003). Grey pricing of stocks based on limited rationality and inefficiency
of stock markets. Journal of Management Engineering, 17(2), 115–117.
Xu, Z. X., & Wu, G. P. (1993). Grey Systems Theory and Grey Prediction on mineral deposits.
Wuhan: Press of Chinese University of Geology.
Yamaguchi, D., Li, G. D., & Nagai, M. (2007). A grey-based rough approximation model for
interval data processing. Information Sciences, 177(21), 4727–4744.
Yan, L. Y., & Mao, L. X. (1998). An application of the GM (1, 1) model. Systems Engineering:
Theory and Practice, 18(10), 104–106.
Yan, S., Si-feng, L., Jian-jun, Z., et al. (2014). The ranking method of grey numbers based on
relative kernel and degree of accuracy. Control and Decision, 29(2), 315–319.
Yan, S., Liu, S., Fang, Z., et al. (2014). Method of determining weights of decision makers and
attributes for group decision making with interval grey numbers. Systems Engineering - Theory
& Practice, 34(9), 2372–2378.
Yan, S., Liu, S., Liu, J., et al. (2015). Dynamic grey target decision making method
with grey numbers based on existing state and future development trend of alternatives.
Jouranl of Intelligent & Fuzzy Systems, 28(5), 2159–2168.
Yang, J. H., & Liu, J. L. (1996). Grey information and applications. In New Advances of Grey
Systems Research (pp. 362–363). Wuhan: Press of Huazhong University of Science and
Technology.
Yang, J., & Wong, W. (2014). Improved unbiased grey model for prediction of gas supplies.
Journal of Tsinghua University (Sci. & Technol), 54(2), 145–148.
Yang, S., Ren, P., & Dang, Y. G. (2009). Opposite directional accumulation generation and
optimization of grey GOM(1,1) model. Systems Engineering: Theory and Practice, 29(8),
160–164.
324 References

Yang, T., Yang, P., Dong, X., et al. (2008). Method for predicting fault status of satellite based on
gray system theory. Computer Measurement & Control, 16(9), 1284–1285, 1307.
Yang, Y. A., Wei, J., Feng, Z. R., et al. (2008). Application of grey incidence analysis in the
selection of the optimal initial orbit of satellites. Systems Engineering and Electronic
Technology, 30(2), 308–311.
Yang, Y., Wang, S. W., Hao, N. L., Shen, X. B., & Qi, X. H. (2009). On-line noise source
identification based on the power spectrum estimation and grey relational analysis. Applied
Acoustics, 70(3), 493–497.
Yang, Y., Liu, S., & John, R. (2014). Uncertainty representation of grey numbers and grey sets.
IEEE Transactions on Cybernetics, 44(9), 1508–1517.
Yao, J., & Hu, W. (2008). Gray evaluation of operational efficiency of OTH ground-wave radar.
Armament Automation, 27(4), 12–14.
Yao, T. X., & Liu, S. F. (2009). Characteristics and optimization of discrete GM (1, 1) model.
Systems Engineering: Theory and Practice, 29(3), 142–148.
Yao, T., Liu, S., & Xie, N. (2010). Study on the properties of new information discrete GM (1, 1)
model. Journal of System Engineering, 25(2), 164–170.
Ye, J., Li, B., & Liu, F. (2014). Forecasting effect and applicability of weakening buffer operators
on GM (1, 1). Systems Engineering-Theory and Practice, 34(9), 2364–2371.
Yeh, Y. L., Chen, T. C., & Lin, C. N. (2001). Effects of data characteristics for GM (1, 1)
modeling. Journal of Grey System, 13(3), 121–130.
Yi, C. H., & Gu, P. L. (2003). Predictions on energy using buffer operators of grey sequences.
Journal of Systems Engineering, 18(2), 189–192.
Yin, J., Liang, X., Xiao, C. et al. (2012). Application of matter-element extension method based on
grey clustering theory in the ground water quality evaluation-example with Taonan City. Water
Saving Irrigation 6, 52–55.
Ying, K. C., Liao, C. J., & Hsu, Y. T. (2000). Generalized admissible region of class ratio for GM
(1, 1). Journal of Grey System, 12(2), 153–156.
Yu, F., Ke, Y., Yingzheng (2009). Decision of fault repair in aircraft assembly system of
automated docking. Computer integrated manufacturing system, 15(9), 1823–1830.
Yuan, Z., Sun, C., Yuan, Z., et al. (2005). Method of grey clustering decision to making to state
assessment of power transformer. Journal of Zhongqing University, 28(3), 22–25.
Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8, 338–353.
Zadeh, L. A. (1994). Soft computing and fuzzy logic. IEEE Software, 11(6), 48–56.
Zhao, G., Sun, Y., Xu, Y., et al. (2007). Gray decision analysis of threat estimation in antimissile
combat of surface warship. Tactical Missile Technology, 3, 32–35.
Zeng, B., Liu, S., & Meng, W. (2011). Development and Application of MSGT6.0 (Modeling
System of Grey Theory 6.0) Based on Visual C# and XML. The Journal of Grey System, 23(2),
145–154.
Zeng, B., & Liu, S. (2014). Prediction model of stochastic oscillation sequence based on amplitude
compression. Systems Engineering Theory and Practice, 34(8), 2084–2091.
Zeng, X. Y., & Xiao, X. P. (2009). Ways to generalize the GM (1, 1) model and applications.
Control and Decision Making, 24(7), 1092–1096.
Zhang, C., Ding, S., & Wang, B. (2014). Study on customization model of aircraft based on grey
incidence analysis. Traffic Information and Safety, 32(4), 131–136.
Zhang, C. K., Nie, M. L., & Wu, J. B. (2007). Partner selections for virtual firms based on
improved grey evaluations. Systems Engineering: Theory and Practice, 27(11), 54–61.
Zhang, D. H., Shi, K. Q., & Jiang, S. F. (2001). Grey predicting power load via GM (1, 1)
modified. Journal of Grey System, 13(1), 65–67.
Zhang, F., Wang, P., Xiao, Z., et al. (2010). Application of grey theory in safety evaluation of
carrier aircraft system. Aircraft Design, 30(3), 56–61.
Zhang, G., Fu, Y., & Yang, R. (2004). Novel self-adjustable grey prediction controller. Control
and Decision, 19(2), 212–215.
References 325

Zhang, J., Liang, S., Zhou, R., et al. (2012). Fault tree analysis of two teeth difference swing
movable teeth transmission based on grey correlation. Machinery Design & Manufacture, 6,
183–185.
Zhang, J. J., Wu, D. S., & Olson, D. L. (2005). The method of grey related analysis to multiple
attribute decision making problems with interval numbers. Mathematical and Computer
Modelling, 42(9), 991–998.
Zhang, K., & Liu, S.-F. (2009). A novel algorithm of image edge detection based on matrix degree
of grey incidences[J]. The Journal of Grey System, 9(3), 265–276.
Zhang, K. (2014). Multi-variables discrete grey model based on driver control. Systems
Engineering Theory and Practice, 34(8), 2084–2091.
Zhang, L., Ren, L. Q., Tong, J., et al. (2004). Study of soil-solid adhesion by grey system theory.
Progress in Natural Science, 14(2), 119–124.
Zhang, M. A., Yuan, Y. B., Zhou, J., et al. (2008). Coordinated urban economic responsibility for
the aftermath of potential disasters as analyzed using grey systems models. Systems
Engineering: Theory and Practice, 28(3), 171–176.
Zhang, N. L., Meng, X. Y., Li, Z. H., et al. (2009). Grey two-layered programming problem and its
solution. Systems Engineering: Theory and Practice, 27(11), 132–138.
Zhang, P., & Sun, Q. (2007). Prediction of the economic development of Baotou city using
comparisons of grey theory and statistics. Mathematical Statistics and Management, 26(4),
595–601.
Zhang, Q. S. (1996). Deviate information theory of grey hazy sets. PhD Dissertation of Huazhong
University of Science and Technology, Wuhan.
Zhang, Q. S. (2001). Difference information entropy in grey theory. Journal of Grey System, 13(2),
111–116.
Zhang, Q. (2007). Improving the precision of GM (1, 1) model using particle swarm optimization.
Chinese Journal of Management Science, 15(5), 126–129.
Zhang, Q. S., Deng, J. L., & Fu, G. (1995). On grey clustering in grey hazy set. The Journal of
Grey System, 7(4), 377–390.
Zhang, Q. S., Guo, X. J., & Deng, J. L. (1996). The entropy analysis method of grey incidences.
Systems Engineering: Theory and Practice, 16(8), 7–11.
Zhang, Q. S., Han, W. Y., & Deng, J. L. (1994). Information entropy of discrete grey number. The
Journal of Grey System, 6(4), 303–314.
Zhang, X., Wang, Z., & Nagai, M. (2006). Research on affective interaction models of robot.
Computer Engineering, 32(24), 6–12.
Zheng, D., Gu, C., & Wu, Z. (2005). On time varying prediction model of the deformation of slope
with multi-factors. Journal of Rock Mechanics and Engineering, 24(17), 3180–3184.
Zheng, Z. N., Wu, Y. Y., & Bao, H. L. (2001). Pathological problems of the GM (1, 1) model.
Management Science of China, 9(5), 38–44.
Zhou, C. S., & Deng, J. L. (1989). Stability analysis of grey discrete-time systems. IEEE Trans.
AC, 34(2), 173–175.
Zhou, C. S., & Deng, J. L. (1986). The stability of grey linear system. Int. J. Control, 43(1), 313–
320.
Zhou, P. (2006). A trigonometric grey prediction approach to forecasting electricity demand.
Energy, 31(14), 2839–2847.
Zhou, T., & Peng, C. H. (2008). Application of grey model on analyzing the passive natural
circulation residual heat removal system of HTR-10. Nuclear Science and Techniques, 19(5),
308–313.
Zhou, W., Dang, Y., & Xiong, P. (2013). Grey clustering model for interval grey number with
variable and fixed weights. Systems Engineering Theory and Practice, 33(10), 2590–2595.
Zhou, Z. J., & Hu, C. H. (2008). An effective hybrid approach based on grey and ARMA for
forecasting gyro drift. Chaos, Solitons & Fractals, 35(3), 525–529.
Zhu, C. H., Li, N. P., Re, D., et al. (2007). Uncertainty in indoor air quality and grey system
method. Building and Environment, 42(4), 1711–1717.
326 References

Zhu, C. Q. (2001). Grey prediction modeling via grey relational weighting. Journal of Grey
System, 13(3), 255–258.
Zhu, J., Huang, Z., Zhai, D., et al. (2012). Research on grey forecasting PID control simulation
based on strengthening buffer operator. Journal of Shanghai University of Science and
Technology, 34(4), 327–332.
Zhu, S., & Shi, L. (2013). Research on supervision of private equity investment fund based on
grey game theory. Journal of Northeast University, 34(7), 1057–1060.
Zhu, X. P., Zhi, X. Z., et al. (2002). Grey predictive control of the vibration of rotor systems.
Mechanical Science and Technology, 21(1), 97–101.
http://igss.nuaa.edu.cn/
Index

A Choice of variables, 184

Accuracy Coefficient, 75
mean relative, 211 clustering of variable weight, 114
relative, 210 constraint, 241
After-event control, 278, 279 decision, 241
All-data GM (1, 1), 166 development, 149, 152, 158–160, 165, 171,
Average image, 70, 74, 235, 237 240
Average weakening buffer operator, 51, 52, 55, equal weight cluster, 117
178, 294 fixed weight cluster, 117
Axioms for buffer operator generation, 57
analytic representation, 49 grey amplifying, 11, 20
fixed points, 48 grey evaluation, 22
information basis, 48 grey impact, 89
grey influence, 85, 89
B grey responsibility, 20, 143, 144, 149
Background value, 20, 142, 149 incidence, 21, 68, 69, 73, 74
Basic predicted value, 213, 215 inverse matrix, 134
Basic value, 33, 113 Combined model
Behavioral sequence prediction model, 9
criterion sequence, 91 Component
economic sequence, 69, 70 the maximum component, 249, 250
horizontal sequence, 144 Consecutive neighbor, 217
time sequence, 69 Constraint condition, 5
Black number, 30, 199 Contour line, 219–221
Bull’s-eye-distance, 233, 234 Contour moment, 220
Contour point, 219, 220
C Control
Candidates, 203, 204, 251 closed loop, 257
Capital elasticity, 190 open loop, 257
Capital input, 190 Control of grey predictions, 278
Catastrophe sequence Control vector, 223, 258, 259, 266, 277, 278
lower catastrophe sequence, 216 Control by remove redundant, 105, 197
upper catastrophe sequence, 215 Countermeasure
Characteristic desirable countermeasure, 231
most favorable, 92, 93, 97 set, 228
quasi-favorable, 93 Critical value, 30
quasi-preferred, 93, 97, 98
variable, 92, 93 D
Characteristic vector, 258 Date sequence of catastrophe, 216

© Springer Science+Business Media Singapore 2017 327

S. Liu et al., Grey Data Analysis, Computational Risk Management,
DOI 10.1007/978-981-10-1841-1
328 Index

Decision making Error-afterward control, 279

coefficient for decision-making, 25 Error-on-time control, 279
coefficient vector for decision-making, 25 Error-prediction control, 279
Decision scheme Error sequence
decision scheme set, 228 modelability, 161, 164
desirable scheme, 231, 232 Even difference GM (1, 1), 21, 27
Degree of grey incidence Even difference model, 147, 148, 150, 152,
absolute degree of incidence, 78 153, 155, 156
Deng’s degree of incidence, 74 Even GM (1, 1), 21, 27, 144
nearness degree of incidence, 27 Event, 6, 10, 14, 209, 227, 228
relative degree of incidence, 22 Exponential function
similitude degree of incidence, 85, 86 homogeneous exponential function, 62
synthetic degree of incidence, 84, 85, 98 non-homogeneous exponential function, 62
three-dimensional degree of incidence, 27, Exponential region, 211, 213
85, 91 Exponential sequence
Derivative, 161, 162, 164, 169, 267 homogeneous exponential sequence, 63
Development domain, 213, 215 non-homogeneous exponential sequence,
Difference information principle, 14 63
Differential restored value, 161
Disaster prediction, 215, 216 F
Distance, 88, 240, 242, 245 Feedback loop, 257, 280
Distance function, 245 Field of grey number, 285
Distinguishing coefficient, 74 Five-step modeling, 221, 222
Fluctuation term, 190, 211
E Four elements for decision-making, 7
Economic reform, 5 Function
Effect, 24, 33, 43, 48, 50, 55, 124, 130, 133, distance, 88
134, 158, 159, 184, 204, 230, 242 grey transfer, 260
Effect mapping, 229 homogeneous exponential, 62
Effect measure inverse, 72
effect measure for benefit type objective, 24 smooth continuous, 211
effect measure for cost type objective, 245 Fuzzy mathematics, 1, 2, 4, 8, 10, 11, 31
lower effect measure, 242, 243, 245
lower effect measure for moderate G
objective, 24 General grey number, 18, 27, 37–43
moderate effect measure, 24 Grey 0-1 programming, 28
synthetic effect measure matrix, 24, 244 Grey action quantity, 149
uniform effect measure, 242, 244, 245, 247 Grey characteristic value, 106
upper effect measure, 24, 242, 243, 245, Grey cluster evaluation
247 basic value of subclass, 113
upper effect measure for moderate cluster based on incidence analysis, 68
objective, 243, 245 clustering decision-making, 249
Effect value, 229–232, 234, 235, 240–246, 299 equal weight clustering coefficient, 117
Efficiency matrix, 242, 244, 245 fixed weight clustering coefficient, 117
Entropy variable weight clustering coefficient, 248
evaluation coefficient vector, 68, 106 Grey control, 2, 17, 25, 26, 257, 258, 260, 261,
information, 68, 106 276, 280
Equal grey matrices, 272, 273 Grey derivative, 161
Equal time interval sequence, 79 Grey differential link, 263, 264
Equation Grey gain matrix, 266
image, 144 Grey incidence
shadow, 77 absolute degree, 22, 27
whitenization, 145, 153, 173–175, 179 cluster, 107
Equivalent scheme, 230, 232 matrix, 109
Index 329

order, 177 layer type, 106

relative degree, 22 non-essential type, 30
relative matrix of, 92 non-synchronous, 31
relative order, 89 reduced form, 37–39, 43
synthetic degree, 22, 27, 81, 84, 85, 96 synchronous, 31, 43
synthetic matrix of, 96 wish type, 6
synthetic order, 22, 81, 84 with only lower limits, 18, 39
Grey incidence analysis, 9, 17, 21, 22, 26, 68, with only upper limits, 18, 30, 39
99, 100, 287, 288, 295 with zero center, 34
Grey incidence control, 278 Grey postponing link, 264
Grey incidence factor Grey prediction control, 282
set of grey incidence factor, 72 Grey proportional link, 262, 264
Grey incidence operator Grey target, 15, 24
average image, 70 Bull’s-eye-distance, 233, 234
averaging operator, 70, 117 one-dimensional decision-making, 231
initial image, 81 s-dimensional decision-making, 232, 233
initialing operator, 70, 73, 82 spherical grey target, 232, 233
interval image, 71 Grey target decision, 229, 233, 241
interval operator, 71 intelligent grey target decision, 24, 241,
reciprocal image, 72 287, 299
reciprocating operator, 72 multi-attribute grey target decision, 245,
reversing image, 72 296, 299
reversing operator, 72 weighted multi-attribute grey target
set of grey incidence operator, 69, 72 decision, 245, 290
Grey induction coefficient, 267, 275 Grey vibration element, 31
Grey inertia link, 262
Grey influence coefficient, 133 H
Grey integral link, 263, 264 Homogeneous Markov chain, 194
Grey linear space, 43
Grey Markov chain, 195 I
Grey matrix Image equation, 144
diagonal, 259 Imagined optimum effect vector, 234, 235
difference of, 99 Imagined optimum model, 234
inverse, 161 Imagined optimum scheme, 234
non-singular, 265 Information, 5
power of, 37 density, 31
product of, 37 fuzzy, 10, 11
scalar, 37 grey, 18, 68
scalar multiplication of, 37 increment, 35
sum of, 33 new, 15, 20, 25, 35, 57, 167, 168
symmetric, 270 old, 57, 58, 167
transposition of, 146 stochastic, 10, 45, 194, 271, 272
triangular, 107 unascertained, 209
unit, 125, 247 Information content, 14
Grey number, 17, 18, 21, 29–31, 33–35, Initial image, 70, 74, 81, 82, 84
37–39, 41–43, 197–199, 201 Input-output, 14, 25, 43, 131, 133, 134, 149,
conceptual type, 32 191
continuous, 30 Internal point, 57
discrete, 30 Internal subsystem, 291
domain, 18 Intersection, 35, 200, 219
essential type, 13, 30, 31, 276 Interval grey number, 17, 18, 20, 29, 31, 35,
information type, 13, 18, 68 37, 106, 200, 201
kernel, 18, 37, 39, 43 Interval prediction, 212, 213
330 Index

Interval prediction (cont.) development, 223

basic predicted value, 213, 215 grey characteristic, 106
development domain, 213, 214 incidence, 92, 107
highest predicted value, 213, 214 mean-value, 106
linear domain, 211 moving probability, 194
lower bound function, 211, 213 nth step moving probability, 194
lowest predicted value, 213, 215 objective structure, 276
predicted domain, 211 perturbation grey, 20
proportional domain, 212 scalar matrix, 37, 82, 83, 250
trumpet-like domain, 211 state, 223, 258, 259
uniform domain, 211 synthetic effect measures, of, 24, 242, 244,
upper bound function, 211, 213 245, 247
value domain, 211 uniform effect measures, of, 241, 245, 247
wrapping domain, 212 Mean, 31, 43, 52, 106, 134, 180, 273
Interval value image, 71 Mean-generating operator, 58
adjacent neighbor, 58
K non-adjacent neighbor, 58
Knowledge base, 7, 10, 14, 45, 183, 197, 200, Mean slope, 72, 212
202, 203, 280 Membership function, 10, 31, 197, 198, 200
K-zigzagged line, 219 Metabolic GM (1, 1), 166
Model
L catastrophe GM (1, 1), 216
Labor elasticity, 190 Cobb—Douglas, 8, 25, 190
Labor input, 190 combined prediction, 9, 187
Laplace transformation, 260, 261, 265 comprehensively balanced, 24, 124, 242,
Law of grey exponent, 63 246
law of negative grey exponent, 63 critical, 19, 30, 200
law of positive grey exponent, 63 error modification, 162
law of quasi grey exponent, 63 error-satisfactory, 210
Length, 33, 38, 40, 58, 77–79, 81, 82, 84–86, general mathematical, 10, 55, 221, 258,
92, 271 271, 280
information field, 34 GM (0, N), 169, 176, 178
Length of a sequence, 77, 78 GM (1, 1), 21, 141, 142, 144, 145, 148,
Levels of accuracy, 286 149, 151, 159, 165, 189, 191
Limit value, 30 GM (1, 1) power, 20
Linear region, 8, 43, 169, 183, 186, 187, 189, grey econometrics, 183, 184, 187
211, 258, 265, 266, 271 grey input-output, 25, 43
Log-linear form, 190 grey production function, 191
Lower bound, 29, 30, 38, 211, 213, 215, 271 grey state, 195
Lower effect measure, 24, 242, 245 grey Verhulst, 21, 173–176, 178
ideal, 24, 242, 244, 245, 286
M incidence-satisfactory, 210
Markov chain, 194, 195 optimal grey input-output, 43
Mathematical model of linear control system, remnant GM (1, 1), 160, 162, 163
259, 260, 265 Modeling
controllable, 259 five-step, 221, 222
observable, 259, 260 grey systems, 55, 58, 60, 64, 149, 177, 191,
output equation, 259 266, 285, 286, 294
output grey matrix, 25 Model simulation, 142
state equation, 259 Moderate effect measure, 24, 245
state grey matrix, 265 Monotonic sequence
Matrix decreasing sequence, 48, 50, 54, 55, 72,
cluster coefficient, 114 240
control, 259 increasing sequence, 48–50, 54, 72
Index 331

Most favorable factor, 92, 93, 97 Partial-data GM (1, 1), 166

Moving probability, 194 Partition, 105
Positioned coefficient, 31
N positioned coefficient of grey number, 31
New data, 57, 168, 280 synchronous, 31
New information GM (1, 1), 20, 166 Possibility function, 10, 14, 22–24, 31–33,
Non-adjacent neighbor generating operator, 57 105, 106, 110–114, 116–118, 121
Non-preference generation, 57 lower measure, 23, 111, 113, 122, 127, 130
Non-uniqueness principle, 191 moderate measure, 112, 113
Norm, 268 typical form, 111–113
continuous function, 269 upper measure, 23, 112, 113, 122, 127, 130
Normality, 73, 243 Power model, 173
Prediction control, 282
O Prediction value, 213, 215
Objective, 229, 242 Preference generation superiority, 9, 92, 93,
Objective function, 233 133, 139, 230
Old data, 169, 280 Preimage, 184
On-time control, 279 Principle of
Operations of grey number, 18 absoluteness of Greyness, 16, 191
addition, 191 informational differences, 14
division, 18, 36, 38, 41, 42 minimal information, 15
equality, 26 new information priority, 15, 25
multiplication, 18, 36, 38, 41 non-uniqueness, 14, 191
multiplication by a scalar, 33, 37, 82, 83 recognition base, 15
negative element, 29 Probability, 10, 31, 39, 40, 67, 180, 194–197,
power, 15, 37, 38 210, 211, 272
reciprocal, 36, 72 Progress in technology, 1, 190
subtraction, 18, 36 Property
Operator closeness, 73, 84–87, 89, 287
averaging, 70, 117 expendability, 175
buffer, 19, 20, 47–50, 52–142, 294 maximum valuation, 138
first order, 48, 49 non-negativeness, 191
grey incidence, 69, 72, 98 symmetry, 281, 282
initialing, 70, 73, 82 Proportional band, 212
interval, 71
reciprocating, 72, 161 Q
reversing, 72 Quasi-favorable factor, 92
strengthening, 19, 49, 50, 54–56, 287 Quasi-optimal countermeasure, 232
weakening, 19, 49–53, 55, 287 Quasi-optimal event, 232
zero starting point, 77, 79, 80, 83, 91, 95 Quasi-optimal scheme, 232, 233, 235
Optimum Quasi-optimal solution, 222
countermeasure, 232, 244 Quasi-preferred factor, 93, 97, 98
decision scheme, 235, 238, 244 Quasi-smooth sequence, 58, 59, 62, 63, 65
effect vector, 232, 234, 235
event, 232, 244 R
Original difference GM (1, 1), 21 Ratio
inverse, 289
P Reciprocal image, 72
Paradox of decision-making, 25, 248, 249, 251, Relative errors, 157, 163, 164, 173, 179, 211,
255 218
Parameter, 5, 8, 20, 28, 142–144, 153, 157, Relative simulation error, 164
175, 184, 186–188, 190, 191, 193, 200, Remnant GM (1,1) model, 160–165
201, 222, 239, 257, 266, 280, 281, 291, Restored value
296 through derivative, 161
332 Index

through inverse accumulation, 161, 165 term of shock disturbance, 55

Retail, 99, 101–103 Simulation error
Reverse image, 72 mean relative, 157, 160, 211
Root-mean-square deviation, 210 relative, 56, 150, 156, 157, 160, 163, 164,
Rule of maximum value, 248, 249 168, 170, 171, 173, 175, 176, 194
Simulation value, 162, 164, 193
S Slope, 32, 72, 212
Satisfactory effect, 231, 232 average slope, 72
Sequence Space
behavioral, 69, 70 euclidean, 231
behavioral horizontal, 69, 186 grey incidence factors, 69
behavioral time, 69 Spherical grey target, 232, 233
catastrophe date, 216, 217 Stochastic process, 45, 194
equal time interval, 79 State equation, 259
errors, 164 State moving probability, 194
first order, 48, 49, 62, 187, 189 State vector, 223, 258, 267, 269, 272
generated with stepwise ratio, 59, 64 Stationary point, 195
grey derivative of, 197, 295 Superfluous term, 276
homogeneous exponential, 63, 149, 150, Superior, 9
152, 153, 155 class, 230, 231
length, 58, 77, 78 countermeasure, 230
lower catastrophe, 216, 217 event, 229, 244
monotonously decreasing, 153 object, 229, 230, 248
monotonously increasing, 153 scheme, 230
non-dimensional, 69, 117, 244 Superiority analysis, 92
non-homogeneous exponential, 62 Synthetic effect measure, 24, 242, 244, 245,
quasi-smooth, 58, 59, 62, 63, 65 247
relative errors, 157, 163, 164, 173, 179, System
211, 218 asymptotically stable, 272
representative, 184 closed loop control, 257
smooth, 20, 58, 63, 64 controllable, 259
smooth ratio, 58, 59 grey linear control, 259, 260, 265
stepwise increasing, 58 observable, 259, 260
upper catastrophe, 215 open loop control, 257
vibrational, 53, 54, 149, 150, 153–155, 219 prediction models, 9, 21, 142, 296
Sequence effected by operator, 59 stable, 267–270, 273, 275, 276
Sequence operator uncontrollable, 51, 259
accumulating, 27, 46 Systems prediction, 17, 19, 21
accumulating generation operator, 27, 60,
61, 65 T
average generation operator, 27 Tendency term, 1, 60, 185, 186, 239, 278
inverse accumulating generation operator, Trap in prediction, 47
27, 61 Triangular possibility function, 123
nth order, 260, 261, 265 center-point triangular possibility function,
smooth ratio, 58 22, 24
stepwise ratio, 58, 59, 64 end-point triangular possibility function, 23
unbiased operator, 21 mixed possibility functions, 106, 110, 121,
Set of events, 227–229, 235, 239, 246 127, 131
Shock wave True behavioral sequence, 47
shock disturbed sequence, 47 Trumpet-like region, 211, 212
shock disturbed system, 19, 27, 47 Truth value, 31
Index 333

Turning point, 111, 122, 127–129 grey constraints, 241

Two stages decision model, 251 grey n-dimensional, 269, 271
uniform effect measures, of, 247
U Vibration sequence, 149, 150, 153–155
Uncertain, 1, 5–7, 10, 11, 14, 15, 30, 105, 141, amplitude, 281, 282
155, 195, 202, 227, 270
Uncertainty W
fuzzy, 10 Wave form prediction, 219
grey, 6 Weighted average weakening buffer operator,
rough, 10 52, 289
stochastic, 10 Weighted coefficient of kernel clustering for
Uniform domain, 58 decision-making, 250
Union, 4, 34, 38, 200, 205 Weighted coefficient vector of kernel clustering
Unitized for decision-making, 249–251, 255
clustering coefficient, 249 Weight of criterion, 241
clustering coefficient vector, 249 Weight vector group of kernel clustering,
Upper effect measure, 24, 242, 243, 245, 247 249–255
Upper domain, 58 Weight vector of kernel clustering, 252, 254
Useless predicted moment, 7 Whitenization of a grey number, 30
mean whitenization, 31, 33, 37
V whitenization equation, 144, 173, 174
Value domain, 58 whitenization matrix, 267
Variance, 43, 67, 193, 210 White number, 30, 33, 38, 41, 45, 199
Vector Wrapping region, 212, 213
cluster coefficient, 114, 116, 248
decision, 25, 250, 251 Z
decision coefficient, 25, 241, 249 Zigzagged line, 72, 76, 77, 84, 211, 219