Syng Yup Ohn

Sung Do Chi (Eds.)

Communications in Computer and Information Science 603

Model Design
and Simulation Analysis
15th International Conference, AsiaSim 2015
Jeju, Korea, November 4–7, 2015
Revised Selected Papers

123
Communications
in Computer and Information Science 603
Commenced Publication in 2007
Founding and Former Series Editors:
Alfredo Cuzzocrea, Dominik Ślęzak, and Xiaokang Yang

Editorial Board
Simone Diniz Junqueira Barbosa
Pontifical Catholic University of Rio de Janeiro (PUC-Rio),
Rio de Janeiro, Brazil
Phoebe Chen
La Trobe University, Melbourne, Australia
Xiaoyong Du
Renmin University of China, Beijing, China
Joaquim Filipe
Polytechnic Institute of Setúbal, Setúbal, Portugal
Orhun Kara
TÜBİTAK BİLGEM and Middle East Technical University, Ankara, Turkey
Igor Kotenko
St. Petersburg Institute for Informatics and Automation of the Russian
Academy of Sciences, St. Petersburg, Russia
Ting Liu
Harbin Institute of Technology (HIT), Harbin, China
Krishna M. Sivalingam
Indian Institute of Technology Madras, Chennai, India
Takashi Washio
Osaka University, Osaka, Japan
More information about this series at http://www.springer.com/series/7899
Syng Yup Ohn Sung Do Chi (Eds.)
•

Model Design
and Simulation Analysis
15th International Conference, AsiaSim 2015
Jeju, Korea, November 4–7, 2015
Revised Selected Papers

123
Editors
Syng Yup Ohn Sung Do Chi
Korea Aerospace University Korea Aerospace University
Goyang-si Goyang-si
Korea (Republic of) Korea (Republic of)

ISSN 1865-0929 ISSN 1865-0937 (electronic)

Communications in Computer and Information Science
ISBN 978-981-10-2157-2 ISBN 978-981-10-2158-9 (eBook)
DOI 10.1007/978-981-10-2158-9

Library of Congress Control Number: 2016946015

© Springer Science+Business Media Singapore 2016

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.
 Preface

AsiaSim (Asia Simulation Conference) is an annual conference organized by the

AsiaSim Federation member society in cooperation with the Asia Simulation Federa-
tion. Since 2001, AsiaSim has provided a forum for scientists, academics, and pro-
fessionals from around Asia to present their latest and exciting research findings in
various fields of modeling, simulation, and their applications. This strict screening
process results in the presentation of the high-quality papers in the fields of modeling
and simulation methodology, virtual reality, aerospace, e-business, manufacturing,
medical, military, networks, transportation, and traffic simulation and general engi-
neering applications.
AsiaSim 2015 received 126 submissions. After a thorough reviewing process, 71
papers were selected for oral presentations and 32 papers were selected for posters.
Among them, only 11 papers were finally accepted for CCIS as full papers (acceptance
ratio of 8.73 %). This volume contains the revised full version of 11 English papers
presented at AsiaSim 2015.
The high-quality program would not have been possible without the authors who
chose AsiaSim 2015 as a venue for their publications. We are also very grateful to the
Program Committee members and Organizing Committee members, who put a
tremendous amount of effort into soliciting and selecting research papers with a balance
of high quality, new ideas and new novel applications. We also thank the external
reviewers for their time, effort, and timely response.
We wish to express our special thanks to General Chair Prof. Yun Bae Kim
(Sungkyunkwan University, Korea) for his advice on all aspects of the conference.
Thank you also to the attendees for joining us at AsiaSim 2015.

June 2016 Seon Yep Ohn

Sung Do Chi
 Organization

Honorary Chairs
Jong-Hyun Kim Yonsei University, Korea
Axel Lehmann Universität der Bundeswehr München, Germany
Bo Hu Li Beihang University, China
Koji Koyamada Kyoto University, Japan

General Chair
Yun Bae Kim Sungkyunkwan University, Korea

General Co-chairs
Zhao Qinping Beihang University, President of the CASS, China
Satoshi Tanaka Ritsumeikan University, President of the ASIASIM and
the JSST, Japan

Organizing Committee
Seong-Yong Jang (Chair) Seoul National University of Science and Technology,
Korea
Byeong-Yun Chang (Chair) Ajou University, Korea
Jinsoo Park Yong In University, Korea
Chul-Jin Park Hanyang University, Korea

International Program Committee

Jong-Sik Lee (Chair) Inha University, Korea
Hae Young Lee Seoul Women’s University, Korea
Lin Zhang Beihang University, China
Xiaofeng Hu Beihang University, China
Yunjie Wu Beihang University, China
Ming Yang Beihang University, China
Osamu Ono Meiji University, Japan
Kyoko Hasegawa Ritsumeikan University, Japan
Gary Tan National University of Singapore, Singapore
Rubiyah Yusof Universiti Teknologi Malaysia, Malaysia
Yong Meng Teo National University of Singapore, Singapore
Yahaya Md Sam Universiti Teknologi Malaysia, Malaysia
Xiao Song Beihang University, China
VIII Organization

Publication Committee
Dong-Won Seo Kyunghee University, Korea
(Proceedings Chair)
Ja-Hee Kim Seoul National University of Science and Technology,
Korea
Sung-Do Chi Korea Aerospace University, Korea
(Special Issue Chair)
Yun-Ho Seo Korea University, Korea
(Special Issue Chair)
Jang-Se Lee Korea Maritime and Ocean University, Korea
Dong-Hyun Choi Korea Aerospace University, Korea
Sol Ha Seoul National University, Korea

Industrial Committee
Dug-Hee Moon (Chair) Changwon National University, Korea
Byung Hee Kim VMS SOLUTIONS Co., Ltd., Korea
Seung Whan Kim Siemens Industry Software Ltd, Korea
Young Suk Park ATWORTH CO., LTD., Korea
Ku-kil Chang Dassault Systems Korea, Korea
Young Gyo Chung SimTech Systems, Inc., Korea
Seong-Hoon Choi Sangmyung University, Korea

Award Committee
Hyung-Jong Kim (Chair) Seoul Women’s University, Korea

International Advisory Committee

Yun Bae Kim (Chair) Sungkyunkwan University, Korea
Gyu Min Lee Pusan National University, Korea
Sam-joon Park Agency of Defense Development, Korea
 Contents

Model and Design

A Regularized Finite Volume Numerical Method for the Extended Porous

Medium Equation Relevant to Moisture Dynamics with Evaporation
in Non-woven Fibrous Sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Hidekazu Yoshioka and Dimetre Triadis

An Integrated Network Modeling for Road Maps . . . . . . . . . . . . . . . . . . . . 17

Zhichao Song, Kai Sheng, Peng Zhang, Zhen Li, Bin Chen,
and Xiaogang Qiu

A Novel Flexible Experiment Design Method. . . . . . . . . . . . . . . . . . . . . . . 28

Gang Zhai, Yaofei Ma, Xiao Song, and Yulin Wu

Modeling Behavior of Mobile Application Using Discrete Event

System Formalism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Yun Jong Kim, Ji Yong Yang, Young Min Kim, Joongsup Lee,
and Changbeom Choi

Multi-modal Transit Station Planning Method Using Discrete Event

System Formalism. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Jaewoong Choi, Bitnal Kim, Onyu Kang, Seonwha Baek,
Yonghyun Shim, and Changbeom Choi

Reliability Analysis of Software Defined Wireless Sensor Networks . . . . . . . 65

Na Gong and Xin Huang

A Path-Line-Based Modeling for Fishing Prediction . . . . . . . . . . . . . . . . . . 79

Koji Koyamada, Katsumi Konishi, Naohisa Sakamoto,
and Marohito Takami

Simulation and Analysis

Numerical Simulation of a Hamilton-Jacobi-Bellman Equation for Optimal

Management Strategy of Released Plecoglossus Altivelis in River Systems . . . 91
Yuta Yaegashi, Hidekazu Yoshioka, Koichi Unami,
and Masayuki Fujihara

Fault Displacement Simulation Analysis of the Kamishiro Fault Earthquake

in Nagano Prefecture Using the Parallel Finite Element Method . . . . . . . . . . 102
Yuta Mitsuhashi, Gaku Hashimoto, Hiroshi Okuda, and Fujio Uchiyama
X Contents

Prediction of 137Cs-Contaminated Soil Losses by Erosion Using USLE

in the Abukuma River Basin, Japan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Carine J. Yi

Charging and Discharging Characteristics of a Quantum Well with an

Initially Lorentzian Wave Packet Incident on a DTM Type Potential . . . . . . . 118
Youichirou Kubo and Norifumi Yamada

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Model and Design
 A Regularized Finite Volume Numerical
Method for the Extended Porous Medium
Equation Relevant to Moisture Dynamics
with Evaporation in Non-woven Fibrous Sheets

Hidekazu Yoshioka1(&) and Dimetre Triadis2

1
Faculty of Life and Environmental Science,
Shimane University, Matsue, Japan
[email protected]
2
Institute of Mathematics for Industry, Kyushu University - Australia Branch,
La Trobe University, Melbourne, Australia
[email protected]

Abstract. The extended porous medium equation (PME) is a degenerate

nonlinear diffusion equation that effectively describes moisture dynamics with
evaporation in non-woven fibrous sheets. We propose a new finite volume
numerical model of the extended PME incorporating regularization of nonlinear
degenerate terms, and apply it to test cases for verification of accuracy, stability,
and versatility. One of the test cases considered is a new exact steady solution of
the extended PME. We also examine a differential equation-based adaptive
re-meshing technique for resolving sharp transitions of solution profiles that may
be optionally incorporated into the procedure above. The computational results
demonstrate satisfactory accuracy of the proposed numerical model, with rea-
sonable reproduction of complicated moisture dynamics involving sharp tran-
sitions and divorce of supports.

Keywords: Moisture dynamics
Evaporation
Non-woven fibrous sheet

Extended porous medium equation
Dual-finite volume method

1 Introduction

Comprehending moisture dynamics occurring in porous media, such as soils, concretes,

papers, and fibrous sheets is essential for a wide variety of real systems, and is thus
currently an important research topic in hydro-environmental and water resources
engineering [1]. Recently, non-woven fibrous sheets have gained attention as cheap and
easily obtained materials with a variety of industrial applications, such as agricultural
plantation sheets and sanitary products. Assessing moisture dynamics in non-woven
fibrous sheets is a crucial step towards creating fit-for-purpose industrial products. As a
very recent example in Japan, non-woven fibrous sheets have been used to vertically
transport water in a small-scale rooftop vegetation system. This is thought to be a
cheap, space-saving, and environmentally friendly plantation system whose source
water solely consists of collected rainwater stored in a plastic tank [1]. The resulting

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 3–16, 2016.
DOI: 10.1007/978-981-10-2158-9_1
4 H. Yoshioka and D. Triadis

moisture dynamics are very complex, and appropriate mathematical and/or numerical
models are needed for a practical comprehension of the system.
Moisture dynamics in porous media can be effectively described with Porous
Medium Equations (PMEs), an important class of degenerate nonlinear diffusion
equations based on the mass conservation principle and constitutive laws [2]. Similar
differential equations arise in other applied study areas, such as plasma physics [3] and
astrophysics [4]. Solving PMEs involves mathematical and computational difficulties
due to degeneracy and nonlinearity of the coefficients. Practical numerical methods for
solving PMEs should have sufficiently high accuracy, stability, and versatility. How-
ever, such methods are still not common and developing a simple numerical method
that is sufficient for practical use is an important research topic.
The main purpose of this article is to present a new numerical model of an extended
PME governing longitudinally one-dimensional (1D) moisture dynamics occurring in
non-woven fibrous sheets based on regularization, finite volume, and a differential
equation-based adaptive re-meshing technique. The proposed numerical model is
extensively verified through test cases with exactly or partially known solutions.

2 Mathematical Model

2.1 Extended Porous Medium Equation (PME)

This article considers moisture dynamics in homogenous thin non-woven fibrous sheets
with a length scale of 10−1 m, width scale of 10−2 m, and thickness scale of 10−3 m to
10−4 m. It is reasonable to assume that transverse variations of the moisture profiles in
such fibrous sheets are insignificant compared to longitudinal variations, thus we
consider a 1D longitudinal mathematical model. The volumetric water content at each
position x of a sheet at the time t is denoted by h ¼ hðt; xÞ, and is normalized to
0  u ¼ ðh  hr Þðhs  hr Þ1  1 where hs and hr ð¼ 0Þ are the maximum (saturated)
and minimum water contents of the sheet. A remarkable difference between modelling
moisture dynamics in non-woven fibrous sheets compared with other materials in
applications, such as soils and concretes, is that evaporation occurs over the entire 1D
or 2D domain in the former case [1] but usually only through the boundary of the
domain in the latter case. Considering this characteristic of moisture dynamics in the
sheets and adopting dimensionless variables as in the literature [5], the extended PME
is proposed in this article as [1, 2, 6]


@u @ mp1 @u @  m1  ~ q
¼ ðm  pÞu þ sin a u u  Es u ð1Þ
@~t @~x @~x @~x

with dimensionless variables and parameters

~t ¼ ðm  pÞKs2 D1 2
x ¼ ðm  pÞKs D1
s hs t; ~
1 ~ 1 2
s hs x; and Es ¼ ðm  pÞ Ks Ds hs Es
2

ð2Þ
 A Regularized Finite Volume Numerical Method 5

where Ds [ 0 is the saturated diffusivity, Ks [ 0 is the saturated permeability, Es  0 is

the evaporation coefficient that depends on both material properties of the sheet and its
surrounding environment, m, p  m  1, and q are positive nonlinearity parameters,
and 0:5p  a  0:5p is the inclination angle of the sheet as defined in Fig. 1.

Fig. 1. The 1D domain of the extended PME (1) indicating the inclination a of the fibrous sheet.

In deriving (1), the pressure head wðuÞ and the permeability K ðuÞ have been
assumed to be physically parameterized as

wðuÞ ¼ p1 Ks1 Ds hs ð1  up Þ and K ðuÞ ¼ Ks um : ð3Þ

Note that the pressure head reduces to the conventional case as p ! þ 0 [2, 6, 7].
The pressure head wðuÞ is continuous and increasing in 0\u  1, vanishes at u ¼ 1,
and diverges as u ! þ 0. The permeability K ðuÞ is continuous and increasing in
0  u  1 and vanishes at u ¼ 0. Figure 2(a) and (b) plot the pressure head wðuÞ and
the permeability K ðuÞ for several p and m, respectively where the coefficients Ks1 Ds hs
for wðuÞ and Ks for K ðuÞ are set to be 1 for the sake of simplicity. The angle a in (1)
may be spatially distributed, but is assumed to be constant in this article. There exist
two contrasting cases on arrangement of the sheets, which are the horizontal (a ¼ 0)
and vertical cases (a ¼ 0:5p). The advection term of (1) vanishes in the horizontal
case (sin a ¼ 0). The standard theory for moisture evaporation from the surface of a soil
body involves a near constant evaporation rate for most values of u. It is only as u
approaches zero that the evaporation rate rapidly decreases to zero [8]. This behavior
has been observed in our preliminary laboratory experiments using thin non-woven
fibrous sheets, which are not discussed here but will be addressed in a succeeding
paper. In the extended PME, this phenomenon occurs with q  1, which has actually
been qualitatively validated through preliminary laboratorial experiments. In addition,
the order of the parameters m and p have experimentally been estimated as O ¼ ð100 Þ
and O ¼ ð101 Þ, respectively, meaning that the relation p  m  1 is satisfied for
moisture dynamics in non-woven fibrous sheets. Hereafter, “~
” representing
non-dimensional variables are omitted for the sake of brevity.

2.2 Regularized Equation

Degeneration of the advection and diffusion terms of the extended PME (1) would
cause severe mathematical and computational issues due to regularity deficits of the
coefficients. The nonlinear degeneration of (1) can be mitigated by regularizing it as
6 H. Yoshioka and D. Triadis

Fig. 2. Plots of (a) the pressure wðuÞ and (b) permeability K ðuÞ. (Color figure online)


@u @ mp1 @u @  
¼ ð m  pÞ ½ he ð uÞ  þ sin a ½he ðuÞm1 u  Es uq ð4Þ
@t @x @x @x
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
using the regularization kernel he ðuÞ ¼ u2 þ e2  e with a small positive parameter
eð¼ 1010 Þ, which mathematically as well as numerically rules out the degeneration of
the advection and diffusion terms. Application of another conventional regularization
method to (1) remains as an option; however, such methods typically rely on trans-
formations of the solution u to a new dependent variable, which in general cannot
numerically realize mass conservation [7].

3 Finite Volume Numerical Model

A numerical method for solving the regularized PME (4), which is referred to as the
Dual-Finite Volume Method (DFVM) [10], is briefly explained in this section.
The DFVM was originally developed for numerically solving Kolmogorov’s forward
equations, which are linear and conservative advection-diffusion-decay equations
(ADDEs), defined on connected graphs [11]. The DFVM concurrently uses primal and
dual computational meshes and evaluates the numerical fluxes based on analytical
 A Regularized Finite Volume Numerical Method 7

solutions to local two-point boundary value problems. Mathematical analysis revealed

that the DFVM for linear ADDEs is unconditionally stable in both space and time with
an appropriate implicit temporal integration algorithm, such as the conventional
backward Euler method.
In the DFVM, (4) is regarded as a nonlinear ADDE. An operator-splitting algorithm
analogous to that of Li et al. [12] specialized for solving (4) is incorporated into the
DFVM, so that numerical instability arising from the nonlinear evaporation term is
avoided. The time increment between each successive time steps is denoted by Dt, the
differential operators defining the advection and diffusion terms of (4) by PAD , and that
for the evaporation term by PE ; namely, the operators PAD and PE for generic suffi-
ciently regular u ¼ uðt; xÞ are expressed as


@ @u @  
PAD u ¼ ðm  pÞ½he ðuÞmp1 þ sin a ½he ðuÞm1 u ð5Þ
@x @x @x

and

PE u ¼ Es uq ; ð6Þ

respectively. Symbolically, the present numerical method temporally discretizes (4) at

each time step as

uðk þ 1Þ ¼ expð0:5DtPE Þ expðDtPAD Þ expð0:5DtPE ÞuðkÞ ð7Þ

where uðkÞ is the solution at the k th time step. Temporal integration in the evaporation
sub-steps is performed with local analytical solutions to the temporal ordinary differ-
ential equation at each node [12]. Temporal integration in the advection-diffusion
sub-step is performed with the conventional backward Euler method and the DFVM
spatial discretization, which is coupled with a Picard algorithm for handling the non-
linear advection and diffusion terms. Although the DFVM is theoretically uncondi-
tionally stable for linear problems, preliminary computation demonstrated that it is not
the case for the extended PME with advection term when Dt is sufficiently large. The
reason for this phenomenon would be the loss of ellipticity of the discretized system
with large Dt as implied in Pop et al. [9].
The present DFVM can be optionally equipped with an adaptive re-meshing
technique based on the Moving Mesh Partial Differential Equation (MMPDE) method
for accurately resolving sharp transitions of solution profiles [13]. The MMPDE
method solves an additional parabolic PDE that governs nodal positions; this is distinct
from the primary PDE to be solved, which in this article is the extended PME. An
advantageous point of the MMPDE method is that it does not alter the topology of the
mesh, which is not common to the other types of adaptive re-meshing methods.
The MMPDE method has successfully been used for numerically approximating
solutions to the Hamilton-Jacobi-Bellman equations governing upstream fish migration
in 1D river flows [14] and Kolmogorov’s forward equations in unbounded domains
governing probability density functions of stochastically-excited systems [15].
8 H. Yoshioka and D. Triadis

4 Application to Test Cases

The present DFVM is extensively applied to test cases for its verification of accuracy,
stability, and versatility. Comparing performance with other published numerical
models is beyond the scope of this article, and will be discussed elsewhere. The test
cases considered in this article are (Sect. 4.1) Barenblatt problems, (Sect. 4.2) hori-
zontal infiltration subject to evaporation, (Sect. 4.3) inclined infiltration subject to
evaporation, (Sect. 4.4) support divorce problem in a horizontal sheet, and (Sect. 4.5)
downward infiltration in a vertical sheet.

4.1 Barenblatt Problems

The Barenblatt solutions are analytical solutions to the conventional PME that do not
have advection and evaporation terms [16]. A Barenblatt solution has a symmetrical
shape in space. The solution has bounded support and it is not differentiable in the
classical sense at the fronts of the support, hence it is a weak solution. The Barenblatt
solutions and their variants have been used as benchmark test cases for numerical
methods for PMEs [16, 17]. Mathematical analysis of Barenblatt solutions and related
exact solutions to the PMEs are reviewed in [18, 19]. A difficulty of approximating the
Barenblatt solutions lies in the existence of the sharp fronts where numerical solutions
would suffer from spurious oscillations as pointed out in Zhang and Wu [17]. For this
case we set parameters p ¼ 0, a ¼ 0, and Es ¼ 0. The initial time is set as t ¼ 0:01 for
avoiding the Deltaic singularity at t ¼ 0:00. The sheet is identified with the domain
X ¼ ð7; 7Þ. Homogenous Dirichlet boundary conditions are specified at the bound-
aries x ¼ 7 for the sake of simplicity. The time increment is set as Dt ¼ 0:01. The
numerical and analytical solutions are compared at the time t ¼ 1:00 for
m ¼ 2; 3; . . .; 8.
Figure 3(a) compares the numerical and analytical solutions for each m without the
MMPDE. The fronts of the analytical solutions become sharper as m increases. The
numerical solutions do not successfully approximate the position and shape of the front
for m ¼ 6; 7; 8, whereas they are accurately captured for smaller m. Figure 3(b) pro-
vides an enlarged view of numerical and analytical solutions around the right front for
m ¼ 8, demonstrating that the numerical solution with the MMPDE can more accu-
rately capture sharp solution profile than that without it for challenging Barenblatt
problems.

4.2 Horizontal Infiltration Subject to Evaporation

This second test considers water infiltration from one side (x ¼ 0) of a horizontal sheet
(a ¼ 0 and p ¼ 0) in an evaporative environment as theoretically considered in
Lockington et al. [6]. Assuming that u ¼ 1 at the boundary x ¼ 0 and that the length of
the sheet is sufficiently long, a straightforward calculation shows that the steady
solution to this test case is found as
 A Regularized Finite Volume Numerical Method 9

Fig. 3. Comparisons of Barenblatt numerical and analytical solutions (a) over the entire domain
and (b) around the right front. (Color figure online)

8 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !mq
2 9
< Es =
uð xÞ ¼ max 1  ðm  qÞ x ;0 : ð8Þ
: 2mðm þ qÞ ;

The solution has bounded support for m [ q. As can be directly seen from (8), this
solution is also a weak solution since it is continuous but not differentiable at an edge of
the support. The focus here is whether the DFVM can capture the solution profiles and
the front position. The sheet is identified with the domain X ¼ ð0; 1Þ, which is uni-
formly discretized into 100 cells. The time increment is set as Dt ¼ 0:001. Steady
numerical solutions are computed starting from the initial guess u ¼ 0 over X. The
parameters Es , m, and q can be chosen such that the support of (8) becomes ½0; 0:5: We
have fixed the parameter q at 0:01, and shown solutions for m ¼ 2; 4; and 6, so that
Es ¼ 8:12; 8:06; and 8:04, respectively. The terminal time of computation is empir-
ically set as t ¼ 20 at which the numerical solutions are observed to be sufficiently
close to steady state.
Figure 4(a) compares numerical and analytical solutions for different values of m
without the MMPDE. The numerical solutions successfully approximate analytical
solutions for all m examined. Figure 4(b) provides an enlarged view of numerical and
10 H. Yoshioka and D. Triadis

Fig. 4. Horizontal infiltration subject to evaporation (a) without and (b) with the MMPDE.
(Color figure online)

analytical solutions around the front for m ¼ 4. Numerical solutions without the
MMPDE appear to perform better than those with it, indicating that using the MMPDE
with the DFVM may not be a good option for the problems with highly singular
evaporation term with small q. Qualitatively similar computational results have been
obtained for the other values of m ¼ Oð100 Þ.

4.3 Inclined Infiltration Subject to Evaporation

An inclined infiltration problem of water from the downstream-end of a sheet is
considered for examining accuracy and stability of the DFVM against more compli-
cated steady problems. As in the previous problem, this test considers water infiltration
from one side (x ¼ 0) of the sheet. Assuming the relationship p ¼ q, we can derive the
implicit analytical solution
Z
m sin a 1
vmq1
x¼ h  i dv ð9Þ
Es u 1 þ W1  exp 1  mv
m sin2 a
ðmqÞEs
 A Regularized Finite Volume Numerical Method 11

where W1 ½ y denotes the second branch of Lambert W-function, which is valid for the
range expð1Þ  y\1. To the authors’ knowledge, this implicit analytical solution has
not been derived elsewhere. The solution (9) can be verified by differentiating twice
with respect to x, using the well known result

dW1 ½ y W1 ½ y
¼ : ð10Þ
dy yð1 þ W1 ½ yÞ

The solution (9) has the bounded support ½0; z where z is the value of the right
hand-side of (9) with u ¼ 0. The domain X is specified as ð0; 2Þ. Parameter values are
specified as a ¼ 0:5p and p ¼ q ¼ 0:01, with four cases of ðEs ; mÞ examined;
ð3:01; 1Þ, ð3:01; 4Þ, ð5:01; 1Þ, and ð5:01; 4Þ, to see the effect that varying nonlinearity
of the advection-diffusion terms, and varying strength of the evaporation term have on
solution profiles. The domain is uniformly discretized into 200 cells and the time
increment is set as Dt ¼ 0:001. The terminal time of computation is empirically set as
t ¼ 20 at which the numerical solutions are observed to be sufficiently close to steady
state. The MMPDE method is not used since the previous problem showed that it does
not to successfully work for the problem with nonlinear evaporation.
Figure 5(a) and (b) compare the numerical and the analytical solution (9) for
stronger (Es ¼ 1) and weaker evaporative environments (Es ¼ 4). The agreement
between two is satisfactory in all cases. The computational results indicate that the
DFVM can handle steady infiltration problems in inclined sheets subject to nonlinear
evaporation. From the form of the exact solution it follows that the support of each
solution is not significantly dependent on the parameter m ¼ Oð100 Þ if 0\q  1, this
is demonstrated in Fig. 5.

4.4 Support Divorce Problem in a Horizontal Sheet

The fourth test case is evaporative moisture dynamics in an initially partially wetted
horizontal sheet. The values p ¼ 0, a ¼ 0, and Es ¼ 1 are specified. The sheet is
identified with the domain X ¼ ð1:5p; 1:5pÞ and the initial condition of u is same as
that in Zhang and Wu [17], a continuous function having bounded support with two
local extremes and one local minimum. The homogenous Dirichlet condition is
assumed at both boundaries of X. According to the literature [20, 21], if m þ q ¼ 2
with m [ 1 and q [ 0, the support of the solution separates into two smaller disjoint
regions before eventually vanishing. This phenomenon is called “divorce of supports”.
The remaining model parameters are set as m ¼ 1:98 and q ¼ 0:02 following Zhang
and Wu [17]. Tomoeda et al. [22, 23] mathematically and numerically analyzed similar
problems in detail, indicating that the problem focused on in this section does not have
classical solutions. The focus here is whether the DFVM can handle this peculiar
phenomenon without computational failure. The domain X is uniformly divided into
320 cells. The time increment is set as Dt ¼ 0:001.
Figure 6(a) shows space-time evolution of the numerical solution where its support
is identified with the contour line of u ¼ 1010 , which is plotted as a black curve.
Figure 6(b) plots the spatial solution profile at the time t ¼ 0:788 examined in Zhang
12 H. Yoshioka and D. Triadis

Fig. 5. Comparisons of numerical and analytical solutions for inclined infiltration subject to
evaporation for Es ¼ 1 (a) and (b) Es ¼ 4.

and Wu [17]. The obtained results are comparable to those in Zhang and Wu [17] who
used a local DG method, which we consider to be a more complicated numerical
algorithm than the DFVM because a given computational mesh will have a higher
degree of freedom. Hence the computation results show that the DFVM can handle
transient problems having an evaporative source term without numerical instability.

4.5 Downward Continuous Infiltration in a Vertical Sheet

The final test case considers a downward water infiltration process from the top of a
vertical sheet (a ¼ 0:5p) subject to an impulsive water input whose analytical
solution is available in Hayek [24]. These analytical solutions have been derived in
order to comprehend moisture dynamics of wetting front in vertical soil columns with a
variety of physical properties. They are therefore not originally considered for moisture
dynamics in fibrous sheets, but are used as one of the test cases here because the
Richards equation [24] and the extended PME in this paper mathematically have a
similar form. The parameters are specified as p ¼ 1, the positive integer m  p þ 1, and
 A Regularized Finite Volume Numerical Method 13

Fig. 6. Numerical solution of the support divorce problem (a) in the x  t phase space and (b) at
the time t ¼ 0:788. (Color figure online)

Es ¼ 0. The focus here is assessing whether the DFVM can capture infiltration fronts
where diffusion and advection terms would degenerate, potentially serving as severe
computational difficulties. The sheet is identified with the domain X ¼ ð0; 5Þ with the
boundary condition F ¼ 0 at x ¼ 0 and 5 where F is the flux associated with (4).
A delta-function initial condition is adopted: uð0; xÞ ¼ dð xÞ. The deltaic condition is
approximated as h1 where h is the size of the leftmost cell. The domain X is uniformly
divided into N regular cells and the time increment is set as N 1 . The values of N
examined are 100
2k for k ¼ 0; 1; 2; 3; 4. The MMPDE method is not used here
because preliminary computation indicated instability of numerical solutions just after
the time t ¼ 0, due to the deltaic initial condition.
Figure 7(a) and (b) compare numerical (N ¼ 200) and analytical solutions with
m ¼ 2 and m ¼ 3 at selected time steps at which the boundary effects at x ¼ 5 are
considered to be sufficiently small. The analytical solutions with m ¼ 2 and m ¼ 3
have unbounded and bounded supports, respectively, and present profiles with different
curvatures. They therefore have qualitatively different mathematical properties.
14 H. Yoshioka and D. Triadis

Fig. 7. Comparisons of numerical and analytical solutions to the downward infiltration problem
with (a) m ¼ 2 and (b) m ¼ 3. (Color figure online)

The figures show that the numerical solutions compare well with the analytical solu-
tions where the maximum values and the tails of the solutions are quite accurately
reproduced at each time step. The computational results with other values of N, not
presented in this article, implied almost first-order convergence of the numerical
solutions in both cases. This observation is consistent with the fact that the DFVM is
equivalent to a fully-upwind FVM for advection-dominant ADDEs [11].

5 Conclusions

A numerical model of the extended PME with a regularization method for the nonlinear
coefficients was proposed and applied to test cases whose solution behaviors are
known. The obtained computational results demonstrated its satisfactory accuracy,
stability, and versatility, indicating that it can potentially serve as an effective tool for
simulating complex moisture dynamics in non-woven fibrous sheets. There were
indications that the MMPDE method may not be a good option for problems with
highly singular evaporation term with small q. A new exact solution for extended
PMEs was derived in this paper, and was accurately reproduced with the DFVM.
 A Regularized Finite Volume Numerical Method 15

The solution may be useful for verification of numerical methods and also for
parameter identification of non-woven fibrous sheets and porous media. Spatially 2-D
extension of the present mathematical and numerical models does not encounter
additional technical difficulties. Future research will address mathematical analysis of
the extended PME, on its regularity and analytical solutions in particular following
[25]. In addition, assessing consistency error between the original and regularized
extended PMEs will be addressed; we expect that the error can be a decreasing function
of the parameter e under an appropriate normed space as in Atkinson and Han [26].
Detailed numerical comparisons of a simplified extended PME with experimental
observations have been performed in Yoshioka et al. [1]. Their approach can be further
sophisticated with the help of the present numerical model for more realistic and
reliable assessment of moisture dynamics in non-woven fibrous sheets. Finally, the
presented regularization method and the DFVM can be applied to other degenerate
nonlinear diffusion equations arising in applied problems, such as Keller-Segel systems
for chemotaxis [27] and population dynamics models [28].

Acknowledgements. We thank Dr. Ichiro Kita and Dr. Kotaro Fukada in Faculty of Life and
Environmental Science, Shimane University, Japan for providing helpful comments and sug-
gestions on this article.

References
1. Yoshioka, H., Ito, Y., Kita, I., Fukada, K.: A stable finite volume method for extended
porous medium equations and its application to identifying physical properties of a thin
non-woven fibrous sheet. In: Proceedings of JSST2015, pp. 398–401 (2015)
2. Landeryou, M., Eames, I., Cottenden, A.: Infiltration into inclined fibrous sheets. J. Fluid
Mech. 529, 173–193 (2005)
3. Wilhelmsson, H.: Simultaneous diffusion and reaction processes in plasma dynamics. Phys.
Rev. A 38, 1482–1489 (1988)
4. Haerns, J., Van Gorder, R.A.: Classical implicit travelling wave solutions for a quasilinear
convection-diffusion equation. New Astron. 17, 705–710 (2012)
5. Broadbridge, P., White, I.: Constant rate rainfall infiltration: a versatile nonlinear model.
Analytic solution. Water Resour. Res. 24, 145–154 (1988)
6. Lockington, D.A., Parlange, J.Y., Lenkopane, M.: Capillary absorption in porous sheets and
surfaces subject to evaporation. Transport Porous Med. 68, 29–36 (2007)
7. Brooks, R.H., Corey, A.T.: Hydraulic properties of porous media. Colorado State
University, Hydrology Papers, Fort Collins, Colorado (1964)
8. Stewart, J.M., Broadbridge, P.: Calculation of humidity during evaporation from soil. Adv.
Water Resour. 22, 495–505 (1999)
9. Pop, I.S., Radu, F., Knabner, P.: Mixed finite elements for the Richards’ equation:
linearization procedure. J. Comput. Appl. Math. 168, 365–373 (2004)
10. Yoshioka, H.: On dual-finite volume methods for extended porous medium equations, arXiv
preprint. arXiv:1507.05281 (2015)
11. Yoshioka, H., Unami, K.: A cell-vertex finite volume scheme for solute transport equations
in open channel networks. Prob. Eng. Mech. 31, 30–38 (2013)
16 H. Yoshioka and D. Triadis

12. Li, Y., Lee, G., Jeong, D., Kim, J.: An unconditionally stable hybrid numerical method for
solving the Allen-Cahn equation. Comput. Math Appl. 60, 1591–1606 (2010)
13. Huang, W., Russel, R.D.: Adaptive Moving Mesh Methods, pp. 27–133. Springer,
Heidelberg (2011)
14. Yoshioka, H., Unami, K., Fujihara, M.: A Petrov-Galerkin finite element scheme for 1-D
tome-independent Hamilton-Jacobi-Bellman equations. J. JSCE. Ser. A2, 71 (in press)
15. Yaegashi, Y., Yoshioka, H., Unami, K., Fujihara, M.: An adaptive finite volume scheme for
Kolmogorov’s forward equations in 1-D unbounded domains. J. JSCE. Ser. A2, 71 (in press)
16. Li, H., Farthing, M.W., Dawson, C.N., Miller, C.T.: Local discontinuous Galerkin
approximations to Richards’ equation. Adv. Water Resour. 30, 555–575 (2007)
17. Zhang, Q., Wu, Z.L.: Numerical simulation for porous medium equation by local
discontinuous Galerkin finite element method. J. Sci. Comput. 38, 127–148 (2009)
18. Aronson, D.G., Caffarelli, L.A.: The initial trace of a solution of the porous medium
equation. Trans. Am. Math. Soc. 280, 351–366 (1983)
19. Fila, M., Vázquez, J.L., Winkler, M., Yanagida, E.: Rate of convergence to Barenblatt
profiles for the fast diffusion equation. Arch. Ration. Mech. An. 204, 599–625 (2012)
20. Rosenau, P., Kamin, S.: Thermal waves in an absorbing and convecting medium. Physica D
8, 273–283 (1983)
21. Nakaki, T., Tomoeda, K.: A finite difference scheme for some nonlinear diffusion equations
in an absorbing medium: support splitting phenomena. SIAM J. Numer. Anal. 40, 945–954
(2002)
22. Tomoeda, K.: Numerically repeated support splitting and merging phenomena in a porous
media equation with strong absorption. J. Math-for-Ind. 3, 61–68 (2011)
23. Tomoeda, K.: Numerical and mathematical approach to support splitting and merging
phenomena in the behaviour of non-stationary seepage. Theor. Appl. Mech. Jpn. 63, 15–23
(2015)
24. Hayek, M.: Water pulse migration through semi-infinite vertical unsaturated porous column
with special relative-permeability functions: exact solutions. J. Hydrol. 517, 668–676 (2014)
25. Vazquez, J.L.: The Porous Medium Equation. Oxford University Press, Oxford (2007)
26. Atkinson, K., Han, W.: Theoretical Numerical Analysis, pp. 449–451. Springer, Heidelberg
(2009)
27. Bian, S., Liu, J.G.: Dynamic and steady states for multi-dimensional Keller-Segel model
with diffusion exponent m > 0. Commun. Math. Phys. 323, 1017–1070 (2013)
28. Naldi, G., Cavalli, F., Perugia, I.: Discontinuous Galerkin approximation of porous
Fisher-Kolmogorov equations. Commun. Appl. Indust. Math. 4 (2013). doi:10.1685/journal.
caim.446
 An Integrated Network Modeling for Road Maps

Zhichao Song1 ✉ , Kai Sheng2, Peng Zhang1, Zhen Li1, Bin Chen1, and Xiaogang Qiu1
( )

1
College of Information System and Management, National University of Defense Technology,
Changsha 410073, Hunan, People’s Republic of China
[email protected]
2
Electronic Engineering College, Naval University of Engineering, Wuhan 430033, Hubei,
People’s Republic of China

Abstract. Critical-location identification on a road map is very helpful to assign

traffic resources reasonably in traffic simulations. To simultaneously identify the
critical levels of roads and junctions in a road map by the same measure of
centrality, we define a novel road network modeling concept: integrated graph,
in which both junctions and roads are abstracted as nodes. Based on this method,
we analyze the importance of locations in a small road network and Beijing’s
main-road network. The results show that this modeling method of road networks
is feasible and efficient.

Keywords: Network modeling · Road network · Centrality indices · Critical-

location identification

1 Introduction

Road network is an important part of environment models in both social simulation and
military simulation. Identifying the critical junctions or roads on the whole road map
and pay more attention to them is very helpful to assign traffic resources reasonably in
traffic simulations. As the traffic resources are always limited, more reasonable the
assignment is means that more important parts of the road network can be protected
better, so these important parts can be restored very soon once they are invalidated and
traffic on the road network can be more stable and more smooth.
Maybe it is because the roads linking with each other can be regarded as a graph or
network naturally. The issue that identifying key parts on road maps is usually dealt with
by graph theory or complex network theory. Urška et al. [1] researched how to identify
critical locations in a spatial network and extracted the key nodes in the street network
of Helsinki Metropolitan Area. Porta et al. [2] studied how centrality works in cities
based on the road graphs. Wu et al. [3] analyzed the topological bottleneck for traffic
networks.
Currently, there are three broadly used road network modeling methods: “Name
Street” method [4], primal graph approach [5] and dual graph approach [6]. In the street
network modeled by “Name Street” method, nodes represent named streets and edges
describe the junctions among these streets. The named-street-centred network reflects a
higher level of abstraction topological connections in a given street network and is very

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 17–27, 2016.
DOI: 10.1007/978-981-10-2158-9_2
18 Z. Song et al.

suitable to analyze the topological character of the street network. Based on the network,
Jiang and Claramunt [4] gave their conclusions that the topology of street networks
reveals a small-world property but has no scale-free property. Whereas, this network
model does not fit the high-resolution critical-location identification in a road map. On
the one hand, the critical-location identification needs to consider geometrical informa‐
tion but lengths of the streets are not described in the model. On the other hand, a named
street is usually composed of many road segments and functions of these segments in a
whole road network should be different, so the critical location analysis in the network
can only give a large-scale location. In the dual graph of a road map, nodes represent
roads and edges represent junctions. This graph can provide a road-segment-level crit‐
ical-location analysis. However the length of each road segment still cannot be consid‐
ered. The critical-location analysis based on a dual graph is just a topological step-
distance concept. The primal graph of a road map is more comprehensive, objective and
realistic for the network analysis of centrality. In the graph, nodes are junctions and
edges are roads. This graph is usually a weighted graph because lengths of roads are
assigned to the edges. Therefore, the primal graph of a road map is the most suitable of
the three for the critical-location analysis on a road map.
Centrality indices, which are usually used to measure the importance of the node
in a network, are also used to identify the critical levels of the locations on a road
map. However, even based on the primal graph method, it still does not satisfy
enough to simultaneously calculate the critical levels of roads and junctions of a road
network by a measure of centrality once. Mostly, centrality scores of nodes are
calculated firstly and then the critical levels of edges are estimated by the scores of
nodes they connect directly. For example each edge is assigned a centrality score
equal to the average score of the pair of nodes on the both sides of it in literature [5]
and several kinds of “edge degree” are defined in literature [7]. In this method, crit‐
ical levels of edges which are estimated by nodes’ scores are totally depended on the
nodes on their both sides. It is in such a local scale that actual effect on the whole
network an edge has is not considered accurately.
To settle this simultaneous calculation for the centrality of nodes and edges, we
propose a new road network modeling method named “Integrated Graph Method” in
this paper. In the integrated graph of a road map, junctions and roads are all abstracted
as nodes and the connective relations are regarded as edges. The remainder of this paper
is organized as follows. In Sect. 2, the integrated graph method is introduced. In Sect. 3,
we present the key parts identification for a small road map. In Sect. 4, a real large-scale
case is given. Our conclusions are given in Sect. 5.

2 Method

We consider that no matter it is a road or a junction the critical locations on the road
map should be identified by one measure of centrality once. As we have introduced, the
road network modeling methods such as primal graph method, dual graph method or
“Named Street” method cannot meet our demands. Among the three methods, primal
graph is much more suitable for discovering critical elements of the network. However,
 An Integrated Network Modeling for Road Maps 19

in primal graph network the importance of nodes and edges still cannot be simultane‐
ously calculated by one measure of centrality. Always the edges’ importance is estimated
by the centrality sores of nodes connecting the edge directly. In this paper, we integrate
the ideas of node presentation in primal graph and dual graph that both roads and junc‐
tions on a road map are represented by nodes. That is why the network modeling method
we proposed is named “Integrated Graph Method”.
Specifically, both roads and junctions on a road map are represented by nodes when
the road network is constructed. If a road is directly connected with a junction, there is
an edge between them and the weight of every edge is half of the road’s length. Based
on this weighted graph or network, the importance of roads and junctions can be
computed simultaneously by one measure of centrality. Figure 1 gives the different

(a) (b)

(c) (d)

Fig. 1. The different network models of a simple road map: (a) is the original road map; (b) is
the primal graph of the road map, in which nodes are junctions and lines are roads; (c) is the dual
graph of the road map, in which nodes mean roads and lines mean junctions; (d) is the network
modeled by integrated graph method, in which red nodes represent junctions and green nodes
represent roads (Color figure online)
20 Z. Song et al.

network models of a simple road map based on primal graph method, dual graph method
and the integrated graph method.

3 Key Parts Identification for a Small Road Map

Being the importance measure methods, centrality indices are often used to identify key
parts on road maps. So far, there have been various measures of structural centrality,
which can be divided into three classes. The first-class measures are based on the idea
that the a central node in a network is near to the other nodes. The second-class measures
are based on the idea that central nodes stand between other node pairs on their paths of
connection. The third-class measure are the ones which combine the two main ideas of
centrality mentioned above. The four measures of centrality: degree centrality [8],
closeness centrality [9], betweenness centrality [10], and information centrality [11] are
four classic ones among the various measures of centrality. The four measures of
centrality cover the three classes mentioned above. Degree centrality is a local-scale
first-class measure, closeness centrality is the global-scale first-class measure, betwe‐
enness centrality is a representative one of the second-class measure, and information
centrality belongs to the third-class measure.
Here we apply the four classic measures of centrality namely degree centrality,
closeness centrality, betweenness centrality, and information centrality to calculate the
importance of nodes and edges in a simple hypothetic road network. To show the differ‐
ences among the four measures, we used a hypothetic road map. Figure 2 gives the
primal graph and the integrated graph of it. From the primal graph we can see that the
simple road map consists of 19 junctions that are labeled by “v1, v2, v3 …, v19” and
18 roads which are labeled by “e1, e2, e3 …, e18”.
The importance of roads and junctions on this road map is calculated by four meas‐
ures of centrality (degree centrality, closeness centrality, betweenness centrality, and
information centrality) based on the primal graph and the integrated graph. Since the
importance of an edge in the primal graph cannot be directly calculated by the node
centrality indices, it is quantified by the average centrality scores of the pair of nodes
on the both sides of it. As the centrality expressions we use are the normalized ones, the
results are limited between 0 and 1.
Figure 3 shows the centrality scores of junctions and roads calculated based on the
primal graph and the integrated graph of the road map. The x-axis is marked by the
sequence numbers of junctions and roads, in which 1–19 represent the nodes “v1, v2,
v3 …, v19” and 20–37 represent the edges “e1, e2, e3 …, e18”. The y-axis is marked
by the centrality scores. The scores are calculated by the four different measures of
centrality. In the integrated graph of road maps, importance scores of the roads are
evaluated by their impacts on the whole network and not only depending on the pair of
junctions on their both ends, so the scores of road nodes are much more reasonable than
the averages.
The centrality scores calculated in the integrated graph are universally lower than
those in the primal graph except the closeness. This is because the primal graph does
not consider the roads as nodes and the normalization factor is related to the node
 An Integrated Network Modeling for Road Maps 21

(a)

(b)

Fig. 2. A simple hypothetic road network: (a) is the primal graph and (b) is the integrated graph

number. If the normalization factors applied to the two network graphs are the same,
the degree centrality sores of junctions in the two networks will be the same. Whereas
the importance of roads are not exact when degree centrality is used in the integrated
graph. We should not ascribe this to the road network model, and it is because the
shortage of degree centrality itself that degree centrality can only give a so local scale
evaluated result. For example, the importance of “v4” and “v5”, of which scores are
very high when closeness centrality, betweenness centrality and information centrality
are used, is not very obvious in Fig. 3(a). For the other three measures of centrality,
effects come from the normalization factor are not obvious. It is counteracted by the
decrease of node pairs’ distances for closeness centrality and the increase of the shortest
path for betweenness centrality. Moreover, the increase of the shortest path makes an
extra contribution to heighten the closeness scores. The normalization factor of infor‐
mation centrality is actually not affected by the node number. The decrease of informa‐
tion centrality scores in integrated graph is only due to the increase of network efficiency
caused by the new road nodes. The trend of the four centrality scores in the two graphs
is similar and the consistency of the closeness centrality scores is the best. The relative
importance of junctions does not change very much in the integrated graph. Especially
when betweenness centrality is used, importance of junctions in these two networks is
22 Z. Song et al.

0.5 0.5
Primal Graph Primal Graph
0.4 Integrated Graph 0.45 Integrated Graph

Centrality Scores
Centrality Scores

0.3 0.4

0.2 0.35

0.1 0.3

0 0.25
0 10 20 30 40 0 10 20 30 40
Sequence Number Sequence Number
(a) (b)
0.8 0.7

0.7 Primal Graph Primal Graph

0.6
Integrated Graph Integrated Graph
0.6
0.5
Centrality Scores
Centrality Scores

0.5
0.4
0.4
0.3
0.3
0.2
0.2

0.1 0.1

0 0
0 10 20 30 40 0 10 20 30 40
Sequence Number Sequence Number
(c) (d)

Fig. 3. Importance of junctions and roads calculated based on the two road networks by different
measures of centrality: (a) is calculated by degree centrality; (b) is calculated by closeness
centrality; (c) is calculated by betweenness centrality; (d) is calculated by information centrality
(Color figure online)

nearly the same (Fig. 3(c)). That is because the measure of betweenness centrality is the
least affected by the new nodes in integrated graphs.
From the comparison of the roads’ scores in the two network graphs, we can conclude
that the evaluation of junctions’ importance is hardly affected by the road nodes and the
calculation of road’s importance can be more exact when integrated graph method is
used, because the scores are calculated based on the contributions of the road nodes in
the whole network.
 An Integrated Network Modeling for Road Maps 23

4 Real Large-Scale Case: An Analysis of Beijing’s Main Road

Network

In this section, we will use the integrated graph method to model the Beijing’s main-
road map which is a large-scale road network and then simultaneously compute the
importance of junctions and road segments by the three classic global-scale centrality
namely closeness centrality, betweenness centrality and information centrality. As
shown in Fig. 4, there are 4192 junctions and 5166 road segments in the map. The line
only depicts the shape, location and length of a main road between two junctions. As
we have discussed, the road network should be modeled using the integrated graph
method before the importance of roads and junctions is computed. The network which
was built using the integrated graph method has 9358 vertices and 10332 edges and is
shown in Fig. 5, in which red vertices represent junctions and green vertices are roads.

Fig. 4. The main road map of Beijing

24 Z. Song et al.

Fig. 5. The integrated graph of the main road map of Beijing (Color figure online)

Based on the integrated network model, the importance of roads and junctions are
computed by closeness centrality, betweenness centrality and information centrality.
Degree centrality is not considered here, because it is a local-scale evaluated method
and is not suitable to evaluate the importance of road nodes in the integrated graph.
Actually, when the information centrality is used, the junctions’ importance is evaluated
by the group information centrality [11]. Because we consider that if a junction is inva‐
lidated the roads that directly connect it are always useless too.
The cumulative distributions of the three measures of centrality are shown in Fig. 6.
The cumulative distribution probability is defined by , where
N(C) is the number of nodes whose centrality scores equal to C and N is the number of
nodes in the whole network. The centrality scores in the figure are all renormalized by
dividing the maximum ones, so the maximum scores are 1. From it we can have conclu‐
sions as follows:
• Closeness centrality is mainly linear and betweenness centrality shows an obvious
exponential distribution. Information centrality which is the third-class centrality
measure also has an exponential character.
• Importance levels computed by closeness centrality are more uniform than the other
two but they do not cover the whole range 0–1. Betweenness centrality and
 An Integrated Network Modeling for Road Maps 25

information centrality have a better resolving power for the nodes’ importance and
are both cover the whole range.
• We can see from the figure that the distribution curve of information centrality scores
locates between the closeness centrality and betweenness centrality. Maybe it is
because information centrality is the third-class measure of centrality which
combines the two ideas of closeness centrality and betweenness centrality.

1
Closeness
0.9 Betweenness
Information
0.8

0.7

0.6
P(C)

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
Importance Level

Fig. 6. The cumulative distributions of the three classic global-scale measures of centrality in
the integrated graph of Beijing’s main road map. (Color figure online)

Considering information centrality combines the two ideas of closeness centrality

and betweenness centrality, we think it should be one comprehensive method to measure
the importance of junctions and roads in a map, so we have mapped importance of
junctions and roads computed by information centrality in the map and show the
geographic distribution of them in Fig. 7. In the figure, the centrality scores are renor‐
malized by dividing the maximum one and the nodes and edges are marked by a series
of gradual changed colors (as shown at the right bottom of Fig. 7). From the result map,
we found that many important junctions and roads we identified are the real high-grade
roads. As road design in a city is always very scientific and reasonable, this phenomenon
shows the validity of our road network modeling method.
26 Z. Song et al.

Importance Level

0 0.5 1

Fig. 7. The geographic distribution of critical roads and junctions in Beijing’s main road map
(Color figure online)

5 Conclusions

This paper proposes a road network modeling method called integrated graph to settle
the simultaneous computation of the importance of roads and junctions on a road map
in traffic simulations. This method that both roads and junctions on a road map are
represented by nodes integrates the ideas of node presentation in primal graph and dual
graph. Based on this, the importance of junctions and roads can be evaluated by a
measure of centrality once.
The method has been validated by analyzing the importance of locations in a small
road network and Beijing’s main road network. The results show that our modeling
 An Integrated Network Modeling for Road Maps 27

method of road networks is feasible and efficient. The calculation of a road’s importance
in the integrated network is based on its contribution to the whole network, and is more
reasonable than the evaluation based on the junctions on its both sides. Furthermore, the
integrated network modeling method makes the importance of roads and junctions
measured by the same standard, so the importance of roads and junctions can be
compared more accurately and clearly. These improvements are very helpful to identify
the critical junctions or roads on the whole road map in traffic simulations and realize a
more reasonable traffic resources assignment in traffic simulations.

Acknowledgements. The authors would like to thank National Nature and Science Foundation
of China under Grant Nos. 91024030, 91224008, 61503402, and 71303252.

References

1. Demšar, U., Špatenková, O., Virrantaus, K.: Identifying critical locations in a spatial network
with graph theory. Trans. GIS 12, 61–82 (2008)
2. Porta, S., Latora, V.: Centrality and cities: multiple centrality assessment as a tool for urban
analysis and design. In: New Urbanism and Beyond: Designing Cities for the Future, pp. 140–
145 (2008)
3. Wu, J., Gao, Z., Sun, H.: Topological-based bottleneck analysis and improvement strategies
for traffic networks. Sci. China Ser. E Technol. Sci. 52, 2814–2822 (2009)
4. Jiang, B., Claramunt, C.: Topological analysis of urban street networks. Environ. Plann. B
31, 151–162 (2004)
5. Porta, S., Crucitti, P., Latora, V.: The network analysis of urban streets: a primal approach.
Environ. Plann. B Plann. Des. 33, 705–725 (2005)
6. Porta, S., Crucitti, P., Latora, V.: The network analysis of urban streets: a dual approach.
Phys. A Stat. Mech. Appl. 369, 853–866 (2006)
7. Nieminen, J.: On the centrality in a graph. Scand. J. Psychol. 15, 332–336 (1974)
8. Holme, P., Kim, B.J.: Attack vulnerability of complex networks. Phys. Rev. E 65, 056109
(2002)
9. Latora, V., Marchiori, M.: Efficient behavior of small-world networks. Phys. Rev. Lett. 87,
198701 (2001)
10. Freeman, L.C.: Centrality in social networks conceptual clarification. Soc. Netw. 1, 215–239
(1979)
11. Latora, V., Marchiori, M.: A measure of centrality based on network efficiency. New J. Phys.
9, 188 (2007)
 A Novel Flexible Experiment Design Method

Gang Zhai, Yaofei Ma(B) , Xiao Song, and Yulin Wu

College of Automation Science and Electrical Engineering, Beihang University,

Beijing 100083, China
[email protected]

Abstract. Latin Hypercube Design (LHD) is a traditional method of

Design Of Experiments (DOE) and is often employed in system analy-
sis. However, this method imposes restriction on experiment trials and
needs much computation capacity to obtain the optimal design. A novel
experiment design method called ETPLHD is proposed in this paper to
solve this problem. ETPLHD can control the number of design points
and thus presents more flexibility to control the number of experiment
trails, which is more efficient compared to the fixed experiment trails in
the traditional LHD method for a same design space. An experiment was
conducted to compare ETPLHD with the other two experiment design
algorithms. The results showed that TPLHD reveals high design perfor-
mance and less time consumption.

Keywords: Experiment design · Simulation · Latin Hypercube Design

1 Introduction

Simulation is a promising way to study the complex systems with high

performance-price ratio [1]. To analysis the characteristics of the target system,
normally a large number of experiment trails need to be run and different level
combinations of the parameters are tested. While the high-performance compu-
tation facilities are widely used to speed up the computation, the computation
capacity required in system analysis still makes it difficult to test each level com-
bination of all parameters, especial for those complex systems with large set of
parameters. Even a simple application with 5 parameters, and each parameter
contains 10 levels, 106 min (over 2 years) is required to test all level combinations
assuming each trail need 10 min to run.
As a result, the DOE technique is widely used in any experiment-based
domains [2] for its capability to analyze the target system with less experiment
trials. The DOE method normally selects a small amount of typical experiment
points in the parameter space, to obtain the comprehensive understanding of the
target system. It is not necessary to test each level combination of the parame-
ters, thus improving the experiment efficiency greatly. The model of the target
system is expected to be derived based on this small amount of experiments,
then the further analysis even the prediction can be made.

c Springer Science+Business Media Singapore 2016
S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 28–39, 2016.
DOI: 10.1007/978-981-10-2158-9 3
 A Novel Flexible Experiment Design Method 29

A critical property is the space-filling property, i.e., how the design points
distributed in the experiment space. Among various DOE methods, the Latin
Hypercube Design (LHD), which was proposed by McKay [3], is most used in
simulations [4]. It is always the concern in different variants of LHD design to
obtain better space-filling property. Park developed a row-wise element exchange
algorithm to obtain the optimal LHD [5]. Morris and Mitchell applied simulated
annealing algorithm to optimize the design [6]. Bates et al. employed the Genetic
algorithm to optimize LHD [7]. Although these variants present good perfor-
mance, the involved number of experiment trails is large. Felipe Viana et al.
proposed a fast optimal Latin Hypercube Design using Translational Propaga-
tion algorithm (TPLHD) [8]. TPLHD generates nearly optimal design instead of
the global optimal design. However, the number of experiment trials of TPLHD
is fixed for discrete experiment space, which is often insufficient to fully explore
the experiment space.
In this paper, an Extended TPLHD method (ETPLHD) is proposed to gen-
erate the design points flexibly. ETPLHD can generate more points in the design
space than traditional LHDs, which will be helpful in study of the target system.
It was shown in the comparison with other DOE methods that ETPLHD is effi-
cient to produce the experiment points, and achieved better evaluation results
about the studied system.
The rest of this paper is organized as follows. Section 2 gives a brief intro-
duction of TPLHD. Then the proposed method ETPLHD is described in detail.
Section 3 introduces the linear regression model for system approximation and
predication. Section 3.2 presents the comparative experiment between ETPLHD
and the other two DOE methods, and the results analysis is given. Finally, the
remarkable features about ETPLHD approach is concluded in Sect. 4.

2 Method
2.1 Review of Latin Hypercube via Translational
Propagation (TPLHD)

TPLHD method works in real time at the cost of finding the near optimal design
instead of the globally optimal design. Suppose a design space with nv variables,
each variable has np levels. The first step in TPLHD is to create a seed design that
contains ns points. This seed design is used as a pattern to fill the experiment
space iteratively. To fill the space, the experiment space is partitioned into nb
blocks firstly:
nb = np /ns (1)
The number of levels contained by each block is determined as follows:

nd = (nb )1/nv (2)

The second step is to fill the seed design into each block. An example is
illustrated in Fig. 1. A seed design containing only 1 point is created in a 9 × 2
30 G. Zhai et al.

(two variables and each has 9 levels) design space. Then the original space is
divided into 9/1 = 9 blocks, and each block has 91/2 = 3 levels. As Fig. 1(a)
shows, the seed design is placed in the left-bottom block firstly. Then the seed
design is iteratively shifted by np /nd = 3 levels along one dimension until this
dimension is filled with the seed design, as Fig. 1(b) and (c) shows. Next, the
design along this dimension is adopted as a new seed design to fill along other
dimension until all blocks are filled, as shown in Fig. 1(d).

(a) Create a seed design (b) Translate the seed de-

sign to another block along
one dimension

(c) Fill one dimension (d) Fill all dimensions

Fig. 1. The procedure of creating 9 × 2 TPLHD

As all the key operations are to translate the design points, The TPLHD
method requires much less computation compared to other optimal methods
which normally need to search the best design among all (np !)nv LHD designs.
A criterion parameter φp , is used to measure the space-filling quality of the
experiment design. For an experiment space, a smaller φp indicates that the
created experiment points are better in distribution to fill up the space.
np −1
 
np
φp = [ d−p
ij ]
1/p
(3)
i=1 j=i+1
 A Novel Flexible Experiment Design Method 31

where p is a pre-selected integer value and dij is the distance between any two
design points xi , xj :


nv
dij = d(xi , xj ) = [ |xik − xjk |t ]1/t (4)
k=1

φp is also adopted in this paper to measure the space-filling quality of exper-

iment design. As suggested by Jin et al. [9], the value p = 50, t = 1 is taken here.
TPLHD method works well with experiment design less than 6 variables. It is
difficult to approximate a good experiment design in high-dimensional space for
2 reasons: (i) the Curse of Dimensionality. The partitioned blocks in experiment
space will grow exponentially with dimensions; (ii) the distribution of experiment
points will be asymmetry in high-dimensional space with linear partition.

2.2 The Extented TPLHD (ETPLHD)

The number of experiment points in TPLHD is constrained by np , which is the

levels of the variable. This characteristic will lead to a small point set most of
the time. For example, a size of 10×5 experiment design space contains total 51 0
level combinations, however, only 10 experiment points would be chosen with
TPLHD method. Obviously, this small amount of experiments is insufficient to
analysis the target system.
To increase the experiment points, the ETPLHD s proposed in this paper
to provide a layered design approach to obtain better distribution of the exper-
iment points. The horizontal interpolation is employed in ETPLHD to design
the experiment in an expanded space and then is scaled into the origin space.
Assuming all variables having the same number of levels, Eq. (2) is modi-
fied as:
d = np1/2 (5)
d is expected to be an integer value, thus np needs to be rounded as the value
that can be squared such as 4, 9 or 16. For instance, np = 15 can be rounded
up to 16, i.e. one extra level is added into the experiment space, then the points
corresponding to the extra level are eliminated at last to correct the design.
ETPLHD holds that each block contains the same number of points as
TPLHD. However, the number of blocks in ETPLHD is much bigger than that
in TPLHD for cases with more than two dimensions, leading to the increment
of the design points in ETPLHD (denoted as n∗ ).

n∗ = dnv = nnp v /2 (6)

Consider a n × m (n levels, m variables) experiment space. Initially, one

dimension (variable) is chosen to be divided into intervals according to Eq. (5).
After this operation, the experiment design is conducted in a subspace n×(m−1)
within each block, where the dimension that has been divided is excluded. The
design in subspace n×(m−1) takes the same steps:chose one dimension to divide,
32 G. Zhai et al.

(a) A TPLHD design (b) Translate within 1 (c) 4 layer of 16×2 space
layer

Fig. 2. The procedure to create 16 × 3 ETPLHD

and then perform the experiment design in a lower dimensional subspace, until
the TPLHD method is applied.
Figure 2 illustrates this process taking a design in 16 × 3 experiment space.
First, one dimension (denoted as the 1d ) of the experiment space is chosen and
partitioned into 4 parts according to Eq. (5). 4 intervals along this dimension are
determined, as shown in Fig. 2(a). Next, a 16 × 2 TPLHD designs (the dimen-
sions are denoted as 2d and 3d ) is conducted in each interval. By this way, the
ETPLHD generates 64 points. For comparison, TPLHD obtains a design with
16 points.
In this example, the experiment points within low-dimensional subspace need
to be expanded to the higher dimension. Generally, if the points set xim has been
obtained within a subspace of m dimensions, the design points can be expanded
to the higher dimension as follows:

xki(m+1) = xim mod d + (k − 1) · d (7)

where k(1 ≤ k ≤ dm+1 ) is the index of the interval along dimension m + 1. The
expansion is repeated until all dimensions are included. As for the example in
Fig. 2, the experiment points form the slanted layer along the dimension 1d , as
shown in Fig. 2(c).
After the expansion to higher dimension, some points may stay in a small
region of the experiment space. Consider the example in Fig. 2, an experi-
ment point [x0 , y0 ] designed in the 2d × 3d subspace will generates a series
experiment points along dimension 1d by taking the expansion operation:
[x0 , y0 , z0 ], [x0 , y0 , z3 ], [x0 , y0 , z7 ], [x0 , y0 , z11 ]. From the direction of the dimen-
sion 1d , they are lines. This problem will be worse when the dimension grows.
As a result, it is necessary to adjust the distribution of the experiment points
before expansion to the higher dimension. Assuming the current dimension is m,
then the points are shifted p(0 ≤ p ≤ dm ) unit along the 2 ∼ m dimension, where
dm is the number of the intervals along dimension m. Equation (7) is changed as
follows:
xki(m+1) = (xim + p)mod d + (k − 1) · d (8)
 A Novel Flexible Experiment Design Method 33

There are two issues should be noted. First, the value of p is normally deter-
mined empirically. Second, a control parameter s can be defined to specify how
many dimensions need to apply the ETPLHD method. For a n × m space, s = 2
means only two dimensions apply the ETPLHD method and the rest m−2 dimen-
sions will apply the TPLHD method. By the control parameters, the number of
experiment points determined by ETPLHD is:
s
+1
n∗ = np · ds = np2 (9)
Figure 3 demonstrates the shift within the 2d × 3d subspace before expansion
to dimension 1d .

Fig. 3. Translate the seed on each layer

The pseudo code of ETPLHD is shown as follows:

Algorithm 1 : ArrayCreateET P LHD(m, n, s)

var
m: the number of the design variables (dimensions) of the
experiment space. The first dim refer to sub-population.
n: the levels of each variable.
s: the control variable specifying how many dimensions should
be applied the ETPLHD design.
begin
d=sqrt(n);
Array seed = CreateTPLHD(m-s,n); //Assuming avaiable
Array res;
for(dimension=m-s+1:m){
for(layer=1:d){
Add the seed design into res;
Translate the seed by current dimension;
}
seed=res;
}
return res;
end.
34 G. Zhai et al.

2.3 DOE Methods Comparison

Three DOE methods, ETPLHD, TPLHD and a random-selection based LHD

method lhsdesign (the implemented function name in Matlab) are compared to
demonstrate the effectiveness of ETPLHD. The lhsdesign method selects the
LHD according to the best max-min criterion, i.e., the maximization of the
minimum distance dmin , from 200 random LHDs. dmin is the other measure of
the points density (similar with φp ); a bigger dmin indicates better distribution
of the points.
In a nv × np (nv levels, np variables) design space, the expected experiments
trials n∗ can be determined by Eq. (9). With TPLHD and M atlabT M method,
nv × n∗ designs were conducted. In order to compare the three methods based
on the same ground, the design values were divided by nsd and then round up
into [1, np ], to generate the same number of design points with ETPLHD. Three
criterions φp in Eq. (3), dmin and Time consumption(s) are compared in the
three methods.
All experiments were conducted in W indowsT M 7 with Intel Core i5-3470
CPU (3.20 GHz), 4 GB RAM, M AT LAB T M (R2012a).
Table 1 shows the results of the three methods. The best φp is shown with
bold. It can be concluded that ETPLHD method performs best under experiment
space with no more than 5 variables. The only exception is the case where
np = 4, nv = 4 and s = 1. For the design space with 6 variables, TPLHD
performs worse than lshdesign with most cases, but ETPLHD is superior to
lshdesign basically. In addition, ETPLHD can generate points with least time
consumption among the three methods. For all cases, ETPLHD and TPLHD cost
less time to obtain an optimal design. By contrast, lhsdesign requires seconds to
minutes for a large experiment space.

3 Apply ETPHLD in System Predication

3.1 Linear Regression Prediction Model

The goal of DOE method is to evaluate the target system with less experi-
ment trails. Many approximation methods such as the linear regression, Kriging
model, neural nets, support vector regression and so on are often employed to
evaluate and then predict the target system. The linear regression is the most
used approach among them10 .
A first-order polynomial can be given by

y = Xβ + e (10)

where y = (y1 , . . . , yn ) is the predicted value of the target system with n experi-
ments. X = (xij )(i = 1, . . . , n, j = 1, . . . , q) is the experiment data recorded in n
trails, where i in the index of experiment trail, and j is the index of data within
each trail. β = (β1 , . . . , βq ) is the regression coefficients; and e = (e1 , . . . , en )
denotes the residuals in each experiment.
 A Novel Flexible Experiment Design Method 35

Table 1. Performance comparison among ETPLHD, TPLHD and lshdesign

ETPLHD TPLHD lhsdesign

np nv s φp dmin Time φp dmin Time φp dmin Time
3 4 1 2 0.5 ≈0 2.056 0.5 ≈0 2.027 0.5 0.1
9 1 2.405 0.444 ≈0 3.178 0.333 ≈0 4.6 0.222 0.1
16 1 3.485 0.312 ≈0 4.322 0.25 ≈0 8.291 0.125 0.3
25 1 3.485 0.312 ≈0 4.322 0.25 ≈0 8.291 0.125 0.3
4 4 1 1.333 0.75 ≈0 1.014 1 ≈0 1.351 0.75 0.1
2 2.065 0.5 ≈0 2.114 0.5 ≈0 2.065 0.5 0.1
9 1 2.281 0.444 ≈0 2.313 0.444 ≈0 3 0.333 0.1
2 3.237 0.333 ≈0 3.294 0.333 ≈0 9.199 0.111 0.43
16 1 2.704 0.375 ≈0 3.304 0.312 ≈0 5.333 0.187 0.3
2 4.439 0.25 ≈0 4.505 0.25 ≈0 16.35 0.062 3.2
25 1 4.283 0.24 ≈0 5.000 0.2 ≈0 6.337 0.16 0.85
2 5.664 0.2 ≈0 5.740 0.2 ≈0 25 0.04 18.2
5 4 1 1.014 1 ≈0 1.333 0.75 ≈0 1.333 0.75 0.1
2 2.027 0.5 ≈0 2.056 0.5 ≈0 2.027 0.5 0.1
9 1 1.521 0.667 ≈0 2.281 0.444 ≈0 1.8 0.555 0.1
2 3.174 0.333 ≈0 4.626 0.222 ≈0 4.562 0.222 0.4
16 1 2.073 0.5 ≈0 3.2 0.312 ≈0 2.286 0.437 0.3
2 4.055 0.25 ≈0 4.222 0.25 ≈0 4.169 0.25 3.2
25 1 2.276 0.44 ≈0 3.671 0.28 ≈0 3.126 0.32 0.9
2 4.285 0.24 ≈0 5.348 0.2 ≈0 6.425 0.16 18.1
6 4 1 0.681 1.5 ≈0 0.8 1.25 ≈0 0.681 1.5 0.1
2 2 0.5 ≈0 2 0.5 ≈0 1.333 0.75 0.1
3 2 0.5 ≈0 2.108 0.5 ≈0 2.056 0.5 0.15
9 1 1.045 1 ≈0 1.533 0.666 ≈0 1.125 0.888 0.15
2 2.383 0.444 ≈0 2.388 0.444 ≈0 2.281 0.444 0.45
3 3.258 0.333 ≈0 4.767 0.222 ≈0 4.562 0.222 3
16 1 1.380 0.75 ≈0 2.173 0.5 ≈0 1.623 0.625 0.3
2 3.468 0.312 ≈0 4.222 0.25 ≈0 3.271 0.312 3.3
3 4.203 0.25 ≈0 8.224 0.125 0.3 5.527 0.187 49
25 1 1.701 0.64 ≈0 3.571 0.28 ≈0 1.927 0.52 0.9
2 3.966 0.28 ≈0 3.488 0.32 ≈0 4.166 0.24 18
3 4.697 0.24 ≈0 5.520 0.2 0.2 8.605 0.12 443.9

Let w = (w1 , . . . , wn ) be the true output of the simulation system, the Sum
of Squared Residuals(SSR) is given by Eq. (11)

SSR = (y − w) (y − w) (11)

36 G. Zhai et al.

The coefficient vector β is defined as follows to compute the predication:

β = (X  X)−1 X  w (12)

The accuracy of the model is normally measured by the mean square error
(MSE ), which is the accurate estimate of the true error of the prediction model.

MSE = SSR/n (13)

where n is the number of experiment trials.

3.2 A Complex Example: The Combat Simulation

There are blue and red force in the combat scenario. The blue force includes a
formation of fighters, who attempts to cross a strait to attack the ships of the
red. On the other hand, the red ships have the anti-air capacity. Once the blue
fighters flight close to the ships, the SAM missiles will be launched to protect
the ships (Fig. 4). The relationship between the performance of SAM missile and
the shoot-down number of blue fighters is concerned. The combat is simulated
using a Computer Generated Force (CGF) platform [10].

Fig. 4. The two dimension display of the scenario

Five parameters of SAM missile are adopted as the design variables and each
has 4 levels, as shown in Table 2.
TPLHD and ETPLHD are tested in this scenario for comparison. In
ETPLHD, s = 3 produces 32 experiment points. Each experiment point is tested
for 5 times to get the mean value of the output for further analysis. The linear
regression approach is employed to obtain an approximated system model that
can be used to predict the output of the combat. Table 3 demonstrates a part
of the results and the predictions of the approximated models based on the two
DOE methods.
 A Novel Flexible Experiment Design Method 37

Table 2. Design parameters and levels

Variables (of the missiles) Levels

1 2 3 4
Velocity X1 (km/h) 700 800 900 1000
Kill Radius X2 (m) 25 50 75 100
Range X3 (km) 250 500 750 1000
Bomb Loads X4 4 6 8 10
Killing Probability X5 0.4 0.5 0.6 0.7

Table 3. The true results and the predictions

(a)TPLHD
No X1 X2 X3 X4 X5 y ypred
1 1 1 1 1 1 8.2 10.2520
2 3 1 1 3 2 17.2 18.1212
3 4 2 2 2 3 14.4 13.6219
4 2 2 4 4 4 27 28.7731
· · · · ··
32 4 4 4 4 4 26.2 25.2967
(b)ETPLHD
No X1 X2 X3 X4 X5 y ypred
1 3 2 1 1 1 8 8.2456
2 3 3 4 2 1 15.6 17.7121
3 2 1 1 3 2 20.4 19.7148
4 4 1 3 2 3 17.2 16.0479
· · · · ··
32 4 3 1 1 3 8 6.7885

Given the data in Table 3, the MSE of the two methods can be calculated
according to Eq. (13). The MSE of ETPLHD is 0.6141, which is much less than
that (1.0533) of TPLHD. This result indicates that the approximated system
model using the ETPLHD method performs better than the model using TPLHD
method.
20 random points are selected from the experiment space to verify the effec-
tiveness of the approximated models. Table 4 demonstrates a part of the results.
ya is the prediction by the approximated model that uses the TPLHD method,
while yb is the predication by the model that uses the ETPLHD method.
The plots of the simulation values and two prediction values are shown in
Fig. 5. The MSE of the two predication is 4.0147 and 3.217 respectively, indi-
cating that the prediction model with ETPLHD is more precise.
38 G. Zhai et al.

Table 4. Part of results of 20 random points with the two methods

No X1 X2 X3 X4 X5 y ya yb
1 3 1 3 3 4 24.2 22.1132 22.0739
2 2 3 2 2 1 17 14.1160 15.4181
3 4 3 1 2 1 13.2 10.4750 11.6806
4 1 3 1 3 1 21.2 18.4052 19.7464
· · · · ··
20 4 4 2 1 2 8 7.0334 7.9351

Fig. 5. The simulation and two prediction values of random points

4 Conclusion

In order to break the limitation of the fixed number of design points in tra-
ditional LHD method, a new flexible DOE method ETPLHD is proposed in
this paper. The comparison shows that ETPLHD method performs better than
TPLHD for no more than 6 variables. The ETPLHD method is also applied
in a complex combat simulation to help to find proper approximation model
of the combat system, which is used to predict the outcomes of the combat
 A Novel Flexible Experiment Design Method 39

with different parameter configurations of the SAM missiles. The results show
that the approximation model using the ETPLHD method has better predicting
accuracy than the model using the TPLHD method.

References
1. Sanchez, S.M., Wan, H.: Work smarter, not harder: a tutorial on designing and
conducting simulation experiments. In: Proceedings of the 2012 Winter Simulation
Conference, pp. 1–15 (2012)
2. Ghosh, S.P., Tuel, W.G.: A design of an experiment to model data base system
performance. IEEE Trans. Softw. Eng. SE-2(2), 97–106 (1976)
3. McKay, M.D., Beckman, R.J.: A comparison of three methods for selecting values
of input variables from a computer code. Technometrics 21(2), 239–245 (1979)
4. Kelton, W.D.: Designing simulation experiments. In: Proceedings of the 1999 Win-
ter Conference, pp. 33–38 (1999)
5. Park, J.S.: Optimal Latin-hypercube designs for computer experiments. J. Stat.
Plann. Inference 30(1), 95–111 (1994)
6. Morris, M.D., Mitchell, T.J.: Exploratory designs for computational experiments.
J. Stat. Plann. Inference 43(3), 381–402 (1995)
7. Bates, S.J., Sienz, J., Toropov, V.V.: Formulation of the optimal Latin hyper-
cube design of experiments using a permutation genetic algorithm. In: 45th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials
Conference, AIAA-2004-2011, Palm Springs, California, pp. 19–22 (2004)
8. Viana, F.A.C., Venter, G., Balabanov, V.: An algorithm for fast optimal latin
hypercube design of experiments. Int. J. Numerical Methods Eng. 82(2), 135–156
(2009)
9. Ye, K.Q., Li, W., Sudjianto, A.: Algorithmic construction of optimal symmetric
latin hypercube designs. J. Stat. Plann. Inference 90, 145–159 (2000)
10. Ma, Y., Gong, G.: A research on CGF reasoning system based on Fuzzy Petri
Net. In: Proceedings of System Simulation and Scientific Computing, Beijing, pp.
406–411 (2005)
 Modeling Behavior of Mobile Application
Using Discrete Event System Formalism

Yun Jong Kim, Ji Yong Yang, Young Min Kim, Joongsup Lee,
and Changbeom Choi ✉
( )

Handong Global University, Newton 206 558 Handong-ro Buk-gu, Pohang,

Gyeongbuk 37554, Republic of Korea
{21100132,21000416,21100114,joongsup,cbchoi}@handong.edu

Abstract. Discrete Event System Specification (DEVS) is a set theoretic

formalism developed for specifying discrete event systems. DEVS features the
specifications of hierarchical system and sequential event. Mobile application
system is one of the typical systems that have hierarchical structure and sequential
processes. However, there has been no attempt to express the mobile application
system as DEVS formalism. This paper describes the design and development of
mobile application behavior using DEVS model and we have classified the types
of the mobile app behaviors as: temporal behavior, non-temporal behavior. Each
behavior is stated as an atomic model and the mobile app events are represented
as a coupled model. By using the DEVS formalism, not only we can simulate
most of the mobile application events using real device; but also we expect that
this work will serve as an effective tool for usability test.

Keywords: DEVS · Mobile application · Simulation environment · DEVS

modeling · Model implementation · Real device

1 Introduction

According to a new market research report published by ‘VisonMobile’, the total global
mobile applications market is expected to be worth US$58.0 billion by the end of 2016.
The market of the mobile application is growing fast and user’s requirement is also
becoming diversified. To further, there have been many studies and attempts to find
solutions that reflects the user’s needs efficiently.
One of the biggest challenges in mobile application development is that the require‐
ments for mobile application change frequently and rapidly [1]. Many developers have
difficulties to keep the code up to date and it takes a lot of time. To deal with this problem,
rapid prototyping has been used [2]. In this methodology, research and development
stages are conducted in parallel to another when creating the prototype [3]. This enables
developer and designer to acquire precise feedback and analyze user requirements by
providing high-fidelity prototypes.
In order to develop a high-fidelity prototype, the developer should consider the
behavior of mobile application. The typical behavior of mobile application can be
in turn categorized into temporal behavior and non-temporal behavior. The temporal

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 40–48, 2016.
DOI: 10.1007/978-981-10-2158-9_4
 Modeling Behavior of Mobile Application 41

behavior does not require users’ input, but instead it is triggered by external sources.
Display of animation on mobile screen and pop-up message from mobile applica‐
tions are such examples. On the other hand, non-temporal behavior is the event that
requires user’s input. Screen touch, swipe and long touch are some examples of this
behavior. We have applied discrete event system specification formalism to monitor
these behaviors.
This paper contains five sections. Section 2 describes the related works concerning
the making of high-fidelity mobile application prototype. Section 3 introduces how to
model the temporal behavior and non-temporal behavior using DEVS formalism, which
sets the theoretical background of our study. Furthermore, we explain the overall struc‐
ture of the simulation using DEVS formalism. Section 4 introduces the case study with
the proposed approach and we conclude our study in Sect. 5.

2 Related Works

Mobile application can be grouped by the three categories: native, web-based, and hybrid
[4, 5]. All three types of mobile applications needs prototyping process under the devel‐
opment in common. To support the process, many prototyping tools currently exist in
the market.
The table below shows the limits of existing tools. We compared four well-known
prototyping tools with our proposed approach. There are several prototyping tools that
support making of high-fidelity prototype. These tools simulate the mobile application
by using the expected behavior information embedded in the pre-designed image. Note
the existing tools only support the non-temporal behavior. Also, it is not possible to
discrete each event with formal representation for the user to obtain additional infor‐
mation about their prototype (Table 1).

Table 1. The comparison between existing prototyping tools and proposed system
Marvel POP Fluid Axure Proposed system
Modeling non- O O O O O
temporal behavior
Modeling temporal X X X X O
behavior
Formal representation X X X X O

3 Modeling and Simulation of Mobile Application Formalism

3.1 Discrete Event System Specification Formalism

The DEVS formalism supports the structure of a system in modular and hierarchical
point of view. There are two kinds of DEVS model: one is atomic model and the other
is coupled model [6]. While a part of system can be represented as an atomic model, a
system represented in DEVS coupled model shows how a system can be coupled
together and also how they interact with each other.
42 Y.J. Kim et al.

The coupled model is composed of various models. This may express larger system
by taking an atomic model or a subordinated coupled model as a child model. A coupled
model ‘CM’ is defined as:

CM = < X, Y, {Mi}, EIC, EOC, IC, SELECT>

X set of internal input events;

Y set of external output events;
{Mi} set of components
EIC external input coupling
EOC external output coupling
IC internal coupling
SELECT the tie-breaking function to arbitrate occurrences of simultaneous events

The Atomic model is the most basic module in the hierarchical structure. ‘Mi’ in the
CM corresponds to the atomic model. It describes the system behavior. An atomic model
‘M’ is defined as follows:

M = < X, S, Y, S, δext, δint, λ, ta > where

X set of internal input events

S set of state
Y set of external output events
δext external transition function specifying state transitions based on the external events
λ output function generating external events as output
ta time advance function

Mobile application can be viewed as the collection of various components and this
can be expressed by using DEVS formalism models. To explain in detail, we use the
term ‘Activity’ in Android, which is an application component that provides a screen
for the users to interact. As shown in Fig. 1, one activity consists of many mobile

Fig. 1. The composition of mobile application activity with DEVS formalism model
 Modeling Behavior of Mobile Application 43

application components. Each mobile application component corresponds to the proper

atomic model concerning the type of behavior. For example, all the behavior related to
the button component is the non-temporal behavior model and this can be expressed as
an atomic model describing non-temporal behavior. On the other hand, the behavior in
which the animation is displayed on screen can be represented as an atomic model and
classified as temporal behavior. Such activities have various components that correspond
to the coupled model of many atomic models.

3.2 Modeling Non-temporal Behavior of Mobile Application

Figure 2 shows how the non-temporal behavior model works. In this model, there are
three states: QUEUE, WAIT and SEND. The initial state is WAIT and this takes place
when the system is ready to receive the input event. As the non-temporal behavior model
receives the event data, the system turns into the SEND state. The model sends out an
execution message through Output1 through state transitions from SEND to QUEUE or
WAIT. When the queue is empty, the system becomes WAIT; and when the queue size
is above zero, it reaches the QUEUE state. Overall, this model generates non-temporal
event, which in turn can be performed in the Execution model, which we will cover in
Sect. 4.

Fig. 2. The details of non-temporal behavior model

3.3 Modeling Temporal Behavior of Mobile Application

Figure 3 describes how the temporal behavior model works. This study conveys one of
temporal behavior examples: the animation model. The animation model simulates the
gradual change of specific area of the screen without user input. This model has three
states: WAIT, PROCESS and RESULT. At the initial state WAIT, the animation area
displays the initial image. Internal transition function occurs after TW seconds and the
state of the model changes into PROCESS. For each stay, it spends TP seconds. Temporal
behavior model stays at PROCESS state N times by increasing count. This means the
model invokes the output function N times and gradually changes the image for each
function-call. When the count becomes equal to N, the system reaches the RESULT
state. In this stage, the model changes the screen image into the last one that is to be
44 Y.J. Kim et al.

shown in the animation. After TR seconds, the state goes back to the initial state and the
cycle is repeated.

Fig. 3. The details of temporal behavior model

3.4 Modeling Execution Model

The execution model sends output message to implement specific event after given time,
Tn. Note the amount of time taken is noted as Tn. Tn corresponds to the number that
the event data includes. Figure 4 demonstrates the details of Execution model.

Fig. 4. The details of execution model

There are two states: WAIT, EVENT. The initial state is WAIT. Once the message
is arrived from the behavior atomic model, the state changes to EVENT. Consequently,
the model turns itself back into WAIT state through the internal transition and sends
message allowing the mobile OS to operate.
 Modeling Behavior of Mobile Application 45

3.5 The Mobile Simulation Using DEVS Formalism

The simulation architecture using DEVS formalism that this paper suggests is depicted
in Fig. 5. This is composed of three parts; Mobile OS, Event convertor for mobile appli‐
cation (ECMA) and DEVS model describing the behavior of mobile application. This
system receives the input event that is going to be simulated through the mobile OS.
Basically, the Input event means the event time and the type of event. These events are
transferred to DEVS formalism models through ECMA. The input message is trans‐
ferred through callback function in the ‘input event convertor’. When the behavior
atomic model receives the input message, the system generates the output message and
the execution model is transferred into DEVS message. The output event convertor in
ECMA then transfers the message via mobile OS by making system API call for each
type of event.

Fig. 5. The simulation structure of mobile application

4 Case Study

In this section, we introduce two specific events which can be expressed by using the
behavior models we have proposed earlier. Note various features of simulation using
the behavior model can be seen from the case study. The events that we simulate are
twofold: (1) Activity conversion by button touch and (2) Image conversion with the
lapse of time.

4.1 Activity Conversion by Button Touch

Activity conversion by touch button is one of the events that is triggered by the user’s
input. Figure 6 describes the event with DEVS formalism models. This event can be
expressed as the coupled model, which comprises the non-temporal behavior model and
46 Y.J. Kim et al.

the execution model. The execution model of non-temporal behavior enables the conver‐
sion to occur and as the result, the image displayed on screen changes into the upper-
left image shown in Fig. 6.

Fig. 6. Activity conversion by button touch

4.2 Image Conversion with the Lapse of Time

The image conversion event in Fig. 7 requires current activity as an input source.
Because the direct user input is unnecessary for this event, the event in Fig. 7 can be
expressed as the coupled model, which comprises the temporal behavior model and the
execution model. The execution model takes some time as temporal behavior model,
and then changes the screen image into the upper-right image shown in Fig. 7.
 Modeling Behavior of Mobile Application 47

Fig. 7. Image conversion with the lapse of time

5 Conclusion

Mobile application system is a highly hierarchical and time-based process [7]. Because
of these characteristics, DEVS formalism is one of the appropriate ways to model mobile
application system. However, there has been no attempt to express mobile application
workings through DEVS.
This study simulates the mobile application by stating the mobile app behavior in
DEVS models and introduces how to model the mobile app behavior. We classified by
the type of the mobile app behavior; and made models of such behavior using the DEVS
atomic model. By combining the behavior atomic model and coupled model we can state
the various mobile application events.
The approach that we proposed represents the temporal behavior which is not
possible to state in existing prototyping tools. This helps to implement the prototype
with high fidelity. In addition, the mobile app simulation using DEVS model enables to
accumulate formal data. This means that the simulation environment that we are
proposing can in turn be used to test usability by analyzing the accumulated data. Many
other industries utilize the data from the prototype to analyze the task-based usability
48 Y.J. Kim et al.

test [8]. Moreover, the formal method, which is ensured by DEVS based simulation
environment, has a mathematical basis [9]; thus with the formal method, the user can
specify, develop, and verify the prototype in systematic manner.
However, the implementation framework still does not exist; hence the user must
develop the entire simulator model from the bottom up. Also, simulator heavily depends
on platform because of the absence of simulation engine. For further research topics we
may look into simulation engine and finding ways to implement the device-independent
framework.

References

1. Joorabchi, M.E., Mesbah, A., Kruchten, P.: Real Challenges in Mobile App Development. In:
International Symposium on Empirical Software Engineering and Measurement, IEEE
computer society, pp. 15–24 (2013)
2. Adelmann, R., Langheinrich, M.: SPARK rapid prototyping environment – mobile phone
development made easy. In: Tavangarian, D., Kirste, T., Timmermann, D., Lucke, U., Versick,
D. (eds.) IMC 2009. CCIS, vol. 53, pp. 225–237. Springer, Heidelberg (2009)
3. Tripp, S.D., Bichelmeyer, B.: Rapid prototyping: an alternative instructional design strategy.
ETR&D 38(1), 31–44 (1990). ISSN 1042-1629
4. Masi, E., Cantone, G., Mastrofini, M., Calavaro, G., Subiaco, P.: Mobile apps development: a
framework for technology decision making. In: Uhler, D., Mehta, K., Wong, J.L. (eds.)
MobiCASE 2012. LNICST, vol. 110, pp. 64–79. Springer, Heidelberg (2013)
5. “Native, web or hybrid mobile-app development,” IBM Software, Thought Leadership
WhitePaper. http://www.computerworld.com.au/whitepaper/371126/native-web-or-hybrid-
mobile-app-development/download/
6. Zeigler, B.P.: Multi-facetted Modeling And Discrete Event Simulation. Academic Press, San
Diego (1984)
7. Zeigler, B.P., Chow, A.C.H.: Parallel DEVS, hierarchical, modular, modeling formalism
(1994)
8. Stone, B., Wang, Y.: AirportLogic: usability testing, prototyping, and analysis of an airport
wayfinding application. In: Stephanidis, C. (ed.) Posters, Part II, HCII 2011. CCIS, vol. 174,
pp. 81–84. Springer, Heidelberg (2011)
9. Seo, K.-M., Choi, C., Kim, T.G., Kim, J.H.: DEVS-based combat modeling for engagement-
level simulation. Simul. Trans. Soc. Model. Simul. Int. 90(7), 759–781 (2014)
 Multi-modal Transit Station Planning Method
Using Discrete Event System Formalism

Jaewoong Choi, Bitnal Kim, Onyu Kang, Seonwha Baek,

Yonghyun Shim, and Changbeom Choi ✉
( )

Handong Global University, Newton 206 558 Handong-ro Buk-gu,

Pohang, Gyeongbuk 37554, Republic of Korea
{21000749,21100089,21400008,21400358,21100384,
cbchoi}@handong.edu
http://www.handong.edu

Abstract. A multi-modal transit station is a complex system in the urban system

and takes important roles in our daily life. Since various transportation systems
a large scale of passenger participates to the station, design and analysis of the
multi-modal transit station are important problem. This paper presents a modeling
and simulation method to support designing the transit station. First, we introduce
the modeling method to build a simulation models to capture the behavior of
vehicles and the crowds using the Discrete Event System (DEVS) formalism.
Then, we introduce a simulation environment for multi-modal transit station
which is based on the DEVSim++. In this simulation environment, we support
requirement verification method based on the untimed DEVS model, and it checks
the requirements by executing the simulation model and the untimed DEVS
model altogether.

Keywords: Transportation simulation · DEVS formalism · Agent based

simulation

1 Introduction

A transportation takes an important role to support a member of modern society to move

places to places. Especially, rapid urbanization has been changed a citizenʼs daily life
dramatically. In urbanized society, large scale of the citizen utilizes various transporta‐
tion method to move from home to their work places, and vice versa. As the urbanization
accelerates, the importance of designing the multi-modal transit station increases. In
multi-modal transit station, various transportation arrives to the station with its own time
schedule and a large scale of passenger may arrive to the station to utilize a transporta‐
tion. As a result, an urban planner or an architect should consider several consideration
points, such as traffic, volume of the passenger flow, and other transit modes to design
the multi-modal transit station. If an urban planner underestimates the traffic flow to the
multi-modal transit station, the access roads to the multi-modal transit station will be
too narrow to handle the traffics. On the other hand, when the planner overestimates the
traffic flow, the size of the access roads will be too wide, and it will be waste of the

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 49–64, 2016.
DOI: 10.1007/978-981-10-2158-9_5
50 J. Choi et al.

resources. Moreover, it should spend a huge budget and time to fix such problems after
the station has been built. Modeling and simulation is one of the alternatives to help the
planner to design multi-modal transit station before the construction. By using modeling
and simulation, an urban planner may consider the traffic flow and volume of the
passenger flow before building the multi-modal transit station. Also, the planner may
consider the geographical layout of access roads to the transit station and the behavior
of the individual passenger to predict the traffic flow and volume of the passenger flow.
Modeling and simulation of urban planning is nothing new to urban planning domain.
There were several studies that utilize the modeling traffic and the layout of the road in
the cities and simulate them to acquire data to find the congestion point. Especially, a
traffic simulation model is widely used to analyze the alternatives of transportation
system without directly changing the real world, such as constructing additional road or
modulating traffic signals [1, 2]. In order to capture such behaviors of traffics, passengers,
and pedestrian, two modeling and simulation methods have been proposed: macro
simulation model and micro simulation model. Both models were used to capture the
characteristics of the traffic of a transportation system. However, the previous studies
were used to predicting the traffic situations. Also, to the best of the authorsʼ knowledge,
there are no studies that confirm whether the design of multi-modal transit station satis‐
fies the requirements or not.
In this paper, we propose Discrete Event System (DEVS) Formalism based modeling
method to capture the behavior of the vehicle and passenger. Also, we propose the
simulation environment to check the transit station model against to the requirement.
The DEVS Formalism is set theoretical formalism which has hierarchical structure and
modular feature. First, we use model of the DEVS Formalism, the DEVS model, to
model spatial layout of the multi-modal transit station in order to capture the behavior
of the vehicles and passengersʼ movements. Especially, we model the spatial layout by
connecting DEVS model consecutively. Also, we model temporal and non-temporal
behaviors of the vehicle and passenger as discrete events from previous DEVS model
to next DEVS model. Also, we model the interaction between vehicle and passenger by
exchanging discrete event from DEVS model to DEVS model.
Second, we propose simulation environment that supports various requirement
specification methods to check the model against to its requirements. In this paper, we
mainly utilize requirement specification method based on the propositional logic.
However, in order to support various requirement specification methods, we unify the
modeling method and the requirement specification method using DEVS model. The
model for requirement specification continuously monitors the event sequences from
model of the multi-modal transit station, and checks the temporal and non-temporal
requirements. A preliminary version of this paper appeared in AsiaSim ʼ15. In the
previous paper, there are no details in modeling method and simulation environment.
This paper is an extended and improved version of the preliminary version. The rest of
this paper is organized as follows. Section 2 presents related works. In Sect. 3 introduces
a modeling method for multi-modal transit station, and Sect. 4 presents the simulation
environment for multi-modal transit station. Finally, Sect. 5 shows the case study of our
proposed method and we conclude this paper.
 Multi-modal Transit Station Planning Method Using DEVS Formalism 51

2 Related Works and Background

In this section, we introduce the previous researches in modeling and simulation in urban
planning domain and modeling and simulation theories. The previous researches can be
categorized by macro simulation and micro simulation. The macro simulation model
captures the characteristics of transportation system in macroscopic point of view rather
than modeling individual elements. On the other hand, the micro simulation model
models each individual transportation element to analyze the interaction among various
transportation elements.

2.1 Related Works

The macro simulation model abstracts the type of the vehicles into the differential equa‐
tions and captures the macroscopic characteristics, such as aggregated traffic flow
dynamics and density of the traffic [3, 4].
A macro simulator has advantage on understanding traffic flows, but hard to predict
behavior of each object. For example, a macro simulation model DYNEMO was
designed for the development, evaluation and optimization of traffic control systems for
motorway networks [4]. Traffic flow model included with the simulation package
combines the advantages of a macroscopic model (computational simplicity) with the
advantages of a microscopic model (output statistics relating to individual vehicles).
On the other hand, most studies about micro simulations were concentrated on indi‐
vidual vehicles, crossroads or traffic signal based simulations. For example, Errampalli
et al. has provided a basis for evaluating the public transport policies such as public
transport priority systems by developing microscopic traffic simulation model [2]. They
have considered two types of public transport policies for evaluation. One is bus lanes
and another is public transport priority systems at traffic signal. Fuzzy logic reasoning
has been incorporated in route choice analysis while choosing the route based on the
level of satisfaction of the available routes. The developed simulation model has been
applied on the part of Gifu city network and it is able to predict the vehicular movements
with a fair amount of accuracy. From this study, it can be concluded that the developed
microscopic simulation model can be applied to evaluate public transport polices partic‐
ularly bus lane and public transport priority systems at intersections with fair amount of
accuracy.
Among various modeling and simulation theories in micro simulation, cellular
automata and agent based simulation are the representative theory that model the trans‐
portation system and simulate them [5]. Especially, Zamith et al. has introduced a flex‐
ible and robust traffic cellular automata model [5]. They consider the motion expectation
of vehicles that are in front of each driver. Then, they define how a specific vehicle
decides to get around, considering the foreground traffic configuration. The model
utilizes stochastic rules for both situations, using the probability density function of the
beta distribution to model three drivers behavior, adjusting different parameters of the
beta distribution for each one. The three drivers are as follows. First is Hunter and
Tailgater that describes the behavior of a driver that intends to stay close to the upfront
vehicle and wants to go fast. Second is Ultraconservative that describes the driver with
52 J. Choi et al.

the opposite profile of the Hunter and Tailgater. This driver prefers to keep a great
distance to the upfront vehicle and drives slower. Third is Flow Conformist that
describes a driver that intends to go according with the flow and is not characterized by
none of the previous profiles. It models only driversʼ behavior and it means they focus
on cars.
Above mentioned previous studies have solid theoretical background to model the
transportation network or the transportation systems. Our approach is similar to the
microscopic simulation. Especially, we utilize the cellular automata to model the spatial
layout and agents to denote the elements of the transportation system. However, our
approach is differentiated by utilizing cellular automata and agents to model the multi-
modal transit station. In our approach we focus to the spatial layout of the multi-transit
station and we model the vehicles and the passengers as agents. Also, we model the
boarding in the transit station by modeling the interaction among the vehicles and the
passengers.

2.2 Discrete Event System Formalism

The DEVS formalism is a set-theoretic formalism developed for specifying discrete
event systems [7–9]. In the DEVS formalism, one can specify basic models and how
these models are coupled in a hierarchical and modular fashion. A basic model, called
an atomic model, specifies the dynamics of a model and is defined as Atomic Model and
Coupled Model.
An Atomic Model is defined as follow:

where
X: a set of external input event types,
Y : an output set,
S: a sequential state set,
δext : Q × X → S, an external transition function
where Q is the total state set of
M = {(s, e)|s ∈ S and 0 ≤ e ≤ ta(s)},
δint : S → S, an internal transition function,
λ: S → Y , an output function,
ta: S → R+ 0,∞ , a time advance function, where the R0,∞
+

is the non-negative real numbers with ∞ adjoined.

An atomic model AM is a model which is affected by external input events X and
which in turn generates output events Y. The state set S represents the unique description
of the model. The internal transition function and the external transition function
compute the next state of the model. If an external event arrives at the elapsed time e
which is less than or equal to ta(s) specified by the time advance function ta, a new state
sʹ is computed by the external transition function . Then, a new ta(sʹ) is computed,
and the elapsed time e is set to zero. Otherwise, an internal event arrives at ta(s), and
then a new state sʹ is computed by the internal transition function . In the case of
 Multi-modal Transit Station Planning Method Using DEVS Formalism 53

internal events, the output specified by the output function λ is produced based on the
state s, which means the output function is processed before the internal transition func‐
tion. Then, as before, a new ta(sʹ) is computed, and the elapsed time e is set to zero.
A Coupled Model is defined as follow:

where
D: a set of component names,
For each i in D,
Mi : a component basic model
(an atomic or coupled model),
Ii : a set of influences of i,
and for each j in Ii ,
Zi,j : Yi → Xj , an i-to-j output translation,
SELECT : 2M − φ → M , a tie-breaking selector.

A coupled model CM consists of components , which are atomic models

and/or coupled models. The influences and i-to-j output translations define the
coupling specification as follows: An external input coupling (EIC) connects the input
event of the coupled model to the input event of one of its components; An external
output coupling (EOC) connects the output event of a component to the output event of
the coupled model; An internal coupling (IC) connects the output event of a component
to the input event of another component. The SELECT function is used to order the
processing of simultaneous internal events for sequential simulation. Thus, all the events
with the same time in a system can be ordered by this function.
Also, the DEVS formalism has various implementations that supports the discrete event
simulation. One of the popular implementation for DEVS formalism is the DEVSim++,
which is implemented base on C++ programming language. The DEVSim++ coordi‐
nates the event schedules of atomic models in a system and provides classes and applica‐
tion programming interface for simulation. Since the DEVSim++ fully utilizes the C++
features, it provides the advantages of object-oriented framework, such as encapsulation,
inheritance, and re-usability and it is utilized to implement various simulators in various
domains [10–12].

3 Modeling Method for Multi-modal Transit Station

In this section we introduce modeling method for multi-modal transit station. A multi-
modal transit station is complex system consist of different types of area. For example,
a modeler should consider the transportation region and passenger region. Following
Fig. 5 shows compositions of general multi-modal transit station. As shown in the
Fig. 5, when a group of vehicles arrives to the station it is categorized by transportation
type and it flows in to the waiting lane. After that, it waits on the waiting lane until the
vehicle picks up a passenger, then it leaves the transit station through the exit roads. On
the other hand, when a passenger arrives to the multi-modal transit station, a passenger
54 J. Choi et al.

decides the transportation method and move to the right platform. Therefore, a passenger
uses the crosswalk to change the platform and a vehicle should wait for the passenger
to cross the road (Fig. 1).

Fig. 1. Representation of general multi-model transit station

In our approach, we model spatial layout of the multi-modal transit station as an

array using a formal representation. Then we model passenger and vehicle as an agent
interacting inside of the multi-modal transit station model. Then, we model the procedure
of taking taxi as interaction between passenger and vehicle models. In order to model
each cell, we utilized the DEVS formalism.

3.1 Modeling Spatial Layout of the Station

A spatial layout of multi-modal transit station is one of the reason that causes traffic
congestion, so it is one of the important consideration for designing multi-modal transit
station. In order to model a spatial layout of the station, we divide a region into cells and
each cell represents the characteristics of the region. For instance, a single lane which
can hold 10 vehicles can be modeled as 10 cells. Then, each cell represents one vehicle
and a modeler can model the behavior of the single lane by array of cells that models
the characteristics of the single lane. Modeling a spatial layout of the multi-modal transit
station using cells have two merits. First, a modeler can decide the resolution of the
simulation by choosing the number of cells. When a modeler uses massive cells to model
the transit station, the model can simulate the movements of passengers and vehicles in
micro scale with high computation power. On the other hand, when a modeler wants to
see the result quickly with the proper resolution, the modeler can reduce the number of
cells. Therefore, a modeler first select a region in the multi-modal transit station, and
 Multi-modal Transit Station Planning Method Using DEVS Formalism 55

selects the proper simulation resolution with respect to the simulation objective. Finally,
each sub-region is mapped to a cell to model the region. Following Fig. 2 shows the
representation of a road using cells. As shown in the Fig. 2, if a modeler wants to model
a road, the modeler should decide the resolution of the simulation. In this case, five cells
represent a sub-region of the given road with 10 m.

Fig. 2. Cell representation of a road

To model the region by dividing smaller region to model the behaviors of the sub-
region, we adopted the cellular automata theory [13]. Therefore, we can model the road
for a vehicle and a passenger by cells that interact with each other.

3.2 Modeling Behavior of Agent

In the real world, a passenger and a vehicle in the multi-modal transit station show certain
tendency to move from one place to other place. They are under controlled by rules
which are given from the society. For example, if we want to model a multi-modal transit
station with taxi, a passenger and a vehicle should follow following rules: “A taxi cannot
overtake the preceding taxi in the single lane.” and “When there is a passenger waiting
in front line, a passenger from backside of the line should not take a taxi”. Such rules
are the propositional statements that the simulation models of the multi-modal transit
station must be followed. A proposition is a declarative sentence that is either true or
false, but not both [14]. Also, we can develop a proposition by connecting various prop‐
ositions using operators, such as conjunction, disjunction, and negation. As a result, we
can develop a complex proposition by combining propositions to give requirement
specification to a given system. Such propositions should always be true during the
simulations. As a result, an urban planner can use propositional logic to check the viola‐
tion of the requirement specifications by monitoring the events during the simulation,
since some events sequence may lead the system into violation status, i.e., some circum‐
stance that the given propositions are not true.
56 J. Choi et al.

Also, the behaviors of the vehicle and the passenger can be captured by automata-
theoretic approach with respect to the propositional logic [15]. Following Fig. 3 shows
the simple single lane road system.

Fig. 3. Single lane road model

The number of cells in the figure denotes the length of the road. There are two cells
in this model, so the length of the transportation system is two. Each cell can have two
states: filled or empty. Each state in the automata shows the circumstance of the trans‐
portation system, for example, state denotes there is one agent is in the cell A, and
A|B denotes there are two agents in the system. Also, four events are defined: ϵ, In,
and Each event denotes, no events are occurred, an agent has arrived to
the system, an agent have departed from cell A, and an agent have departed from cell
B, respectively.
Since we can consider the passenger line as a queue, we can use automata to represent
the behavior of the passengers. Also, since there are two lanes, one for the taxis and
other for the passengers, we should cross product two automata to represent the
combined behavior between taxis and passengers. Figure 4 shows the cross product of

Fig. 4. Cross product of two single lane transportation system

 Multi-modal Transit Station Planning Method Using DEVS Formalism 57

the two single lane transportation system. As shown in the figure, all transitions are the
valid behaviors in the multi-modal transit station. Therefore, a user can detect the viola‐
tion of the requirement specifications by monitoring the transitions of the transportation
system.

3.3 Modeling Interaction Between Vehicle and Human

To model the multi-modal transit station, an urban planner should consider the vehicle
road and the human road. Each road may contain vehicle agents and passenger agents,
respectively. Since we are modeling the multi-modal transit station, each passenger
interacts with vehicle. When a vehicle can ride a passenger, a passenger will disappear
from the human road.
Figure 5 shows a boarding in the multi-modal transit station. In the multi-modal
transit station, boarding process shown in Fig. 5 is occurred multiple times in parallel.
Since both of the roads are the queuing model, simulation model of the vehicle and the
passenger can be considered agents that wait in the line. The behavior of each roads is
similar, each agent in the road should follow the lane and it cannot overtake the agent
in the front. In addition, we can model the boarding process by matching proper pair of
the vehicle and the passenger. Figure 6 shows the valid transitions of the automata for
a multi-modal transit station considering the matching between a vehicle and a
passenger.

Fig. 5. Boarding process in a multi-modal transit station

58 J. Choi et al.

Fig. 6. Automata of the multi-modal transit station considering matching

4 Simulation Environment of the Multi-modal Transit Station

Building a simulation model for multi-modal transit station is not an easy task, especially
to the urban planners. In this section we introduce the simulation environment of the
multi-modal transit station. First, we explain the key simulation model that models the
spatial layout of the multi-modal transit station. Then, we introduce the verification
method of the simulation environment.

4.1 Simulation Models for Multi-modal Transit Station

As we aforementioned, we control the resolution of the simulation using cellular
automata based on the DEVS formalism. To model the spatial layout of the station, we
utilize the different types of cellular automata to represent road for vehicles and passen‐
gers. Each road is differentiated by type of the agents, and different type of the agent
cannot enter the road. Each road has multiple aligned cells and when an agent enters the

Fig. 7. Layout of the road models in DEVS formalism

 Multi-modal Transit Station Planning Method Using DEVS Formalism 59

road, it moves forward in their lane. In order to capture such semantics, we adopted one-
dimensional cellular automata using DEVS formalism. Figure 7 shows the layout of
each road model.
Each agent in the road model checks the state of the next cell and decides whether
to move or stop. After the cells agent moves, the state of the cell is changed. This infor‐
mation is delivered to previous cell, and previous cell decides whether to go or stop.
Therefore, the information propagates backward. The DEVS formalism is suitable to
adopt this backward propagation mechanism.
The Fig. 8 shows the Coupled Model and Atomic Model which each cell checks the
state of the next cell and decides whether to send agent to next cell or not. The difference
between vehicle road model and human road model is the agent which enters to the
model.

Fig. 8. Cell DEVS model in the transportation simulator

Figure 9 shows the Traffic Manager Model of the road model. The Traffic Manager
Model generates the agents to the road, and urban planner may control the traffic flow
by changing parameters in the model. The Traffic Manager Model has two model the
Poll Manager model and Agent Generator model.

Fig. 9. Traffic manager model in the road model

60 J. Choi et al.

The Agent Generator Model generates agent objects and sends it to first road cell.
Each road cell has Poll Manager Model and Transfer Manager Model as shown in
Fig. 8. When a Poll Manager Model sends state information of the cell to corresponding
Transfer Manager Model when the state is changed. If a Poll manager gets this acknowl‐
edgement message, the road cell delivers agent object to next cell or matches the taxi
and human. On the other hand, it will be blocked because other agent is existing in the
front cell.

4.2 Verification Model for Multi-modal Transit Station

To check the simulation model that satisfy the requirement specification, an urban
planner must provide rigorous requirement specification. One of the rigorous require‐
ment specification method is specifying requirements by logic. Following Fig. 10 shows
the example of the requirement specification of transit station with length two.

Fig. 10. Requirement specification of a transit station

As we introduced earlier, we have specified the valid behavior by propositional logic

and we have generated automata with respect to the propositional logic. In Fig. 10, solid
line circle denotes the valid state and dotted line circle denotes the invalid state. In order
to check the entry point to the invalid state, the simulation environment monitors the
simulation events. When a road simulation model generates events, it passes the event
to the verification model. The Fig. 11 shows the conceptual figure of the verification
model of the proposed simulation environment.
 Multi-modal Transit Station Planning Method Using DEVS Formalism 61

Fig. 11. Verification model of the proposed simulation environment

The verification model has two models, the Validity Check Model and Invalidity
Check Model. The Validity Check Model continuously monitors the event sequences
from the transit station model to check the transition model that generates valid events.
On the other hand, the Invalidity Check Model only check the invalid events that cannot
be existed through the simulation. For example, the Invalidity Check Model check the
events that lead to the abnormal states, which is matching has been occurred at the back
of the line despite empty vehicle exist at the front or customer has ridden to the vehicle
despite the customer at the front did not ride a vehicle.

5 Case Study: Taxi Station of Pohang KTX Station in South Korea

This section introduce a case study of taxi station of Pohang KTX (Korean Train
eXpress) Station in South Korea. The Pohang KTX station has opened in 2015, and
many of the passenger had suffered terrible traffic jam while using transit station. One
of the reason for traffic jam was the Taxi station, so we have built an abstracted simu‐
lation model to model the taxi station of Pohang KTX station. The main objective of the
simulator is to check the fairness of the taxi station system. In order to check the fairness
of the system, we first specify the requirement specifications for the taxi and the passen‐
gers. Following propositions are the requirements of the tax station.

Proposition 1. A taxi cannot overtake the preceding taxi in the single lane.

Proposition 2. A passenger must be served in a First Come First Serve (FCFS) fashion.
Based on the propositions, a passenger should be served as FCFS fashion and the
Taxi should leave the transit station one by one. By simulating the model, we have found
the following situations: (1) ideal case and (2) abnormal case. The ideal case, every
passenger and the taxi are matched in FCFS fashion, as shown in the left part of the
Fig. 12.
62 J. Choi et al.

Fig. 12. Taxi station without fence

In Fig. 12, the white circle denotes the passenger and the black circle denotes the
taxi. However, in the abnormal case as shown in the right part of the Fig. 12, a passenger
can be matched to a taxi, although there exists a passenger in the line. This abnormal
case happens when the passenger does not recognize the line. Since the traffic light
prevents the taxi to enter the single lane, a passenger, who arrived to the taxi station,
may think he/she is the first passenger in the line, so he/she will take a taxi although
there exists a passenger in the line. We have also analyzed the alternative solution of
the taxi station. Figure 13 shows the alternative solution, which is installing fence. In
order to prevent random matching at the taxi station, a fence can be installed to the transit
station. Fortunately, the decision maker of the Pohang KTX station has installed a fence
to prevent a passenger to take a taxi anywhere. Figure 14 shows the initial phase of the
taxi station and current taxi station of Pohang KTX station.

Fig. 13. Taxi station with fence system

Fig. 14. Pictures of the taxi station of Pohang KTX station

 Multi-modal Transit Station Planning Method Using DEVS Formalism 63

6 Conclusion

Designing multi-modal transit station is one of important consideration in urban plan‐

ning. Since urban infrastructure cannot be modified easily, a decision maker may
consider the various alternative designs before building a transit station. To assist the
urban planners, modeling and simulation of transit station is one of the most adequate
method to draw alternatives efficiently.
In this paper, we presented a design method for multi-modal transit station and
proposed a simulation environment based on the DEVS formalism. In order to capture
the behavior of the vehicles and the passenger in the transit station, we model the spatial
layout of the station as multiple cells using DEVS Formalism. Then, we implemented
the simulation model using DEVSim++. Moreover, in order to verify the requirement
specifications, we provided various method to check the requirement specification, such
as propositional logic. To verify the requirement specification, we developed the
untimed DEVS model to check the target transportation system against to the require‐
ment specification. For the further research, a multi-lane system and temporal behavior
of the traffic jam in the transportation network can be considered.

References

1. Hernndez, J., Ossowski, S., Garca-Serrano, A.: Multiagent architectures for intelligent traffic
management systems. Transp. Res. Part C Emerg. Technol. 10(56), 473–506 (2002)
2. Errampalli, M., Okushima, M., Akiyama, T.: Microsopic simulation model considering public
transport policy. J. Eastern Asia Soc. Transp. Stud. 6, 2718–2733 (2005)
3. Papageorgiou, M.: Concise encyclopedia of traffic & transportation system: simulation
program, pp. 491–502 (1991)
4. Schwerdtfeger, T.: Dynemo: a model for the simulation of traffic flow in motorway networks.
In: The 9th International Symposium on Transportation and Traffic Theory, pp. 65–87. VNU
Science Press (1984)
5. Zamith, M., et al.: A new stochastic cellular automata model for traffic flow simulation with
driversʼ behavior prediction. J. Comput. Sci. 9, 51–56 (2015)
6. OBrien, E.J., Lipari, A., Caprani, C.C.: Micro-simulation of single-lane traffic to identify
critical loading conditions for long-span bridges. Eng. Struct. 94, 137–148 (2015)
7. Zeigler, B.P.: Multifacetted Modeling and Discrete Event Simulation. Academic Press,
London (1984)
8. Zeigler, B.P.: Object-Oriented Simulation with Hierarchical, Modular Models. Academic
Press, London (1990)
9. Zeigler, B.P., Kim, T.G., Praehofer, H.: Theory of Modeling and Simulation. Academic Press,
London (2000)
10. Song, H.S., Kim, T.G.: Application of real-time DEVS to analysis of safety critical embedded
control systems: railroad-crossing control example. Simul. Trans. Soc. Model. Simul. Int.
81(2), 119–136 (2005)
11. Hong, J., Seo, K.M., Kim, T.G.: Simulation-based optimization for design parameter
exploration in hybrid system: a defense system example. Simul. Trans. Soc. Model. Simul.
Int. 89(3), 362–380 (2013)
12. Seo, K.M., et al.: DEVS-based combat modeling for engagement-level defense simulation.
Simul. Trans. Soc. Model. Simul. Int. 90(7), 759–781 (2014)
64 J. Choi et al.

13. Wolfram, S.: Theory and Applications of Cellular Automata, vol. 1. World Scientific,
Singapore (1986)
14. Rosen, K.: Discrete Mathematics and Its Applications, 7th edn. McGraw-Hill Education, New
York (2012)
15. Vardi, M.: An automata-theoretic approach to linear temporal logic. In: Moller, F., Birtwistle,
G. (eds.) Logics for Concurrency. LNCS, vol. 1043, pp. 238–266. Springer, Heidelberg
(1996)
16. Tag Gon Kim, DEVSim++ Manual. http://smslab.kaist.ac.kr
Reliability Analysis of Software Defined Wireless
Sensor Networks

Na Gong and Xin Huang(B)

International Business School in Suzhou, Department of Computer Science

and Software Engineering, Xi’an Jiaotong-Liverpool University, Suzhou, China
[email protected], [email protected]

Abstract. Software defined wireless sensor networks (SDWSN), which

separate the control and data forwarding functions, have been proposed
to solve the management dilemma of conventional wireless sensor net-
works (WSN). This technique can achieve a dynamic and centralized
network management. However, the study on its reliability is still defi-
cient. In this paper, SDWSN is formally modeled using Continuous-Time
Markov Chains (CTMCs), and its reliability is verified using probabilis-
tic model checking techniques. The verification results are expected to
give suggestions to the future SDWSN designs.

Keywords: Model checking · PRISM · Software defined network · Wire-

less sensor network

1 Introduction
Nowadays, as an important technique in the next generation wireless and mobile
networks, the wireless sensor network (WSN) is becoming more and more com-
plex and demanding [6,7,17]. It has been widely used in many areas, for exam-
ple military, navigation and environmental monitoring [15]. Typical WSNs are
self-configuring networks [10,15], and their topologies are frequently changed
[11]. This feature of WSNs poses the demand for dynamic network manage-
ment. Therefore, software-defined wireless sensor network (SDWSN), which is
a combination of WSN and software-defined networking (SDN), is proposed
[5,10,11,15,18,23].
SDN is a new paradigm of networking system that provides dynamic net-
work management and configuration. Why SDN is needed? Traditional networks
consist of many devices for forwarding and processing data packets [21]. It is
extremely time-consuming to deploy and maintain traditional networks, because
the packet routing control functions are located in routers, and they need to be
updated locally in most cases [11]. To remedy this limitation, SDN is proposed.
It separates the control plane and the data plane. The data plane composed
by physical network devices is responsible to perform packet forwarding and
processing functions. The control layer is in charge of the whole network man-
agement [22]. The applications hosted in the application layer can program the
devices in data plane [14].

c Springer Science+Business Media Singapore 2016
S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 65–78, 2016.
DOI: 10.1007/978-981-10-2158-9 6
66 N. Gong and X. Huang

Similar to SDN, SDWSN aims at separating the control plane from the
data forwarding plane in WSNs. These days, the dramatically raising applica-
tion of the wireless and sensing technology expands the requirement for the
dynamic, flexible and secure management of WSN. Hence, to satisfy these
needs, the Software-Defined Wireless Sensor Network (SDWSN) is proposed
[5,6,11,15,18,23]. Compared with the traditional WSN, SDWSN has several
advantages:
– The network becomes programmable. This allows for more dynamic network
segmentation and utilization without changes in other devices [12].
– The centralized controller maintains a global view of the whole network. Thus,
security and flexibility of SDWSN could be improved.
However, one major concern about SDWSN is its reliability. Because SDWSM
has a central controller, it may becomes the bottleneck of the whole system. If
it fails, the whole system will be down. However, so far, few researchers study
the reliability of the SDWSN which however is not a negligible issue. This paper
therefore aims to study the reliability of the SDWSN by using the probabilistic
model checking technique with the tool PRISM.
The reminder of this paper is organized as follows. In Sect. 2, SDWSN is
described. The probabilistic model checking technique is explained in Sect. 3.
Section 4 describes the modeling of the SDWSN structure. Section 5 analyzes
the results. Finally, Sect. 6 provides the conclusion.

2 Software Defined Wireless Sensor Network

The widespread application of WSN attracts much awareness from both the
industrial and academic fields. Recently, many researchers focuses on enhanc-
ing the capability of WSN reconfiguration adaption. First of all, many MAC
protocols such as T-MAC, C-MAC and snapMAC have been proposed [3,19,20].
Also, a centralized architecture is proposed for WSNs [9]. In addition, using SDN
techniques in WSNs gains attentions of many researchers [5,6,10,11,15,18,23].
A SDN is composed by the following layers (shown in Fig. 1):
– Data plane: it consists of network devices such as switches to maintain the
packet forwarding functionality [22].
– Controller plane: it is in charge of the whole network management especial
the management of the packet processing rules [21].
– Application plane: it contains various applications such as Could Computing
Service and Data Warehouse [1].
– APIs: it allows the centralized controller entity in the intermediate plane to
transfer instructions and information from or to the data and application plane
[4]. The OpenFlow protocol now is a popular one [12].
The construction of SDWSN is inspired by three important ideas in SDN,
namely, the separation of controller and data planes, standardized application
interfaces and centralized management. As presented in Fig. 2, the differences
between our SDWSN and SDN are:
Reliability Analysis of Software Defined Wireless Sensor Networks 67

Fig. 1. The architecture of SDN

Fig. 2. The architecture of SDWSN

68 N. Gong and X. Huang

– Replaces the data plane in SDN with a WSN;

– Increases the number of controllers to enhance its reliability.

3 Probabilistic Model Checking

The behaviors of many real-life systems have stochastic characteristic. To model

these systems, a formal verification technique, Probabilistic Model Checking, is
widely used [13]. It is able to evaluate the correctness, performance and depend-
ability of those systems [13]. Three procedures are involved in the model check-
ing:

– Construction of a mathematical model.

– Specify properties of the system.
– Analysis these properties.

In the second step, the property specification is usually represented using

probabilistic temporal logics. It usually uses specific operators to reason about
the probability or Boolean value of events occurrences.

3.1 Continuous-Time Markov Chain

Continuous-time Markov Chains (CTMCs) usually describe a state transition

system augmented with probabilities. Figure 3 is a typical dense model of time
with discrete states. Its transitions among states could occur at any time unit in
CTMC. These arrows present the state transferring with particular possibility
in the system. The most important feature of this model is the Markov property
that is “memorylessness”. This could be described as the following equation:

P(Xn+1 = x|X1 = x1 , X2 = x2 , . . . , Xn = xn ) = P(Xn+1 |Xn = xn )

where Xn stands for the state of the system at time n and P(Xn = xn ) is the
corresponding probability. The probability distribution of system state at time
n + 1 (the future state) depends only on the state at time n (the current state)
and does not relate with the states at time (1, n) (the past states).
A continuous-time Markov chain (CTMC) M is a four-tuple (S, s, R, L).
Assume each component of the system has n states, then S is a finite set of
states where S = {s0 , s1 , s2 , . . . , sn } and s ∈ S; s is the initial status of the
component; R is the transition rate matrix of |S| × |S|, and L is the labeling
function with atomic propositions [2]. Figure 3 is an example of CTMC:

S = {s0 , s1 , s2 , s3 , s4 };

s = s0 ;
 Reliability Analysis of Software Defined Wireless Sensor Networks 69

⎡ ⎤
0 0.5 0.2 0.3 0 0
⎢ 0 0 0 0 0.7 0 ⎥
⎢ ⎥
R=⎢
⎢0 0 0 0 0 0 ⎥;
⎥
⎣ 0 0 0 0 0 0.2 ⎦
0 0 0 0 0 0.2

L(s0 ) = {OK}, L(s1 ) = L(s3 ) = L(s4 ) = {}, L(s2 ) = {Failed}, L(s5 ) = {Waiting}.

Fig. 3. CTMC example.

Time in CTMC is generally modeled using exponential distribution [16]. For

a continuous random variable X, if its density function is given by:
 −λt
λe , if t > 0
f (t) =
0, otherwise
where λ is the rate, X is exponential with the parameter λ > 0. In this system, all
delays and repair time are assumed as X here, therefore, the property P(X ≤ t)
can be computed by
t
F (t) = P(X ≤ t) = λe−λt dx
0

and the property P(X > t) is e−λt .

3.2 Continuous Stochastic Logic

Continuous stochastic logic (CSL) is used for specifying properties of CTMCs. It
is defined by two types of syntaxes: state formula (Φ) and path formula (Ψ ). CSL
allows many reliability related properties to be specified such as the reachability
property [13]. This usually outputs a failure rate where the transaction of the
system modeled by CTMC finally shut down, or a success rate. In this project,
we mainly focus on the failure rate of the system.
70 N. Gong and X. Huang

3.3 PRISM
PRISM is a popular probabilistic model checking tool. It now mainly sup-
ports four types of probabilistic model: discrete-time Markov chains (DTMCs),
continuous-time Markov chains (CTMCs), Markov decision processes (MDPs)
and Probabilistic timed automata (PTAs) [8].
The model in PRISM is specified using the PRISM modeling language. Each
System is described as modules. Each module contains its local variables which
is used to indicate the possible states. The status transition behavior in modules
is represented in command. For example:
[] x=0 -> 0.8:(x’=0) + 0.2:(x’=1);
This command claims that when the variable x has value 0, the update state
will remain at 0 with probability 0.8 and change into 1 with probability 0.2.
Additionally, the module uses cost and reward to permit reasoning about the
quantitative measure of the system such as the expected power consumption and
expected time.
PRISM’s property specification language is based on temporal logics, namely,
PCTL, CSL, probabilistic LTL and PCTL∗ [13]. It mainly has three operators:
– P operator: this is used to reason about the probability of an event occurrence,
e.g. what is the probability of y = 6:
P=? [y = 6].
– S operator: this refers to the steady-state behavior of the model, e.g. what is
the long-run probability of the queue being more than 75 % full:
S=? [q > 0.75].
– R operator: this is used for reasoning about the expected cost of the model.
For example: what is the expected number of lost requests within 15.5 time
units
R=? [C <= 15.5].

4 Model Checking SDWSN

This section illustrates a SDWSN formal model and presents its implementation
details.

4.1 Formal Model of a Typical SDWSN

PRISM is used to analyze SDWSN reliability. To achieve this, the general SDN
architecture shown in the Fig. 1 is modified into the structure presented in the
Fig. 2. The improved SDWSN has following features:
– Application Layer: Because this paper does not consider the effects of appli-
cations, all of them are regarded as a whole with very low failure rate.
– Controller Layer: If the controller fails, the system will shut down. Hence, the
modified structure adds one controller to increase the reliability.
– WSN Layer: This consists of several sensors to collect and measure information
and other physical quantities. Similar with control plane, if all sensors fail, then
controller will shut down the system.
 Reliability Analysis of Software Defined Wireless Sensor Networks 71

4.2 Implementation
In the formal model, two modules (controller layer and WSN layer) are realized.
The detail structures of modules are as follows:
– Module 1 - Controller layer: Two controllers are c1 and c2 . The states are:
both two controllers are working (m = 2), only one is working (m = 1) and
fail (m = 0). Each controller has the same failure rate λc . The transition rate
of the controller layer whose initial status is m = 2 to the state m−1 is m×λc .
– Module 2 - WSN layer: At least two sensors are considered in this module.
Additionally, the number of sensors can be manually changed. Each sensor
has the same failure probability λa .
To verify the dependability of this model, the following failure types are
considered:
– Controller layer fails if none of controllers is working.
– WSN fails if none of sensors is working.
– System fails if the controller layer fails.
– System fails if the number of working sensors is less than the lower bound.
– System fails if the skipping cycle is greater than the upper limit. (The con-
troller can automatically skip the current operation cycle when the application
or WSN layer is abnormal, then a Max COUNT is set to limit the skipping.)

5 Results
The main results are presented in this section. The factors of influencing the
reliability of SDWSN are also analyzed.

Fig. 4. The performance of WSN layer in 24 h with different number of sensor nodes.
The y-axis is the failure probability of the WSN layer; and the x-axis is the duration
(24 h). The number of sensor nodes is represented in the right; e.g., “5 sensors” means
that the WSN layer contains 5 sensors. The failure rate of each sensor node is once a
day.
72 N. Gong and X. Huang

Fig. 5. The performance of WSN layer in 30 days with different number of sensor nodes.
The y-axis is the failure probability of the WSN layer; and the x-axis is the duration
(30 days). The number of sensor nodes is represented in the right; e.g., “5 sensors”
means that the WSN layer contains 5 sensors. The failure rate of each sensor node is
once a day.

Fig. 6. The performance of WSN layer in 24 h with different number of sensor nodes.
The y-axis is the failure probability of the WSN layer; and the x-axis is the duration
(24 h). The number of sensor nodes is represented in the right; e.g., “5 sensors” means
that the WSN layer contains 5 sensors. The failure rate of each sensor node is once per
30 days.

In Fig. 4, the failure rate of the WSN layer is shown. Its setting is:
– Failure rate of each sensor: once per day;
– Duration: 24 h;
– The number of sensors: 5, 10, 15, and 20.
As shown in Fig. 4, increasing the number of sensors can decrease the failure
probability of the entire WSN layer. However, if we set the duration to 30 days,
this strategy will be challenged. The result is shown in Fig. 5. From this figure,
we can see that the failure probability of the WSN plane is 100 % after 4 days.
 Reliability Analysis of Software Defined Wireless Sensor Networks 73

Fig. 7. The performance of WSN layer in 30 days with different number of sensor nodes.
The y-axis is the failure probability of the WSN layer; and the x-axis is the duration
(30 days). The number of sensor nodes is represented in the right; e.g., “5 sensors”
means that the WSN layer contains 5 sensors. The failure rate of each sensor node is
once per 30 days.

Fig. 8. The performance of WSN layer in 24 h with different failure rate of individual
sensor nodes. The y-axis is the failure probability of the WSN layer; and the x-axis is
the duration (24 h). The failure rate of sensor nodes is represented in the right; e.g.,
“s-1/2” means that the failure rate is once per 2 days. The number of sensor nodes is 5.

This suggests that when the failure probability of individual sensor node is high
(once per day), increasing the number sensor nodes is not a good way of enhanc-
ing the reliability of the WSN layer.
In Figs. 6 and 7, the failure rate of each sensor in the WSN plane is changed
to once per month. Compared with Figs. 4 and 5, the failure probability of the
WSN plane is significantly lower. This depicts that decreasing the failure rate of
each sensor node can decrease the failure probability of the sensor layer.
In order to study the relation between the sensor node failure rate and the
WSN layer failure rate, we did several new experiments. In Figs. 8 and 9, we can
74 N. Gong and X. Huang

Fig. 9. The performance of WSN layer in 30 days with different failure rate of individual
sensor nodes. The y-axis is the failure probability of the WSN layer; and the x-axis is
the duration (30 days). The failure rate of sensor nodes is represented in the right; e.g.,
“s-1/2” means that the failure rate is once per 2 days. The number of sensor nodes is 5.

Fig. 10. The performance of the whole system (SDWSN) in 24 h with different failure
rate of individual sensor nodes. The y-axis is the failure probability of the system; and
the x-axis is the duration (24 h). The failure rate of sensor nodes is represented in the
right; e.g., “s-1/2” means that the failure rate is once per 2 days. The number of sensor
nodes is 5. The failure rate of the controller is?

see that lower sensor node failure rate is corresponding to lower WSN failure
rate. In Figs. 10 and 11, we can see that lower sensor node failure rate can also
enhance the system (SDWSN) reliability given a fixed controller failure rate.
Except the sensor nodes, the system reliability is also related to controllers’
status. We have the following findings:
 Reliability Analysis of Software Defined Wireless Sensor Networks 75

Fig. 11. The performance of the whole system (SDWSN) in 30 days with different
failure rate of individual sensor nodes. The y-axis is the failure probability of the
system; and the x-axis is the duration (30 days). The failure rate of sensor nodes is
represented in the right; e.g., “s-1/2” means that the failure rate is once per 2 days.
The number of sensor nodes is 5. The failure rate of the controller is once per 30 days

Fig. 12. The performance of the whole system (SDWSN) in 24 h with different failure
rate of one controller. The y-axis is the failure probability of the system; and the x-axis
is the duration (24 h). The failure rate of the single controller is represented in the
right; e.g., “c-1/2” means that the failure rate is once per 2 days. The number of sensor
nodes is 5. The failure rate of the sensor node is once per 30 days.

– From Figs. 12 and 13, we can see that reducing the failure rate of each con-
troller can enhance the system reliability.
– From Figs. 12 and 14, we can see that increasing the number of controllers can
also enhance the system reliability.
76 N. Gong and X. Huang

Fig. 13. The performance of the whole system (SDWSN) in 30 days with different
failure rate of one controller. The y-axis is the failure probability of the system; and the
x-axis is the duration (30 days). The failure rate of the single controller is represented
in the right; e.g., “c-1/2” means that the failure rate is once per 2 days. The number
of sensor nodes is 5. The failure rate of the sensor node is once per 30 days.

Fig. 14. The performance of the whole system (SDWSN) in 24 h with different failure
rate of the two controllers. The y-axis is the failure probability of the system; and the
x-axis is the duration (24 h). The failure rate of the two controllers is represented in
the right; e.g., “c1c2-1/2” means that the failure rate is once per 2 days. The number
of sensor nodes is 5. The failure rate of the sensor node is once per 30 days.

6 Conclusion
In this paper, to verify the reliability of SDWSN, a probabilistic model has
been realized based on CTMC using PRISM. After experiments, the results
illustrate the impacts of controllers and sensors on the network performance.
 Reliability Analysis of Software Defined Wireless Sensor Networks 77

Firstly, increasing the number of controllers and sensors can enhance the system
reliability. Secondly, the lower failure rate of controllers and sensors makes the
network more dependable. Moreover, these two factors are more cognizable in
long term.
Nevertheless, the limitation of the assumptions inevitably results in obstacles
to find more potential correlation among components in SDWSN. For example,
the relationship between the controller and sensor is omitted and this will be
considered in the further research.

Acknowledgment. This work has been supported in part by the XJTLU RDF140243,
by the Natural Science Foundation of Jiangsu Province under Grant BK20150376, and
by the Suzhou Science and Technology Development Plan under Grant SYG201516.

References
1. Ali, S.T., Sivaraman, V., Radford, A., Jha, S.: A survey of securing networks using
software defined networking. IEEE Trans. Reliab. 64(3), 1086–1097 (2015)
2. Aziz, A., Sanwal, K., Singhal, V., Brayton, R.: Model-checking continuous-time
Markov chains. ACM Trans. Comput. Logic (TOCL) 1(1), 162–170 (2000)
3. De Mil, P., Jooris, B., Tytgat, L., Hoebeke, J., Moerman, I., Demeester, P.: snap-
Mac: a generic MAC/PHY architecture enabling flexible MAC design. Ad Hoc
Netw. 17, 37–59 (2014)
4. Ding, A.Y., Crowcroft, J., Tarkoma, S., Flinck, H.: Software defined networking
for security enhancement in wireless mobile networks. Comput. Netw. 66, 94–101
(2014)
5. Granelli, F., Gebremariam, A.A., Usman, M., Cugini, F., Stamati, V., Alitska, M.,
Chatzimisios, P.: Software defined, virtualized wireless access in future wireless
networks: scenarios and standards. IEEE Commun. Mag. 53(6), 26–34 (2015)
6. Han, Z.J., Ren, W.: A novel wireless sensor networks structure based on the SDN.
Int. J. Distrib. Sens. Netw. 2014 (2014)
7. Harrop, P., Das, R.: Wireless sensor networks 2010-2020. Networks 2010, 2020
(2010)
8. Hinton, A., Kwiatkowska, M., Norman, G., Parker, D.: PRISM: a tool for automatic
verification of probabilistic systems. In: Hermanns, H., Palsberg, J. (eds.) TACAS
2006. LNCS, vol. 3920, pp. 441–444. Springer, Heidelberg (2006)
9. Hunkeler, U., Lombriser, C., Truong, H.L., Weiss, B.: A case for centrally controlled
wireless sensor networks. Comput. Netw. 57(6), 1425–1442 (2013)
10. Jacobsson, M., Orfanidis, C.: Using software-defined networking principles for wire-
less sensor networks. In: 11th Swedish National Computer Networking Workshop
(SNCNW), 28–29 May 2015, Karlstad, Sweden (2015)
11. Jagadeesan, N.A., Krishnamachari, B.: Software-defined networking paradigms in
wireless networks: a survey. ACM Comput. Surv. (CSUR) 47(2), 27 (2014)
12. Kirkpatrick, K.: Software-defined networking. Commun. ACM 56(9), 16–19 (2013)
13. Kwiatkowska, M., Norman, G., Parker, D.: Probabilistic model checking in prac-
tice: case studies with PRISM. ACM SIGMETRICS Perform. Eval. Rev. 32(4),
16–21 (2005)
14. Lin, Y.-D., Lin, P.-C., Yeh, C.-H., Wang, Y.-C., Lai, Y.-C.: An extended SDN
architecture for network function virtualization with a case study on intrusion
prevention. IEEE Netw. 29(3), 48–53 (2015)
78 N. Gong and X. Huang

15. Luo, T., Tan, H.-P., Quek, T.Q.S.: Sensor openflow: enabling software-defined wire-
less sensor networks. IEEE Commun. Lett. 16(11), 1896–1899 (2012)
16. Menshikov, M., Petritis, D.: Explosion, implosion, and moments of passage times
for continuous-time Markov chains: a semimartingale approach. Stoch. Process.
Appl. 124(7), 2388–2414 (2014)
17. Oualhaj, O.A., Kobbane, A., Sabir, E., Ben-othman, J., Erradi, M.: A ferry-assisted
solution for forwarding function in wireless sensor networks. Pervasive Mob. Com-
put. (2015)
18. Xu, R., Huang, X., Zhang, J., Lu, Y., Wu, G., Yan, Z.: Software defined intelligent
building. Int. J. Inf. Secur. Priv. (2016)
19. Steiner, R.V., Mück, T.R., Fröhlich, A.A.: C-MAC: a configurable medium access
control protocol for sensor networks. In: 2010 IEEE Sensors, pp. 845–848. IEEE
(2010)
20. Van Dam, T., Langendoen, K.: An adaptive energy-efficient MAC protocol for
wireless sensor networks. In: Proceedings of the 1st International Conference on
Embedded Networked Sensor Systems, pp. 171–180. ACM (2003)
21. Wood, T., Ramakrishnan, K.K., Hwang, J., Liu, G., Zhang, W.: Toward a software-
based network: integrating software defined networking and network function vir-
tualization. IEEE Netw. 29(3), 36–41 (2015)
22. Xie, J., Guo, D., Zhiyao, H., Ting, Q., Lv, P.: Control plane of software defined
networks: a survey. Comput. Commun. 67, 1–10 (2015)
23. Lu, Y., Huang, X., Huang, B., Xu, W., Zhang, Q., Xu, R., Liu, D.: A study on
the reliability of software defined wireless sensor network. In: 2015 IEEE Interna-
tional Conference on Smart City (IEEE Smart City 2015), Chengdu, China, 19–21
December 2015
 A Path-Line-Based Modeling for Fishing
Prediction

Koji Koyamada1(&), Katsumi Konishi2, Naohisa Sakamoto1,

and Marohito Takami1
1
Academic Center for Computing and Media Studies,
Kyoto University, Kyoto, Japan
[email protected]
2
Department of Computer Science,
Kogakuin University, Tokyo, Japan

Abstract. The development of a predictive model for fishing grounds is

required due to climate change and global warming to maintain steady fish
catches. In general, the fish catches are modeled using environment variables in
the time-varying ocean simulation dataset. The current modeling techniques
consider ocean simulation dataset only on the fishing day, which is too simple to
explore a causal relationship between the catches and environment variables. To
solve this problem, we use environmental variables, such as water temperature,
velocity and salinity, sampled on flow path-lines, which terminate at fishing
locations, and we employ a standard least squares method with the L0 norm
regularization technique to construct an accurate fishing ground model. We
apply the approach to construct a predictive model of neon flying squid to
confirm the effectiveness by demonstrating the model accuracy distribution
along the path-lines.

Keywords: Fishing ground modeling
L0 norm regularization
Visualization

1 Introductions

Global warming is causing climate change, which increases the variance in environ-
mental variables, such as temperature, salinity and velocity. Climate change results in a
change of locations of the habitats of fish because the habitats of fish are always looking
for a favorable environment. For fishery grounds modeling, such an environment has
been represented as a function of environmental variables. If the number of the habitats
and the environmental variables can be measured, we can construct a predictive model
for the population.
There is a strong need for the development of a predictive model for fishing to
maintain steady fish catches. Fishing is under greater risk due to global warming
because the fisherman’s tacit knowledge may be useless. To avoid wasting fuel oil in
fishing, the fishermen understand that they should have a fishery prediction technique.
The fishermen asked the academic community to develop a highly accurate predictive
model for their fishery. They recorded fish catches and made the results open to the
community.

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 79–88, 2016.
DOI: 10.1007/978-981-10-2158-9_7
80 K. Koyamada et al.

Recently, fishermen have recorded their fish catches when they are on their fish
boats for future reference. The catches are represented using a unit named the catch per
unit effort (CPUE). In fisheries, the CPUE is an indirect measure of the abundance of a
target species. Changes in the catch per unit effort are inferred to signify changes to the
true abundance of the target species. The CPUE is recorded with the location at which
the fishing is conducted. The location includes the latitude and longitude of the fishing
boat but not the depth at which the fish are caught.
In fishery ground prediction, the CPUE is generally represented as a function of
environment variables. It is difficult to measure sufficient variable data. High perfor-
mance computing facilities enable us to calculate a highly accurate environmental
model that includes the temperature, salinity and velocity. With down-scaling and data
assimilation techniques, we can obtain highly accurate environmental variables with
sufficient resolution around possible fishing grounds.
Current modeling techniques consider an ocean simulation dataset only on the
fishing day. With such a technique, which does not use time-varying datasets, it is
difficult to explore the causal relationship between the CPUE and environment vari-
ables, even if their fitness has been maximized. The food chain is important when we
estimate fishery prediction. It originates from nutrient salts that precipitate at the bottom
of the sea. When the salts move upwards where the sunlight arrives along an upwelling
current, phytoplankton can reproduce. Then, zooplankton and fish eat the phyto-
plankton and the zooplankton, respectively. It is natural that we assume that these food
chain components follow the sea water current. The phytoplankton and the zooplankton
follow the ocean current passively in their moving.
In our research, we employ a standard least squares method with the L0 norm
regularization technique [3] to sparsely model the causal relationship because we have
a large number of possible explanatory variables considering a time-varying ocean
simulation dataset. The technique is a type of signal processing technique for efficiently
acquiring and reconstructing a signal by finding solutions to underdetermined linear
systems based on the principle that, through optimization, the sparsity of a signal can
be exploited to recover it from far fewer samples than required by the Shannon-Nyquist
sampling theorem.
The remainder of the paper is organized as the follows. In Sect. 2, we describe the
related work on fishery grounds predictive modeling. In Sect. 3, we explain our visual
analytics system, which supports the discovery of the food chains along a path-line in
the ocean flow. In Sect. 4, we describe the L0 norm regularization technique for sparse
modeling. In Sect. 5, we apply the technique to construct a predictive fishery model for
neon-flying fishes to confirm the effectiveness by comparing the proposed technique
with the previous one.

2 Related Work

Tian et al. developed two types of HSI models for predicting neon squid fish catches
and compared the models [1]. The HSI model represents an index that indicates
whether an animal feels comfortable by using a value from 0.0 to 1.0. The index is
defined by the United States Environmental Preservation Agency (EPA). In this HSI
 A Path-Line-Based Modeling for Fishing Prediction 81

model, a Suitability Index (SI) is first constructed so that the CPUE can be described
using a single environmental variable, such as temperature, salinity or velocity. We call
this method Tian’s method. For example, when we consider a sea animal, we select one
of the temperature, salinity and velocity variables and construct a function of the CPUE
with the selected variable. To develop an HSI model, we integrate several SI indexes
into a single HSI model using a mathematical method, such as the geometric mean.
Xinjun et al. proposed an HSI model in which the CPUE is explained using a set of
weighted summations of the environmental variables and evaluated the effect caused by
the weighting methods [2]. We call this method Xinjun’s method. These previous
models considered only the correlation between the CPUE and the environmental
variables but did not consider any causal relationship in the food chain.
Koyamada et al. proposed a modeling technique that employs environmental
variables, such as water temperature, velocity and salinity, sampled on flow path-lines
that terminate at fishing locations [6]. Although the results indicated that the accuracy
of the model was improved when it considered the variables sampled along the
path-lines, they did not determine how many sampling points along the path-line are
required for effective modeling.

3 Visual Analytics System for a Path-Line-Based Sampling

For the exploration of the causal relationship in the food chain, we focus on a path-line
that represents a curved line whose tangent is parallel to a velocity vector at a
spatio-temporal point. In this research, we assume that the food chain components,
such as plankton and fish, move along the ocean current. Although the number of the
target fish that the fishermen try to catch may eat small fishes can be measured by the
CPUE, it is difficult to measure those of the other components. It is naturally inferred
that each can have a set of environmental variables and specific value ranges.
Along a path-line that passes through a fishing location, we sample environmental
variables, such as the water temperature, velocity and salinity. We regard these vari-
ables as the factors that can be regarded as the occurrences of the food chain com-
ponents. In other words, these variables can be explained by the latent variables of the
occurrences.

Fig. 1. Fishing ground points and the corresponding path-lines. (Color figure online)
82 K. Koyamada et al.

We developed a visual analytic system to confirm that a path-line starts at Oyashio

region, which can provide nutrient salts, and terminates at a CPUE point. Figure 1
shows the CPUE points and the corresponding path-lines that pass through these
points. The left image shows a visualization result of the fishing day. The second, third
and fourth images show the results of the previous ten, thirty, and sixty days,
respectively. The blue and orange regions represent Oyashio and Kuroshio regions,
respectively. If the food chain is conducted along the path-line, the expert can divide it
into multiple segments that are related to food chain components. Because the first
component is phytoplankton, which often reproduces by the upwelling current, it is
preferable to identify a location where the reproduction starts to occur. We asked
domain experts to use and to evaluate the visual analytic system of whether the food
chain can be identified along the path-line. The domain experts agreed to its effec-
tiveness and suggested that the existence of vortex flow may be related to the formation
of the food chain.

4 Method

We develop a fishing ground model using an ocean simulation dataset and a fish catch
dataset called CPUE. The simulation dataset is generated from a data assimilation
system for the Northwest Pacific Ocean. The values of the environmental variables
were obtained from an ocean simulation based on the 3D-VAR data assimilation
product MOVE (MRI Multivariate Ocean Variational Estimation). The simulation
generated values of temperature (T), salinity (S), and current velocity (U, V, W). Thus,
the number of variable types is five. Because the grid size is 0.1° between 43° N and

Fig. 2. Path-lines that terminate at a single CPUE point with varying depth values. (Color figure
online)
 A Path-Line-Based Modeling for Fishing Prediction 83

35° S and 141° W and 147° E in the North-West Pacific Ocean, the grid resolution is
90  70 in the horizontal plane. We calculate the path-lines whose terminating points
are located at the CPUE points. In the path-line, environmental variables, such as water
temperature, velocity and salinity, are sampled [5]. Figure 2 shows the path-lines (left)
and the sampled variables on the lines (right). In the lower right of the figure, several
path-lines for a single CPUE point are displayed with the depth values changed.
In the fishing ground model, we assume that the fish catch is explained using
environmental variables sampled along path-lines that are terminated at the CPUE
points. We regard the depths of the CPUE points as unknown variables because there is
no information on the depths where the fish were caught. Thus, we have 16,200
(¼ 5  54  60) variables at each CPUE point because the depth and time resolutions
are 54 and 60, respectively. In our research, we consider two methods, Tian’s and
Xinjun’s methods, to confirm the effectiveness in which time series data sets are
employed for constructing the HSI model.

4.1 Xinjun’s method

In Xinjun’s method, we assume that the log CPUE can be described as a type of
generalized linear model as follows:

X
60 X
54  
logCPUE ¼ hTk;j logTk;j þ hSk;j logSk;j þ hUk;j logUk;j þ hVk;j logVk;j þ hWk;j logWk;j ;
k¼1 j¼1

ð1Þ

where Tk;j , Sk;j , Uk;j , V and Wk;j denote the temperature, salinity and velocity compo-

nents of sea water at depth j and time step k, respectively, and hk;j ¼ hTk;j hSk;j
hUk;j hVk;j hWk;j T are design parameters. To simplify the problem, the value of log CPUE
is roughly quantized to be in f1; 2g, that is, in (1), we replace the value with
qa ðlog CPUEÞ, where qa is a quantization function defined by

1 if x\a
qa ðxÞ ¼ :
2 if x [ a

This paper proposes the following HSI model,

60 X
X 54  
qa ðlogCPUE Þ ¼ hTk;j logTk;j þ hSk;j logSk;j þ hUk;j logUk;j þ hVk;j logVk;j þ hWk;j logWk;j
k¼1 j¼1

ð2Þ

In the numerical experiments of the next section, we use the average of log CPUE as
the threshold a.
Next, we focus on a scheme to estimate the parameters hk;j . Let us consider N
ðiÞ ðiÞ ðiÞ ðiÞ ðiÞ
CPUE points CPUEðiÞ and their environmental variables Tk;j , Sk;j ,Vk;j ,Uk;j and Wk;j for
84 K. Koyamada et al.

i ¼ 1; 2; . . .; N. The parameters can be obtained using a standard least squares method,

and we use L0 norm regularization to avoid overfitting and propose the following
problem,

minimize ky  Ahk2 þ kkhk0 ; ð3Þ

where k
k0 denotes the L0 norm of a vector, that is, the number of non-zero entries of a
vector, y 2 RN is a constant vector whose i th entries are CPUE ðiÞ , A 2 RN16200 is a
ðiÞ ðiÞ ðiÞ ðiÞ ðiÞ
constant matrix consisting of log Tk;j ; log Sk;j ; log Vk;j ; log Uk;j and log Wk;j as follows,
h i
ðiÞ ðiÞ ðiÞ ðiÞ ðiÞ
aik;j ¼ log Tk;j log Sk;j log Vk;j log Uk;j log Wk;j ;
2 3
a11;1 a11;2 ... a11;54 a12;1 . . . a160;54
6 a2 a12 ... a21;54 a22;1 . . . a260;54 7
6 1;2 7
A¼6 .. 7
1;2
6 .. .. .. .. 7;
4 . . . . . 5
aN1;2 aN1;2 ... aN1;54 aN2;1 . . . aN60;54

and h 2 R16200 is a design variable vector. To solve this problem, we apply the iterative
reweighted least squares (IRLS) algorithm [4]. In IRLS, the solution at the ðt þ 1Þ th
iteration is obtained as
 2
hðt þ 1Þ ¼ arg minky  Ahk2 þ kW ðtÞ h ; ð4Þ
h

ðtÞ
where W ðtÞ is a diagonal matrix whose diagonal elements Wii are calculated using the
optimal solution of the previous iteration as follows,

ðtÞ 1
Wii ¼   ; ð5Þ
ðkÞ
 hi  þ e

and e is a small constant. The empirical results show that we can obtain the optimal
solution of (3) by iterating (4) and (5) until convergence.
IRLS algorithm for (3)
Step 1. Set W ð0Þ to be the identity matrix for i ¼ 1; 2;


; 16200:
ii
Step 2. Set t 1:  
Step 3. Repeat Step 3-1 to Step 3-3 until  
hðt þ 1Þ  hðtÞ \tol
Step 3-1. Calculate hðt þ 1Þ using (4).
Step 3-2. ðt þ 1Þ
Calculate Wii for i ¼ 1; 2;


; 16200 using (5).
Step 3-3. t tþ1
 A Path-Line-Based Modeling for Fishing Prediction 85

4.2 Tian’s Method

In Tian’s method, we assume that the log CPUE, that is the SI model, can be described
as a type of generalized linear model for each variable as follows

60 X
X 54
SIX ¼ logCPUE ¼ hXk;j logXk;j ; ð6Þ
k¼1 j¼1

where X denotes S, T, V, U or W. Then, using the quantization function defined in the

previous section, this method defines the following HSI model:

SIS þ SIT þ SIV þ SIU þ SIW
HSI ¼ qa : ð7Þ
5

Next, we focus on a scheme to estimate the parameters hXk;j . Let us consider N CPUE
ðiÞ
points CPUEðiÞ and their environmental variables Xk;j for i ¼ 1; 2; . . .; N, where X
denotes S, T, V, U or W. The parameters can be obtained using Eq. 3, in which
ðiÞ
A 2 RN3240 is a constant matrix consisting of zik;j ¼ log Xk;j as follows:
2 ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ 3
z1;1 z1;2 ... z1;54 z2;1 ... z60;54
6 ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ 7
6 z1;1 z1;2 ... z1;54 z2;1 ... z60;54 7
A¼6
6 . .. .. .. .. 7
7
4 .. . . . . 5
ðNÞ ðNÞ ðNÞ ðNÞ ðNÞ
z1;1 z1;2 ... z1;54 z2;1 ... z60;54

and h 2 R3240 is a design variable vector. To solve this problem, we apply the IRLS
algorithm described in the previous section for each variable S, T, V, U or W.

5 Experimental Results and Discussion

In this research, an important point is how sparsely the CPUE can be modeled, that is,
how much the number of variables is minimized. To examine the method proposed in
the previous section, we perform some numerical experiments. In all experiments, we
use e ¼ 108 and tol ¼ 105 . Because the optimal solution of (3) depends on the value
of k, we increase it from 0.01 to 1 by 0.1 and select the best k.
To show the efficiency of the proposed method, we compare the proposed method
with the previous method, in which a time-varying dataset is not employed. In Xinjun’s
method, the CPUE can be described by leaving terms in which k equals 1 in Eq. 6.

X
54  
logCPUE ¼ hTi logT1;i þ hSi logS1;i þ hTi logU1;i þ hTi logV1;i þ hTi logW1;i ;
j¼1
86 K. Koyamada et al.

In Tian’s method, we assume that the log CPUE, that is, the SI model, can be
described in the same way.

X
54
SIX ¼ logCPUE ¼ hXk;j logXk;j
j¼1

where X denotes S, T, V, U or W.
Here, the parameters are estimated by a least squares method. We examine these
methods using fish cash data of January 2006. The parameters are estimated using 45
CPUE points of January 1 to 10, and we examine the accuracy rate of the estimated 137
CPUE points from January 11 to 31. Because the values of logCPUE should be 1 or 2,
we round them into 1 or 2 and calculate the percentage of correct answers.
Table 1 shows the results that indicate that the accuracy of the model is improved
when it considers the variables that are sampled along the path-lines.
The purpose of this research is to demonstrate the advantage that the environmental
variables sampled along path-lines can explain the CPUE points that are measured at
the terminating points. Numerical experiments indicate that the value of CPUE depends
highly on the velocity of water. We showed the effectiveness by comparing the pro-
posed model with the previous one, which does not consider the path-line sampling and
the efficiency of the L0 norm regularization to select important environmental variables.

Table 1. Percentage of correct answers.

Proposed method with best k 75.91 %
Proposed method with k ¼ 0 49.64 %
Previous method using Eq. 6 54.95 %

Fig. 3. Model accuracy distribution along the number of days. (Color figure online)
 A Path-Line-Based Modeling for Fishing Prediction 87

By considering a food-chain in the fishery, the model accuracy can be affected by

the 60-day data fields at the maximum. To estimate the change of the accuracy of the
models that are constructed using the 0-, 10-, 20-, 30-, 40-, 50- and 60-day data fields,
we constructed the related prediction models. Because we employ five variables, which
are the temperature, salinity, velocity vectors (U, V, W), we had five SI models, which
were evaluated with respect to the accuracy. The HSI model is constructed by taking
the geometrical average of the SI models. Figure 3 shows that the model accuracy
cannot be improved by extending the number of days for the model construction. In
many cases, the accuracy is maximized when 30-day data fields are used, which
suggests that the data fields are not effective for more than 30 days.

6 Conclusion

To construct an accurate fishing ground model, we employed environmental variables,

such as water temperature, velocity and salinity, sampled on flow path-lines that ter-
minate at fishing locations and employ a standard least squares method with the L0
norm regularization technique. By comparing the developed model with a previous
method that employs environmental variables only on the fishing day, we showed that
the accuracy of the model is improved when it considers variables sampled along the
path-lines. In addition, we calculate the model accuracy by changing the number of
days in which the path-lines are traced back and understand that more than 30 days may
not improve the model accuracy.
The food chain is governed by a physical law in which the sea current transports the
environmental resources from the upstream regions to the downstream regions. In our
model, we consider the treatment that reflects the physical law. In this research, we
assume that the food chain is governed only by the sea current. In the future, we will
consider physical transportation caused by the gradient vector flow of a scalar variable,
such as salinity. We will also consider a vortex flow to explain a fish catch as the
domain experts suggested after they experienced our visual analytics system. One piece
of evidence shows the formation of twin vortices located near the fishing grounds [5].
The flow direction indicates that one of the two vortex centers attracts flow from the
bottom of the ocean, which suggests that upwelling flow occurs near the vortex center
and the upwelling may facilitate the food chain.

References
1. Tian, S., Chen, X., Chen, Y., Xu, L., Dai, X.: Evaluating habitat suitability indices derived
from CPUE and fishing effort data for Ommatrephes bratramii in the northwestern Pacific
Ocean. Fish. Res. 95(2–3), 181–188 (2009)
2. Xinjun, C., Siquan, T., Bilin, L., Yong, C.: Modeling a habitat suitability index for the eastern
fall cohort of Ommastrephes bartramii in the central North Pacific Ocean. Fish. Res. 29(3),
493–504 (2011)
3. Daubechies, I., Devore, R., Fornasier, M., Gntrk, C.S.: Iteratively reweighted least squares
minimization for sparse recovery. Comm. Pure Appl. Math 63(1), 1–38 (2010)
88 K. Koyamada et al.

4. Futami, K., Onoue, Y., Sakamoto, N., Koyamada, K.: Visualization of ocean current effects
for fishing ground formation. J. Jpn Soc. Fluid Mech. 34(2), 83–86 (2015)
5. Uenaka, T., Sakamoto, N., Koyamada, K.: Visual analysis of habitat suitability index model
for predicting the locations of fishing grounds. In: Proceedings of IEEE Pacific Visualization
2014 (Visualization Notes), pp. 306–310 (2014)
6. Koyamada, K., Konishi, K., Sakamoto, N., Takami, M.: A path-line-based approach for
developing a fishing ground model. In: Proceedings of AsiaSim 2015 (2015)
Simulation and Analysis
 Numerical Simulation
of a Hamilton-Jacobi-Bellman Equation
for Optimal Management Strategy of Released
Plecoglossus Altivelis in River Systems

Yuta Yaegashi1(&), Hidekazu Yoshioka2, Koichi Unami1,

and Masayuki Fujihara1
1
Graduate School of Agriculture, Kyoto University, Kyoto, Japan
[email protected],
[email protected],
[email protected]
2
Faculty of Life and Environmental Science,
Shimane University, Matsue, Japan
[email protected]

Abstract. A stochastic differential equation model for population dynamics of

released Plecoglossus altivelis (Ayu) in a river system subject to feeding damage
by Phalacrocorax carbo (Great Cormorant) and fishing activity by human is
proposed. A stochastic optimal control problem to maximize the sum of the cost
of countermeasure to prevent the feeding damage and the benefit of harvesting
the fish is formulated, which ultimately reduces to solving a Hamilton-Jacobi-
Bellman equation. Application of a Petrov-Galerkin finite element scheme to the
equation successfully computes the optimal management strategies for the
population dynamics of P. altivelis in a real river system and ecological and
economical indices to verify them.

Keywords: Inland fishery
Plecoglossus altivelis (Ayu)
Phalacrocorax

carbo (Great Cormorant)
Population dynamics
Feeding damage
Stochastic
optimal control
Hamilton-Jacobi-Bellman equation

1 Introduction

Plecoglossus altivelis (Ayu) is annual and diadromous fish species in Japan (Photo 1:
left panel) [1]. P. altivelis accounts for 7.5 % (2,353 t) of total fish catch (31,264 t) of
Japanese inland fisheries in 2013 [2]. Life history of P. altivelis is unique as reviewed
in Tanaka et al. [3]. During autumn, adult fishes spawn eggs in the downstream reaches
of their living river and die soon afterwards. Hatched larval fishes descend to coastal
areas of the downstream water body of the river, which is a sea or a brackish lake in
general, and grow up to juveniles with feeding on zooplanktons till the next spring. The
grown fishes ascend the river toward its midstream and upstream reach where
rock-attached algae, which are staple foods of P. altivelis, are available in riverbed.
They feed on the algae to mature till the coming autumn when they descend the river.

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 91–101, 2016.
DOI: 10.1007/978-981-10-2158-9_8
92 Y. Yaegashi et al.

Photo. 1. Adult P. altivelis (left panel) and adult P. carbo (right panel). (The photo of P. carbo
is taken from Photo by (c)Tomo.Yun http://www.yunphoto.net).

Literatures report that P. altivelis has significant influences on diversity of species in

river ecosystems, such as aquatic insects, algae, and other fish species [4, 5]. Appro-
priately managing population of P. altivelis is therefore a key topic from both eco-
logical and economical point of views.
Fish catch of P. altivelis in Japan has recently been rapidly decreasing, which is
considered mainly due to decrease of its population [1]. A cause of the population
decrease would include climate changes and manipulation of river environments such
as installations of physical barriers like dams and weirs. Another not negligible cause is
feeding damage by Phalacrocorax carbo (Great Cormorant), which is a piscivorous
bird species that reigns as a top predator of river ecological systems in Japan (Photo. 2:
right panel) [6]. Each individual of adult P. carbo. eats on average 0.5 (kg) of fish per
day and has high ability to swim and to capture the preys in rivers. P. carbo lives in
groups and creates colonies for nesting and spawning in riparian forest where there are
less signs of human life. Total number of individuals of P. carbo in Japan was about
three thousands during 1970 due to decline and decrease of habitats but has been
recently increasing. Currently, total number of individuals of the bird in Japan is
estimated to exceed more several tens of thousands [7]. Population increase of the bird
has been causing significant feeding damage to inland fishery resources in Japan where
P. altivelis is not an exception. Releasing juvenile fishes of P. altivelis in rivers under
the initiative of local fishery cooperatives during every spring is currently a common
way to artificially maintain population of P. altivelis subject to the predation from
P. carbo. Fishery cooperatives and public bodies in Japan have been enthusiastically
developing a number of direct and indirect countermeasures with different qualities to
prevent feeding damage by P. carbo [6, 7], though decisive conclusion to determine the
most optimal countermeasure has not been obtained yet. Theoretical investigation
based on an appropriate mathematical model for population dynamics of P. altivelis
considering predation from P. carbo would be useful for comparing performances of
different countermeasures. However, only a few such researches have been conducted
so far in Japan, which is the main motivation of this paper.
The purpose of this paper is to present a mathematical model for population
dynamics of released P. altivelis in a river system subject to predation by P. carbo and
fishing activity by human. An optimal control problem to maximize the sum of the
 Numerical Simulation of a Hamilton-Jacobi-Bellman Equation 93

profit by harvesting the fish and the cost of countermeasure to prevent the feeding
damage is then presented. Finding the optimal management strategy of the population
of P. altivelis reduces to solving a Hamilton-Jacobi-Bellman equation (HJBE), which is
a time-backward degenerate parabolic nonlinear partial differential equation (PDE).
The HJBE is numerically solved with a stable finite element scheme.
The rest of this paper is organized as follows. Section 2 presents the mathematical
model used in this paper. Section 3 applies the mathematical model to optimal pop-
ulation management of released P. altivelis in Hii River, Shimane Prefecture, Japan.
Section 4 concludes this paper and presents future perspectives of our research.

2 Mathematical Model

2.1 Stochastic Differential Equation Model

A stochastic process model for population dynamics of released P. altivelis in a habitat,
which is a river system, is presented. Let ð0; T Þ with the terminal time T ð [ 0Þ (day) be
a period during which population dynamics of P. altivelis is considered. The time t ¼ 0
(releasing time of juvenile P. altivelis) and t ¼ T (beginning of a closed season of
fishing P. altivelis) are taken in summer and autumn, respectively within a year. The
total biomass of P. altivelis in this river system at the time t (day) is denoted as Xt (kg).
The initial condition X0 (kg) at the time t ¼ 0 (day) is assumed to be deterministic,
which is considered to be valid for the case where most part of population of P. altivelis
is introduced through intensive release events under the initiative of local fishery
cooperatives [8]. Population growth of P. altivelis is assumed not to be prescribed by
environmental capacity as in the usual ecological modelling [9] but rather by the
growth curve of individuals, considering that the population of P. altivelis in such a
case would not saturate in the river system. The governing Itô’s stochastic differential
equation (SDE) of the process Xt is proposed as [10]


dXt ¼ aðt; Xt Þ  RXt  kðut ÞXt  vft  Tc g ct Xt dt þ bðt; Xt ÞdBt ð1Þ

with the Verhulst-type coefficients arising from the growth curve of fishes as
 
aðt; xÞ ¼ r 1  K 1 x x ð2Þ

and

bðt; xÞ ¼ rx ð3Þ

where Bt (day1/2) is the 1-D standard Brownian motion, K ¼ mX0 (kg), m (−), r
(1/day), and r (1/day1/2) are positive parameters, R (1/day) is the natural mortality rate
(1/day), k ðuÞð  0Þ (1/day) is the predation pressure of P. carbo as a function of the
control variable ð0  Þuð  1Þ (−), which is effort to prevent the predation,
ð0  Þcð  cM Þ (1/day) is the fishing pressure, cM is the maximum fishing pressure,
ð0\ÞTc ð\T Þ (day) is the opening time of catching P. altivelis, and vS is the
94 Y. Yaegashi et al.

characteristic function for generic set S. The term bðt; Xt ÞdBt describes temporal
fluctuation of the population of P. altivelis due to natural and artificial environmental
changes in the habitat [11, 12]. A major difference with the present and conventional
stochastic Verhulst models [9] is that the parameter K of the former depends on X0 but
that of the latter does not. Population dynamics of P. carbo is not directly considered in
the SDE (1), but its influences on the population dynamics of P. altivelis are considered
in the term k ðut ÞXt to formulate a minimum mathematical model.
The generator Au;c for the coupled process Yt ¼ ðt; Xt Þ conditioned on Ys ¼ ðs; xÞ
with s\t for sufficiently regular u ¼ uðs; xÞ is expressed as [10]

@u
 @u 1 @ 2 u
Au;c u ¼ þ a  Rx  k ðuÞx  vfs  Tc g cx þ b2 2 : ð4Þ
@s @x 2 @x

2.2 Optimal Control Problem

The variables u and c are taken as control variables in the model and are assumed to be
Markov controls. The admissible ranges U and C of the control variables u and c are
specified to be the compact sets as

U ¼ fuj0  u  1g ð5Þ

and

C ¼ fcj0  c  cM g; ð6Þ

respectively. The objective function to be maximized, which represents the total profit
and is denoted as v ¼ vðs; T; X; u; cÞ (kg), is proposed as
Z T Z T
vðs; T; X; u; cÞ ¼  f ðut Þdt þ vft  Tc g ct Xt dt ð7Þ
s s

where f ð  0Þ (kg/day) with f ð0Þ ¼ 0 is an increasing function. The first and second
terms in the right hand-side of (7) represent the minus of the total cost of operating a
countermeasure to prevent the feeding damage by P. carbo and the total amount of
harvested P. altivelis by human during the period ðs; T Þ, respectively. The coefficients k
and f are set as

kðuÞ ¼ k0 ð1  auÞ ð8Þ

and

f ðuÞ ¼ xðexpðbuÞ  1Þ; ð9Þ

respectively where k0 (1/day) is the predation pressure from the bird without any
countermeasures, ð0\Það  1Þ modulates the effectiveness to decrease the predation
 Numerical Simulation of a Hamilton-Jacobi-Bellman Equation 95

pressure with the control u, bð [ 0Þ (−) modulates the cost to prevent feeding damage,
and x (kg/day) is a parameter serving as a weight that determines the balances on the
cost and the benefit in the objective function. The parameter a represents the effec-
tiveness of the countermeasure, while b represents its inefficiency to prevent the
feeding damage. Values of these parameters would depend on the countermeasures
chosen [6, 7].
The goal of the optimal control problem is to find the optimal controls u ¼ u and
c ¼ c that maximize the objective function v. The maximized objective function
V (kg) is referred to as the value function, which is expressed as

V ðs; x; T Þ ¼ Es;x ½vðs; T; X; u ; c Þ ð10Þ

where Es;x ½
 is the expectation conditioned on Xs ¼ x. According to the dynamic

programming principle for stochastic control problems [10], the HJBE governing
spatio-temporal evolution of the value function V ¼ V ðs; x; T Þ is expressed as
n o  
sup Au;c V  f ðuÞ þ vfs  Tc g cx ¼ Au ;c V  f ðu Þ þ vfs  Tc g c x ¼ 0: ð11Þ
u2U;c2C

The optimal controls u ¼ u and c ¼ c are analytically expressed through the value
function V as

u ¼ b1 minfmaxf0; lnðk0 acÞ  lnðbxÞg; 1g ð12Þ

and

c ¼ vfc  1g cM ; ð13Þ

respectively where the notation c ¼ x @V 

@x has been used. In this model, c is thus given
as a bang-bang type control for each ðs; xÞ. The computational domain of the HJBE
(11) is set as X ¼ ð0; LÞ with a positive constant L (kg) specified later. The HJBE (11)
has to be equipped with appropriate terminal and boundary conditions for its
well-posedness. The terminal condition to be equipped  with the HJBE is set as Vs¼T ¼
0 and the boundary conditions as Vx¼0 ¼ 0 and @V 
@x x¼L ¼ 0, respectively. The former
boundary condition means that when the population of P. altivelis becomes extinct in
the river system, the total profit V is 0, and the latter means that the total profit V
doesn’t increase when the population of P. altivelis exceeds L. Solving the HJBE yields
the optimal controls u and c in the spatio-temporal domain ð0; T Þ  ð0; LÞ.

2.3 Indices for Validating Optimal Management Strategy

Once the optimal controls u and c are computed from the HJBE, a variety of indices
for validating optimal management strategy can be obtained considering the link
between SDEs and terminal and boundary value problems of linear PDEs [10]. The
index J defined as
96 Y. Yaegashi et al.

Z T 
J¼E s;x
gðXt Þdt ð14Þ
s

with an univariate function g solves the Kolmogorov’s backward equation (KBE)

 
Au ;c J þ g ¼ 0 ð15Þ

subject to appropriate
 terminal and boundary conditions, which are J ¼ 0 at the time
t ¼ T, Jx¼0 and @J  ¼ 0. Hereafter, the notation J0 ¼ J ð0; X0 Þ is used for the sake of
@x x¼L  
brevity. Ecologically and economically relevant gðXt Þ can be k ut Xt , vft  Tc g ct Xt ,
 
and, f ut with which the index J0 represents the total weight of predated P. altivelis
by P. carbo, the total weight of harvested P. altivelis as fish catches, and the total cost
to prevent the feeding damage from P. carbo, respectively. The indices with the
above-mentioned three gðXt Þ are denoted in order as J0;pr , J0;cau , and J0;cos , respectively.

3 Application

3.1 Numerical Scheme

The Conforming Petrov-Galerkin Finite Element (CPGFE) scheme is used for
approximating solutions to the HJBE (11) and the governing equations of the indices
(15). The CPGFE scheme is based on the fitting technique where both the trial and test
functions are determined from exponential analytical solutions to element-wise local
two-point boundary value problems [13]. Spatial and temporal discretization procedure
of the scheme is provided in Yoshioka et al. [13] and is therefore not presented here.
For steady linear problems, numerical solutions with the scheme applying an appro-
priate lumping method to the mass matrix guarantee the parabolic discrete maximum
principle [14]. The scheme has been applied to numerically solving a nonlinear elliptic
problem for upstream fish migration in open channel flows [15]. Computational
accuracy of the scheme for linear time-independent problems has been verified, and
theoretical and numerical analyses indicated that its accuracy in space is at least first
order in the L1 -sense at each node. In this paper, a Picard iteration method is equipped
in the spatial discretization procedure of the CPGFE scheme so that nonlinearity of the
HJBE can be numerically handled at each time step.

3.2 Computational Conditions

Parameters of the present mathematical model are estimated from the collected data in
and around Hii River system, San-in area, Japan where most part of population of
P. altivelis are introduced through intensive release events under the initiative of Hii
River fishery cooperatives. The total length of the mainstream and the catchment area
of the river system are 153 (km) and 2,070 (km2), respectively [16]. Hydrological
characteristics of Hii River are well documented and statistically analyzed in detail in
Sato et al. [17, 18], and are not explained in this paper. Two brackish water bodies,
 Numerical Simulation of a Hamilton-Jacobi-Bellman Equation 97

which are Lake Shinji and Lake Nakaumi from upstream, are connected to the main-
stream of Hii River. Hatched larval fishes of P. altivelis are thought to descend to Lake
Shinji; however, it is not known where actually the larval fishes survive during winter.
Constituent members and union members of Hii River cooperatives say that the
number of juvenile fishes ascending the river during spring is significantly decreasing
[8]. In this river system, juvenile fishes of P. altivelis are released in its midstream and
upstream reach during May. The fishes grow up in the system and spawn in down-
stream reaches during October to November.
The initial time t ¼ 0 for numerical computation is set on a day in May. The
terminal time T is set as 180 (day), which is in the autumn in the same year. The
parameters r and K are set as 8:7  102 (1/day) and 20X0 (kg), respectively, con-
sidering a deterministic counterpart of (1) with k ¼ c ¼ 0 [19]. The natural mortality R
is set as 4:6  103 (1/day) [20]. The range of the parameter r is assumed to be
restricted to r\0:4 (1/day1/2), which is the condition that (1) with k ¼ c ¼ 0 has a
non-trivial and non-singular PDF for sufficiently large T [21]. The value of r ¼ 0:25
(1/day1/2) is thus assumed. The other parameters are k0 ¼ 4:0  103 (1/day) [8],
cM ¼ 0:01 (1/day), and x ¼ 1 (kg/day). a ¼ 1 (−) is fixed unless otherwise specified.
The initial condition is X0 ¼ 1:5  103 (kg), which amounts to 3:0  105 juveniles of
P. altivelis [8]. The domain of the population x is set as X ¼ ð0; 6:0  104 Þ (kg), which
is discretized into a mesh with 300 elements and 301 nodes. The time increment for
temporally integrating the HJBE is 0.01 (day). Increasing spatial and temporal reso-
lution of the computation does not significantly change the obtained computational
results below.

3.3 Computational Results

Figures 1, 2 and 3 plot the computed value function V, the optimal control u , and the
optimal control c , respectively. The panels (a)–(c) in each figure present the compu-
tational results with b ¼ 1, b ¼ 5, and b ¼ 10, respectively. The solution V for each b
does not have numerical oscillations, indicating that the present CPGFE scheme can
reasonably handle the HJBE derived in this paper. Figures 2(a)–(c) indicate that the
optimal control u significantly depends on the inefficiency b of the countermeasure, and
the optimal strategy common to the cases with small and moderate b is to intensively

Fig. 1. Value function V for (a) b ¼ 1, (b) b ¼ 5, and (c) b ¼ 10.

98 Y. Yaegashi et al.

Fig. 2. Optimal control u for (a) b ¼ 1, (b) b ¼ 5, and (c) b ¼ 10.

prevent feeding damage by P. carbo after the opening time Tc except for the neigh-
borhood of the terminal time T. In these cases, it is also indicated that the intensive
countermeasure should also be performed before the opening time Tc if the population of
P. altivelis is small. For the largest b, the optimal strategy is not to perform the coun-
termeasure. On the other hand, Figs. 3(a)–(c) indicate that the optimal control c does
not significantly depend on b; the optimal strategy is to harvest P. altivelis after the
opening time Tc except when the population of P. altivelis is small.
Comparisons of a variety of countermeasures to prevent the feeding damage, which
are characterized through the non-dimensional parameters a and b, are next carried out.
The indices considered are the total weight of predated P. altivelis (J0;pr ), the total
weight of caught P. altivelis (J0;cau ), and the total cost to prevent the feeding damage by
P. carbo (J0;cos ) presented in Sect. 2.3. The values examined are a ¼ 0:1i and b ¼ j
where i and j are nonnegative integers such that 0  i; j  10. In total 121 numerical
computations are performed for each index. Figures 4(a)–(c) plot the computed indices
J0;pr , J0;cau , and J0;cos in the a  b phase space. These figures can validate the coun-
termeasures in terms of the indices. The focus here is comparing the methods with high
efficiency and low effectiveness (small b and small a) and those with low efficiency and
high effectiveness (large b and large a). This is because countermeasures with low
efficiency and low effectiveness (large b and small a) should not be used in practice,
and those with high efficiency and high effectiveness (small b and large a) would
merely exist; all the existing countermeasures would have both advantages and dis-
advantages on efficiency and effectiveness. Figures 4(a) and (c) indicate that the

Fig. 3. Optimal control c for (a) b ¼ 1, (b) b ¼ 5, and (c) b ¼ 10.

 Numerical Simulation of a Hamilton-Jacobi-Bellman Equation 99

Fig. 4. The indices J0;pr , J0;cau , and J0;cos for a variety of countermeasures.

method with higher efficiency and lower effectiveness (small b and small a) can more
effectively reduce the total weight of predated P. altivelis and the cost to preventing the
feeding damage. In addition, Fig. 4(b) indicates that the method with higher efficiency
and lower effectiveness more effectively increases the total fish catch. The obtained
computational results thus recommend using a countermeasure with higher efficiency
and lower effectiveness for managing the population of P. altivelis.

4 Conclusions

This paper applied a stochastic control theory to optimal management problem of

population dynamics of P. altivelis in river systems subject to predation by P. carbo
and fishing activity by human. Application of the dynamic programming principle to
the SDE of population dynamics led to the HJBE governing the optimal strategy to
maximize the sum of the cost to prevent the feeding damage and the profit of harvesting
the fish. The model was applied to a real problem of inland fishery management in Hii
River, Japan where most of the juvenile fishes are thought to be introduced through
release events in spring. The CPGFE scheme successfully discretized the HJBE and
computed the value function and the optimal management strategies under a variety of
countermeasures. The presented mathematical and numerical modelling can serve as an
effective tool to control population dynamics of P. altivelis in river systems in Japan.
Future research will extend the present mathematical model to a model with
multiple habitats as considered in Unami et al. [12]. Yoshii Weir and Hinobori Weir
installed in the mainstream of Hii River (Photo 2) have been thought to be physical
barriers for upstream migration of juvenile P. altivelis that possibly fragment habitats of
the fish. Another interesting extension of the present model is to consider hydraulic and
hydrological conditions into the coefficients and parameters, such as the relations
between water quality indices and the growth rate [22, 23]. These extensions would
reduce to numerically solving an HJBE with larger number of independent variables
and/or higher nonlinearity. Mathematical analysis on the HJBE of the present and
extended models will also be performed for better comprehending their mathematical
properties. Possible difficulties to be encountered in the mathematical analysis include
degeneration of the diffusion term [24, 25] and discontinuity of the optimal controls.
Such solutions should be dealt with from the viewpoint of viscosity solutions [26, 27].
100 Y. Yaegashi et al.

Exploring links between the present population dynamics model and the other kind of
related mathematical models, such as the local swimming behavior model [15], is also
an important future research topic for more effectively analyzing migration of
P. altivelis in river systems.

Photo. 2. Yoshii Weir (left panel) and Hinobori Weir (right panel) installed in the mainstream
of Hii River.

Acknowledgements. We thank Dr. Kayoko Kameda of Lake Biwa Museum, and officers in
Hii-River Fishery Cooperatives, Yonago Waterbirds Sanctuary, and the Ministry of Environment
for their valuable comments and providing data. This research is supported by the River Fund in
charge of The River Foundation and JSPS Research Grant No.15H06417.

References
1. Takahashi, I., Azuma, K.: The Up-to-now Knowledge Book of Ayu. Tsukiji-shokan, Tokyo
(2006). (in Japanese)
2. MAFF: Statistics of catch and production amounts of aquatic resources in Japan during 2015
(2015). http://www.maff.go.jp/j/tokei/sokuhou/gyogyou_seisan_13/ (in Japanese)
3. Tanaka, Y., Iguchi, K., Yoshimura, J., Nakagiri, N., Tainaka, K.: Historical effect in the
territoriality of ayu fish. J. Theor. Biol. 268(1), 98–104 (2011)
4. Iguchi, K., Tsuboi, J., Tsuruta, T., Kiryu, T.: Foraging habits of Great Cormorant in relation
to released Ayu stocks as a food source. Aquacult. Sci. 56(3), 415–422 (2008)
5. Katano, O., Abe, S., Nakamura, T.: Relationships between ayu Plecoglossus altivelis and
other organisms in stream communities. Bull. Fish. Res. Agen. Suppl. 5, 203–208 (2006)
6. Yamamoto, M.: What Kind of Bird is the Great Cormorant. http://www.naisuimen.or.jp/
jigyou/kawau/01-1.pdf (in Japanese)
7. Yamamoto, M.: Stand face to face with Great Cormorant II. http://www.naisuimen.or.jp/
jigyou/kawau/03-1.pdf (in Japanese)
8. Hii River Fishery Cooperative. Personal Communication (2015)
9. Tsoularis, A., Wallace, J.: Analysis of logistic growth models. Math. Biosci. 179(1), 21–55
(2002)
10. Øksendal, B.: Stochastic Differential Equations. Springer, Berlin (2003)
11. Doyen, L., Thébaud, O., Béné, C., Martinet, V., Gourguet, S., Bertignac, M., Fifas, F.,
Blanchard, F.: A stochastic viability approach to ecosystem-based fisheries management.
Ecol. Econ. 75, 32–42 (2012)
 Numerical Simulation of a Hamilton-Jacobi-Bellman Equation 101

12. Unami, K., Yangyuoru, M., Alam, A.H.M.B.: Rationalization of building micro-dams
equipped with fish passages in West African savannas. Stoch. Environ. Res. Risk Assess.
26(1), 115–126 (2012)
13. Yoshioka, H., Unami, K., Fujihara, M.: Mathematical analysis on a conforming finite
element scheme for advection-dispersion-decay equations on connected graphs. J. JSCE Ser.
A2 70(2), I_265–I_274 (2014)
14. Mincsovics, M.E.: Discrete and continuous maximum principles for parabolic and elliptic
operators. J. Appl. Math. Comput. 235(2), 470–477 (2010)
15. Yoshioka, H., Unami, K., Fujihara, M.: A Petrov-Galerkin finite element scheme for 1-D
time-independent Hamilton-Jacobi-Bellman equations. J. JSCE Ser. A2. 71, I_149–I_160
(2016). (in press)
16. MLIT (2002). http://www.mlit.go.jp/river/toukei_chousa/kasen/jiten/nihon_kawa/87072/
87072-1 (in Japanese)
17. Sato, H., Takeda, I., Somura, H.: Secular changes of statistical hydrologic data in Hii river
basin. J. JSCE Ser. B1 68(4), 1387–1392 (2012). (in Japanese with English Abstract)
18. Sato, H., Takeda, I., Somura, H.: Secular changes of statistical dry-season discharges in Hii
river basin. J. Rainwater Catchment Syst. 19(2), 51–55 (2014). (in Japanese with English
Abstract)
19. Miyaji, D., Kawanabe, H., Mizuno, N.: Encyclopedia of Freshwater Fishes in Japan.
Hoiku-sha, Tokyo (1963). (in Japanese)
20. Murayama, T.: Why population of Ayu annually changes? In: Furukawa, A., Takahashi, I.
(eds.) River Works to Grow up Ayu. Tsukiji-shokan, Tokyo (2010). (in Japanese)
21. Grigoriu, M.: Stochastic Calculus: Applications in Science and Engineering. Birkhäuser,
Boston (2002)
22. Breitburg, D.L., Rose, K.A., Cowan, J.H.: Linking water quality to larval survival: predation
mortality of fish larvae in an oxygen-stratified water column. Mar. Ecol. Prog. Ser. 178,
39–54 (1999)
23. Budy, P., Baker, M., Dahle, S.K.: Predicting fish growth potential and identifying water
quality constraints: a spatially-explicit bioenergetics approach. Environ. Manage. 48(4),
691–709 (2011)
24. Chernogorova, T., Valkov, R.: Analysis of a finite volume element method for a degenerate
parabolic equation in the zero-coupon bond pricing. Comput. Appl. Math. 34(2), 619–646
(2015)
25. Chernogorova, T., Valkov, R.: Finite volume difference scheme for a degenerate parabolic
equation in the zero-coupon bond pricing. Math. Comput. Model. 54(11), 2659–2671 (2011)
26. Crandall, M.G., Lions, P.L.: Viscosity solutions of Hamilton-Jacobi equations. Trans. Am.
Math. Soc. 277(1), 1–42 (1983)
27. Fleming, W.H., Soner, H.M.: Controlled Markov Processes and Viscosity Solutions.
Springer Science & Business Media, New York (2006)
Fault Displacement Simulation Analysis of the Kamishiro
Fault Earthquake in Nagano Prefecture Using the Parallel
Finite Element Method

Yuta Mitsuhashi1 ✉ , Gaku Hashimoto2, Hiroshi Okuda2, and Fujio Uchiyama1

( )

1
Disaster Reduction and Environmental Engineering Deptartment,
Kozo Keikaku Engineering, Inc., 4-5-3 Chuo, Nakano-Ku, Tokyo 164-0011, Japan
[email protected]
2
Graduate School of Frontier Sciences, University of Tokyo, 5-1-5 Kashiwanoha,
Kashiwa-shi, Chiba 277-8563, Japan

Abstract. In recent years, the evaluation of fault displacement has been required
for evaluating the soundness of underground structures during an earthquake.
Fault displacement occurs as the result of the rupture of the earthquake source
fault, and studies have been conducted using the finite difference method, the
finite element method, etc. The present study used the nonlinear finite element
method to perform a dynamic rupture simulation analysis of the Kamishiro fault
earthquake in Nagano Prefecture on November 22, 2014. A model was prepared
using a solid element for the crust and a joint element for the fault surface. The
Kamishiro fault earthquake in Nagano was a reverse-fault earthquake whose fault
plane included a part of the Kamishiro fault and extended northward from there.
The total extent was 9 km, and the surface fault displacement confirmed was
approximately 1 m at maximum. Initial stress was applied to the fault to inten‐
tionally rupture the hypocenter to perform a propagation analysis of the rupture,
and the displacement and response time history obtained in the analysis were
compared with observational records. At this time, joint elements according to
Goodman et al. that had been expanded were introduced to the finite element
method code FrontISTR, which can analyze large-scale models, and the simula‐
tion analysis was performed.

Keywords: Fault distance · FEM · Joint element · Dynamic rupture simulation ·

Parallel computing

1 Introduction

In recent years, the evaluation of fault displacement has been required for evaluating
the soundness of underground structures during an earthquake. Fault displacement
occurs as the result of the rupture of the earthquake source fault, and studies have been
conducted using the finite difference method (i.e. [1]) and the finite element method (i.e.
[2]). In particular, many studies centering on the finite difference method have been
performed using dynamic rupture simulation analysis, which reproduces the sponta‐
neous rupture process of a fault using a slip-weakening model (i.e. [1]).

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 102–109, 2016.
DOI: 10.1007/978-981-10-2158-9_9
 Fault Displacement Simulation Analysis 103

The present study used the 3D nonlinear finite element method to perform a simu‐
lation analysis of the Kamishiro fault earthquake in Nagano Prefecture on November
22, 2014. The Kamishiro fault earthquake in Nagano (Mj 6.7, Mw 6.2) was a reverse-
fault earthquake whose fault plane included a part of the Kamishiro fault and extended
northward from there, with a length of approximately 20 km and a depth of approxi‐
mately 10 km. The total extent was 9 km, and the surface fault displacement confirmed
was approximately 1 m at maximum [3]. Also, a maximum acceleration of approxi‐
mately 600 cm/s2 was observed in the K-NET [4] observation points near the fault. A
40 km × 40 km × 20 km model that included the earthquake source fault was prepared
using a solid element for the crust and a joint element for the fault. A model was created
for the rupture process of the fault according to the nonlinear constitutive law incorpo‐
rating stress drop, and for the initial conditions, initial stress was applied to the joint
element to rupture the fault, and the ground surface response time history caused by
propagation of the rupture was compared with the observational records of K-NET. Also,
the surface fault displacement observed in studies of the actual site after the earthquake
was compared with the displacement obtained in the present analysis. This research was
conducted using the finite element method code FrontISTR [5], which can perform
parallel computation of large-scale models. This allowed a wide area of the crust to be
analyzed using relatively fine mesh.

2 Analytical Conditions of the 3D Finite Element Method

2.1 Creating a Model of the Fault by Joint Elements

In the present study, a model of the fault was created by the joint elements shown in
Fig. 1. Those joint elements are finite elements that easily simulate the contact/sliding/
exfoliation between two physical bodies. Several joint elements have been proposed,
such as those proposed by Goodman et al. [6] and those that have been formulated on
the basis of 3D isoparametric elements. The shape of the elements according to Goodman
et al. was hypothesized to be rectangular, and the deformation between the two
contacting surfaces is represented by the 6-mode combination shown in Fig. 2. The
authors expanded this to the desired shape of a triangle or quadrilateral, and implemented
FrontISTR. This allowed a high-precision analysis to be performed, even with distorted
mesh.

6 8 7
5

3 4 3 6
2
4 5
2
1 1

Fig. 1. Joint elements

104 Y. Mitsuhashi et al.

q1 q3 q5

q2 q4 q6

Fig. 2. Deformation mode of the joint elements

The relationship between the shearing stress and relative displacement of the joint
element used in this study is shown in Fig. 3(a). The joint elements have a bi-directional
degree of freedom with respect to the translation deformation in the surface (q1 and q2
shown in Fig. 2), so the shearing stress τ and the relative displacement ε are evaluated
in Eqs. (1) and (2). However, fq1 and fq2 are the shearing stress according to modes q1
and q2, and δq1 and δq2 are the displacement of modes q1 and q2.

(1)

(2)

As shown in Fig. 3(a), sliding rupture occurs when the shearing stress τ of joint elements
reaches the yield stress τy, and a stress drop to τ0 occurs. According to past studies (i.e.
[1]), the behavior of the shearing stress after sliding rupture is that it does not rapidly
drop to τ0 but has a certain degree of inclination, and with the slip-weakening model in
the finite difference method, a frequently used model is one in which the shearing stress
drops linearly to critical slip displacement Dc shown in Fig. 3(b). For convenience sake,

τ τ

τy Surface energy
ks τy
Surface energy
τi τi
Stress drop Δτ Stress drop Δτ
τ0 τ0
δ δ
Critical slip distance Dc
(a) This study (b) Slip-weakening model

Fig. 3. Stress displacement relationship

 Fault Displacement Simulation Analysis 105

the present study used a model in which the shearing stress dropped to τ0 the instant that
sliding rupture occurred. Also, according to past studies [2], the analytical results depend
on the relative quantity Δτ = τy − τ0, not on the absolute quantity τ0. Therefore, this study
used an analysis that focused on the relative value Δτ when τ0 = 0MPa. In addition,
referring to Dan et al. ([1]), the initial shearing stiffness ks of the joint elements was
specified so that the surface energy calculated from Dc = 25 cm in the slip-weakening
model was equivalent, and the vertical stiffness kv was made to be linear and a sufficiently
rigid value.

2.2 Movement Equation

The movement equation is given by Eq. (3).

(3)
Here, x is the displacement; M, C, and K are the mass, damping, and stiffness matrix,
respectively; F is the external force vector. For damping, stiffness-proportional damping
was used, and referring to the maximum transmission frequency of the model and the
study by Mizumoto et al. [2], it was established that there would be 2 % damping at
1 Hz. Also, the effect of the reflected wave was eliminated by setting the viscous boun‐
dary to the model periphery that excluded the crust surface. According to the constitutive
law of the joint elements shown in Fig. 3(a), this problem had non-linearity, so a conver‐
gent calculation according to the Newton-Raphson method was performed.

3 Analysis Model

The model used in analysis is shown in Fig. 4, and the fault parameters are shown in
Table 1. The fault shape was established by referring to the AIST active fault database
[7] and aftershock distribution, and a strike angle of 12°, dip angle of 50°, length of
18 km, and depth of 10 km (width of 12.2 km) were established. As for generally used
physical values, ν = 0.25 was hypothesized with the share modulus of stiffness μ set to
30 GPa and the unit weight γ was set to 2.5 t/m3. Referring to the records on the hypo‐
center of this earthquake, the hypocenter was set to a position at a 5-km depth, and the
τy of the joint elements was set at a large value. The Kamishiro fault earthquake was a
reverse-fault earthquake. But the earthquake was also indicated to have been accompa‐
nied by a left-lateral slip. So we assume the direction of initial stress λ as 90° (pure
reverse-fault), 75°, 60° and 45°, and we compare their results(See Fig. 4). By setting an
initial stress, rupture would occur at the same time that the analysis started. Since ground
surface displacement was not observed on the north side of the fault, the shape of the
fault surface might not have been a simple rectangle, but for convenience sake, this study
used a rectangular fault.
106 Y. Mitsuhashi et al.

λ =75° λ =60°
Observation point North
(K-NET Hakuba)

12.2km
λ =90°

20km
18km
λ =45°
hypocenter
40km fault surface model
40km

Fig. 4. Analysis model.

Table 1. Fault parameters

Fault width W 12.2 km
Fault length L 18.0 km
Strike angle θ 12 degree
Dip angle δ 50 degree
Stress drop Δτ 1.00 MPa
Fault shearing stiffness ks 1.20 × 104 kN/m/m2
Fault vertical stiffness kv 1.20 × 104 kN/m/m2

The stress drop was uniformly set to 1 MPa, the condition in which the moment
magnitude Mw, calculated from the fault displacement Δu of each element in the final
state of the analysis by using Eqs. (4) and (5), with the fault surface as Σ, was close to
the value observed in the actual earthquake. τy was established according to Eq. (6),
referring to Andrew [8].

(4)

(5)

(6)

The maximum frequency which can be covered by the mesh fmax, that is calculated in
Eq. (7), is 0.875 Hz when n = 4. However, β is the shear wave velocity at the crust.

(7)
 Fault Displacement Simulation Analysis 107

The analysis was a consecutive nonlinear response analysis according to the

Newmark-β method (parameter β = 0.25, γ = 0.5). The integration time of the analysis
Δt was 0.01 s, and the duration T was 20 s.

4 Analytical Results

The analytical results are shown. The simulation analysis matched well when initial
stress direction λ is 60°. So deformation diagram, fault rupture time, fault displacement
only the case is shown as a representative case.

4.1 Model Deformation Diagram

The deformation diagram of the analysis final time is shown in Fig. 5.

C A
C
N

A
Cross section B

Unit: m
B Deformation
magnification: ×7000

Fig. 5. Analysis model deformation diagram and vertical displacement contour diagram.

4.2 Fault Rupture Time and Fault Displacement

A contour diagram of the time that each joint element of the fault surface ruptured is
shown in Fig. 6(a). The average rupture propagation time Vr was evaluated as 3 km/s
from Fig. 6(a), and the S-wave speed β at the crust was 3.46 (km/s); therefore, the ratio
Vr/β equals 0.87. According to recipe [9], the general ratio Vr/β is 0.72, but the fact that
larger values have been obtained is also described, so the results of this test are not
believed to greatly diverge from the reality. Also, the rupture propagation speed has
been confirmed to increase as a result of changing the surface energy required at the
time of rupture (See Fig. 3).
Contour diagrams of the fault displacement that occurred in the joint elements in the
analysis final time are shown in Fig. 6(b). The seismic moment M0, which is the index
of the size of the earthquake, computed from the fault displacement of the final time was
2.77 × 1018 Nm, and the moment magnitude Mw calculated from M0 was 6.2. This agreed
well with the results for the actual earthquake according to the CMT analysis presented
by the Japan Meteorological Agency in which the seismic moment was 2.98 × 1018 Nm
and the moment magnitude was 6.2.
108 Y. Mitsuhashi et al.

Fig. 6. Fault rupture time and fault displacement.

In addition, the surface displacement of the final time was approximately 62 cm, and
the vertical displacement observed for the actual earthquake after it had occurred was
approximately 90 cm. Therefore, the simulation was believed to have simulated the
actual phenomenon well.

4.3 Response Time History

The response time history at the observation point location was compared with the results
actually observed by K-NET Hakuba, the response acceleration on the ground surface.
In this study, the displacement time history was compared to focus on the relatively
long-term results. So we can neglect the effect of the surface layer. The acceleration
response time history of K-NET Hakuba was corrected by using the method of Boore
et al. [10], and the displacement time history waveform that were obtained by time
integration were compared with the waveform obtained in the analysis. A comparison
of the displacement time history obtained from the observational records with the

Fig. 7. Comparison of displacement response time history. (Color figure online)

 Fault Displacement Simulation Analysis 109

analytical results is shown in Fig. 7. However, the time axis origin has been adjusted as
appropriate.
Because the observation points of K-NET Hakuba were relatively close to the fault,
a relatively large, permanent displacement occurred in the analysis final time. The simu‐
lation analysis matched well when initial stress direction is 60°.

5 Conclusion

A fault dynamic simulation analysis of the Kamishiro fault earthquake in Nagano

Prefecture was performed using the finite element method. By adjusting the stress drop
Δτ, analytical results close to the seismic moment in the actual earthquake were able to
be obtained, and the results for the surface fault displacement and displacement time
history matched well with the observations of the actual earthquake. In the future, a more
comprehensive study that includes a more detailed fault shape and physical property
value variations is desired.

Acknowledgments. This study used the K-NET strong motion seismograms from the National
Research Institute for Earth Science and Disaster Prevention.

References

1. Dan, K., Muto, M., Torita, H., Ohhashi, Y., Kase, Y.: Basic examination on consecutive fault
rupturing by dynamic rupture simulation. In: Annual Report on Active Fault and
Paleoearthquake Researches, no.7, pp. 259–271 (2007). (In Japanese)
2. Mizumoto, G., Tsuboi, T., Miura, F.: Fundamental study on fault rupture process and
earthquake motions on and near a fault by 3D-FEM. J. JSCE, no.780, I-70, 27–40 (2005). 1
(In Japanese)
3. Japan Association for Earthquake Engineering: Report about the Earthquake at the Nagano
Prefecture north in 2014 (2015). (In Japanese)
4. National Research Institute for Earth Science and Disaster Prevention: Strong-motion
Seismograph Networks. http://www.kyoshin.bosai.go.jp/. Accessed 01 Aug 2015
5. FrontISTR Workshop HP. http://www.multi.k.u-tokyo.ac.jp/FrontISTR/. Accessed 01 Aug
2015 (In Japanese)
6. Goodman, R.E.: Methods of geological engineering in discontinuous rocks, Chap. 8, pp. 300–
368. West Publishing Company (1976)
7. National Institute of Advanced Industrial Science and Technology: Active fault data base.
https://gbank.gsj.jp/activefault/index_gmap.html. Accessed 01 Aug 2015 (In Japanese)
8. Andrews, D.J.: Rupture velocity of plane strain shear cracks. J. Geophys. Res. 81, 5679–5687
(1976)
9. Earthquake Research Committee, Headquarters for Earthquake Research Promotion, Strong
motion forecasting method of the earthquake which specified an earthquake source fault
(Recipe). Revision on 21 Dec. 2009 (In Japanese)
10. Boore, D.M., Stephens, C.D., Joyner, W.B.: Comments on baseline correction of digital
strong-motion data: examples from the 1999 Hector Mine, California, Earthquake. Bull.
Seismol. Soc. Am. 92(4), 1543–1560 (2002)
Prediction of 137Cs-Contaminated Soil Losses by Erosion
Using USLE in the Abukuma River Basin, Japan

Carine J. Yi ✉
( )

International Research Institute of Disaster Science, Tohoku University, Sendai, Japan

[email protected]

Abstract. The 2011 Great East Japan Earthquake and tsunami triggered a
significant nuclear power plant accident. Subsequent measurements of the
concentration of cesium-137 (137Cs) showed that the litter and surface layers in
the forest areas near the plant were significantly contaminated. This study applied
the Universal Soil Loss Equation (USLE), which has been widely used to estimate
soil losses from erosion, in the Abukuma River Basin. The greatest soil loss was
predicted to be 1762.75 t ∙ yr−1 ∙ ha−1. To predict losses of 137Cs-contaminated
soil, a 137Cs-soil transfer factor was applied in place of a crop factor, and it
yielded an average contaminated-soil loss rate of 190.65 t ∙ yr−1 ∙ ha−1, whereas
the standard USLE calculation yielded an estimated average soil loss rate of
184.14 t ∙ yr−1 ∙ ha−1. Higher soil losses were predicted in steeper areas west of
the river. However, contaminated soil may be deposited along a comparatively
flat area, such as that on the east side of the river.

Keywords: Fukushima Daiichi Nuclear Power Station explosion · Cesium-137 ·

USLE · Abukuma river basin

1 Introduction

Studies of nuclear events were first conducted during World War II and focused on topics
such as the life cycle, mechanisms, and environmental impacts of fallout. Cesium-137
(137Cs) is among the fission products that generate the most concern because of its beta
and gamma emissions, high radioactivity, and relatively long, 30-year half-life
(Okumura 2003), and it is only produced during nuclear fission events (Ritchie and
McHenry 1990). According to Carter and Moghissi (1977), 137Cs was first released
during early nuclear tests in 1945, and studies on the global dispersal of 137Cs into the
environment began with high-yield thermonuclear tests in November 1952. Perkins and
Thomas (1980) and Ritchie and McHenry (1990) developed the earliest studies of the
dispersal of 137Cs. 137Cs fallout is strongly influenced by local precipitation patterns
and rates (Longmore 1982). However, 137Cs concentrates on the soil surface when the
soil is left undisturbed (Gerspar 1970; Ritchie et al. 1972). Once in contact with the soil,
137Cs is firmly adsorbed to finer soil particles, such as clay, and further movement of
137Cs by natural chemical processes is limited (Tamura, 1964; Ritchie et al. 1982;
Guimarães et al. 2003). Rogowski and Tamura (1965) investigated the difference
between predicted and experimental soil losses by spraying 137Cs onto areas of bare

© Springer Science+Business Media Singapore 2016

S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 110–117, 2016.
DOI: 10.1007/978-981-10-2158-9_10
 Prediction of 137Cs-Contaminated Soil Losses by Erosion 111

clipped meadow and tall meadow cover. Walton (1963) determined the vertical radio‐
activity profiles associated with weapons tests in several New Jersey soils by examining
four radionuclides: strontium 90, ruthenium 106, 137Cs, and 144Cs. McHenry and
Ritchie (1973) found that 137Cs concentrated in accumulation sites, such as floodplains,
lakes, and reservoirs. Okumura (2003) described the life cycle of radioactive 137Cs and
analyzed its material flow in the U.S. Guimarães et al. (2003) described the vertical
distribution of 137Cs in the soil profile and showed its exponential decrease with
increasing soil depth at undisturbed sites. Mizugaki et al. (2008) analyzed suspended
sediment sources using 137Cs and 210Pbex in Japan before the Fukushima Daiichi
Nuclear Power Station explosion.
The 1986 Chernobyl accident, which was classified as level 7 (major accident)
according to the International Nuclear Event Scale (Cardis et al. 2006), has been the
only nuclear accident to provide in-field data that were used to identify contamination
areas and trace radioactive materials, determine the impacts to human health from
external or internal contamination via long-term monitoring, and evaluate the impact of
radioactive fallout on the environment and agricultural activities. Subsequent to the
Chernobyl accident, the Great East Japan Earthquake and Tsunami occurred in 2011
and triggered an explosion at the Fukushima Daiichi Nuclear Power Station. Residents
within a 20-km radius of the Fukushima Daiichi Nuclear Power Station were forced to
leave their homes. Japanese government agencies continue to measure the level of radi‐
oactive contamination in urban areas, agricultural areas, forests, rivers and ocean. A
total of 3,793 observation locations are measured daily throughout the country and used
to chart the current and daily average levels and other measurements of the radioactivity
associated with this disaster (a national radioactivity information list is available online
at http://new.atmc.jp/).
Since the Fukushima Daiichi Nuclear Power Station explosion in 2011, a number of
measurement-based studies have been conducted. Yasunari et al. (2011) applied a
Lagrangian particle dispersion model known as FLEXPART to estimate the total
deposits of 137Cs by integrating daily observations of 137Cs in each prefecture in Japan.
Tonosaki et al. (2012) measured radiation levels in surface soils and found that they
contained greater concentrations of 134Cs and 137Cs relative to deeper soil layers.
However, Hayashi et al. (2012) measured 134Cs and 137Cs contamination in areas
around Mt. Tsukuba, Ibaraki, Japan and found that the levels were higher in forest than
in open areas, such as paddy fields and rural towns. In addition, they found that the litter
layers and surfaces in the forest were significantly contaminated.
Radioactive material released from a nuclear accident follows a process in which it
falls to the ground surface, is partially retained in the soil and vegetation, and to some
extent is then transported into rivers and distributed into bottom sediment (NIES,
2012). In this study, the Universal Soil Loss Equation (USLE) is used to estimate the
transfer of 137Cs through soil erosion into the Abukuma River Basin, which is the closest
river basin to the site of the incident. The USLE is an erosion model designed to predict
the average rate of soil erosion for each feasible alternative combination of crop system
and management practice in association with a specified soil type, rainfall pattern, and
topography (Wischmeier and Smith 1978).
112 C.J. Yi

The Abukuma River runs from southwest to northeast in the Fukushima Prefecture,
which has a population of 1,380,000. The river is 239 km long, and the total river network
is 1,814 km long. The river basin area is 5,405 km2. The river is a first-class fresh water
resource that is surrounded by mountainous areas that have summit heights in excess of
1,000 m and occasionally experience heavy rains (MLIT; Water and Disaster Manage‐
ment Bureau 2003).

2 USLE Application

The USLE is a widely applied erosion model used to estimate soil loss on slopes due to
runoff in specific field areas subjected to particular cropping and management systems
(Warren et al. 2005). Wischmeier and Smith (1978) developed the USLE for conserva‐
tion planners, and the soil equation is defined as follows:

(1)
where
A = soil loss (tons/ha/year),
R = rainfall erosivity factor (tf ∙ m2/ha ∙ h),
K = soil erodibility factor (h/m2),
L = slope length factor,
S = slope steepness factor,
C = cover-management factor, and
P = supporting practice factor (management factor).

The variable A is the calculated soil loss per unit area. It is expressed in the same units
selected for K and includes the same period selected for R. In practice, these parameters
are usually selected such that A can be calculated in tons per hectare per year. However,
other units may be selected if desired. The USLE is an erosion model designed to calcu‐
late long-term soil loss overages from sheet and rill under specified conditions. The
model is also useful for construction sites and other non-agricultural conditions,
although it does not predict deposition or calculate sediment yields from gully, stream
bank, or streambed erosion. Other studies have applied the USLE to Japanese soil (Higa
and Manmoto 2000; Imai and Ishiwatari 2006; Imai and Ishiwatari 2007; Unoki et al.
2010).
R is the rainfall-runoff erosivity factor, and it is calculated from daily precipitation
data. In this study, the Automated Meteorological Data Acquisition System (AMeDAS)
averages of September 2009 were reviewed to determine the R factor, which is defined as

(2)

where I is the rainfall intensity (cm/h) and R is the amount of rainfall (cm). The variable Ei
is defined as follows:

(3)
 Prediction of 137Cs-Contaminated Soil Losses by Erosion 113

K is the soil erodibility factor, and it represents the soil loss rate per erosion index
unit for a specified soil type. The soil erodibility factor is measured on a unit plot defined
as a 72.6-foot long, uniform 9 % continuous slope in clean-tilled fallow soil. The K factor
according to soil type is summarized based on work by Imai and Ishiwatari (2007).
L is the slope length factor, and it is the ratio of soil loss from the field slope length
to the soil loss from a 72.6-foot (22.13-meter) slope length under identical conditions.
S is the slope steepness factor, and it is the ratio of the soil loss from the field slope
gradient to that from a 9 % slope under otherwise identical conditions. The S factor may
be evaluated by combining the L factor for each land cell (Wischmeier and Smith 1978):

(4)

where m is 0.5, 0.4, 0.3, and 0.2 for tan β > 0.05, 0.03 < tan β ≤ 0.05, 0.01 < tan β ≤ 0.03,
and tan β ≤ 0.01, respectively (λ is the slope length in meter and ß is the slope angle in
degrees).
C is the cover and management factor, and it is the ratio of the soil loss from an area
with specific cover and management characteristics to that from an identical area under
tilled, continuous fallow conditions. In Fukushima, the cover/management categories
and corresponding C factors are as follows: orchard (0.2), farm (0.4), rice (0.01), live‐
stock (0.01), flowers (0.01), and grass crops (0.05); these values were suggested by the
Okinawa Science and Technology Promotion Center (OSTC) and are based on the
Okinawa Prefecture land-use map (2001).
P is a supporting practice factor (management factor), and it takes on the following
values: 0.3 for orchards, 0.3 for farms, 0.1 for rice, 0.1 for livestock, 0.1 for flowers, and
0.3 for grass crops.

3 Results

Figure 1 depicts the results of the USLE calculations. In the figure, the color changes
progressively among adjacent regions within the range of white to black, with white
indicating no soil loss and black indicating maximum soil loss. The standard USLE
calculations (left panel) predicted an average loss of 184.14 t ∙ yr−1 ∙ ha−1 of 137Cs-
contaminated soil from the Abukuma River Basin. The greatest predicted soil loss,
1762.75 t ∙ yr−1 ∙ ha−1, was concentrated in the steep areas west and north of the center
of the Abukuma River Basin.
After performing the standard USLE calculations, the P factor was replaced by the
137Cs transfer factor, Tp. The Radioactive Waste Management Center (1988) proposed
137Cs transfer implementation coefficients for agricultural products, for example,
0.00070 for tomatoes, 0.012 for beans, 0.0011 for radishes and 0.0010 for apples. The
Tp value is calculated as follows:

(5)
114 C.J. Yi

Fig. 1. Results of USLE calculations

The right-hand panel in Fig. 1 depicts the distribution of eroded soil contaminated
with 137Cs, which replaced the supporting practice factor as the soil transfer factor. A
maximum soil loss of 449.897 t ∙ yr−1 ∙ ha−1 (an average of 190.65 t ∙ yr−1 ∙ ha−1) was
predicted in the Abukuma River Basin using this calculation. Active soil loss was
predicted upstream of the river, whereas other areas in the hydrology network indicated
minimal or no soil loss.
The Abukuma River Basin is a mountainous region with 77,202 ha of cultivated area
(Abukuma River Basin Risk Reduction Committee 2006), as shown in Fig. 2. In the study
area, an average of 190.65 t ∙ yr−1 ∙ ha−1 (a maximum soil loss of 449.8s97 t ∙ yr−1 ∙ ha−1)
of 137Cs-contaminated soil was predicted by the USLE to be lost through soil erosion.

Fig. 2. Cultivated land in the Abukuma River Basin

 Prediction of 137Cs-Contaminated Soil Losses by Erosion 115

Fig. 3. 137Cs (2012.12.28) air concentrations overlaid on the results of the 137Cs transfer factor
USLE

Higher soil erosion areas were predicted to occur in association with steep slopes and fallow
soil; these areas were found west and north of the center of the study area. However, soil
erosion occurs widely in the study area near the main streams of the Abukuma River.
Figure 3 shows the 137Cs concentration data from December 28, 2012 overlaid with the
results of the 137Cs transfer factor USLE for the 23 of 59 towns in Fukushima Prefecture
that lie within the Abukuma River Basin. The predicted losses of contaminated soil on the
east side of the river are small compared with those of the west side, as apparent in the
figure from the numerous dark gray to black areas in the west, and the 137Cs concentra‐
tions on the east side of the river are still high. High 137Cs air concentrations gradually
spread out from the east coast to the northwest and the southwest. Relatively lower 137Cs
air concentrations occur along the Abukuma River, which consists largely of flatland areas.

4 Conclusions

137Cs has a 30-year half-life cycle and disappears naturally by radioactive decay. This
study showed that 137Cs transfer can be estimated according to soil erosion rates in
flatland areas that contain 137Cs contaminated soil and where predicted soil loss is less
than in steep areas. Therefore, further attention might be required and the findings
described here can help inform disaster recovery strategies. Intense rainfall may be a
significant factor in the transportation of contaminated soil to downstream areas and
should be monitored in future watershed risk management strategies.
116 C.J. Yi

References

Carter, M.W., Moghissi, A.A.: Three decades of nuclear testing. Health Phys. 33(1), 55–71 (1977)
Cardis, E., Howe, G., Ron, E., Bebeshko, V., Bogdanova, T., Bouville, A., Carr, Z., Chumak, V.,
Davis, S., Demidchik, Y., Drozdovitch, V., Gentner, N., Gudzenko, N., Hatch, M., Ivanov, V.,
Jacob, P., Kapitonova, E., Kenigsberg, Y., Kesminiene, A., Kopecky, K.J., Kryuchkov, V.,
Loos, A., Pinchera, A., Reiners, C., Repachol, M., Shibata, Y., Shore, R.E., Thomas, G.,
Tirmarche, M., Yamashita, S., Zvonova, I.: Cancer consequences of the Chernobyl accident:
20 years on. J. Radiol. Prot. 26, 127–140 (2006)
Gerspar, P.L.: Effect of American Beech trees on the gamma radioactivity in soils. Soil Sci. Soc.
Am. Proc. 34, 318–323 (1970)
Guimarães, M.F., Filho, V.F.N., Ritchie, J.: Application of cesium-137 in a study of soil erosion

ᯘㄔ஧ ㉺ᕝᫀ⨾ Ώ㑔ᮍ᮶ Ώ

and deposition in Southeastern Brazil. Soil Sci. 168(1), 45–53 (2003)

㑔ᆂྖ 㘊⧊㐩ၨ ⏣୰ᩔ Ⲉᇛ┴⟃Ἴᒣ᳃ᯘᇦ࠿ࡽ ᨺᑕ

Hayashi, S ( )., Koshikawa, M ( )., Watanabe, M ( )., Watanabe, K (

ᛶࢭࢩ࣒࢘ὶฟ≉ᛶ ᮏ㝣ỈᏛ఍ ᅇ኱఍ ྡྂᒇ ྠㅮ₇せ᪨㞟

)., Nishikori, T ( )., Tanaka, A ( ).: の
. 77 2012 . 89 (2012). (in

ẚ჆ᴿ୕㑻 ‶ᮏ⿱ᙲ ᘧ࡟ࡼࡿᅵተὶฟண ᪉ἲ ᖹᡂ

Japanese)

ᖺᗘὶᇦ㉥ᅵὶฟ㜵Ṇሗ࿌᭩ Ἀ⦖
Higa, E ( )., Manmoto, H ( ).: USLE . 12

௒஭ၨ ▼Ώ ㍤ኵ
( ) (2000). (in Japanese)
Imai, K ( )., Ishiwatari, T ( ).: Regional characteristics of USLE factors based on

ᖹᡂ ᖺᗘ㎰ᴗᅵᮌᏛ఍኱఍ㅮ₇఍ㅮ₇せ᪨㞟
literate data. Extended abstract for 2006 the conference of Japanese society of Irrigation,
drainage and rural engineering, 18 (2006).

௒஭ၨ ▼Ώ ㍤ኵ ⤫ィ㈨ᩱ➼ࢆ⏝࠸࡚ᩚ⌮ࡋࡓ㒔㐨ᗓ┴ู ᅵ
(in Japanese)

ተ౵㣗ᅉ ᆅᇦᛶ࡟ࡘ࠸࡚ ᐮᆅᅵᮌ◊✲ᡤ ᭶ሗ

Imai, K ( )., Ishiwatari, T ( ).: の
の , № 645 (July, 2007)
Ritchie, J.C., McHenry, J.R.: Application of radioactive fallout cesium-137 for measuring soil
erosion and sediment accumulation rates and patterns: a review. Int. J. Environ. Qual. 19(2),
215–233 (1990)
Ritchie, J.C., McHenry, J.R., Bubenzer, G.D.: Redistribution of fallout 137Cs in Brunner Creek
watershed in Wisconsin. Wis. Acad. Sci. Arts Lett. 70, 161–166 (1982)
Longmore, M.E.: The caesium-137 dating technique and associated applications in Australia - a
review. In: Ambrose, W., Duerden, P. (eds.) Australian Archaeometry Conference; Sydney
(Australia), 396 p., pp. 310–321. Australian National University, Canberra, 15-18 February

እᓮ┿⌮㞝 㔠Ꮚ┿ྖ Ύ㔝჆அ ᨺᑕᛶࢭࢩ࣒࢘

1982. ISBN 0 86784 239 3, 37 refs

࡟ࡼࡿ᳃ᯘࡸᮌᮦ࡬ ᙳ㡪࡟ࡘ࠸࡚ ᮌᮦ᝟ሗ ᪥ᮏᮌᮦ⥲ྜ᝟ሗࢭࣥࢱ࣮ ᳃ᯘ⥲ྜ

Tonosaki, M ( )., Kaneko, S ( )., Kiyono, Y ( ).:

◊✲ᡤ
の . ( ,
.) 249, 1–6 February 2012. (in Japanese)
McHenry, J.R., Ritchie, J.C.: Accumulation of fallout cesium 137 in soils and sediments in
selected watersheds. Water Resour. Res. 9(3), 676–686 (1973)

ᅜᅵ஺㏻┬Ἑᕝᒁ Ṋ㝰ᕝỈ⣔ ὶᇦཬࡧἙᕝ ᴫせ ᱌ ཧ⪃

Ministry of Land, Infrastructure, Transportation and Tourism (MILIT)-Water and Disaster

㈨ᩱ ᖹᡂ ᖺ ᭶
Management Bureau ( ): の の ( )-
4-1, (Report for Abukuma River Basin management) 15 11 , November 2003.

㈈ ள⇕ᖏ⥲ྜ◊✲ᡤ
(in Japanese)

ᖹᡂ ᖺᗘ ෆ㛶ᗓጤク஦ᴗ ࠕ⌈⍚♋࡟㛵ࡍࡿㄪᰝࠖ ฼⏝࡟ࡼࡿ㝣ᇦᙳ㡪࡟㛵

Okinawa Science and Technology Promotion Center (OSTC, former: ( ) ):

ࡍࡿㄪᰝ◊✲
13 : GIS
(2001). (in Japanese)
Perkins, R.W., Thomas, C.W.: Worldwide fallout. In: Hanson, W.C. (ed.), pp. 53–82. Technical
Information Center, Oak Ridge (1980)
 Prediction of 137Cs-Contaminated Soil Losses by Erosion 117

Ritchie, J.C., McHenry, J.R., Gill, A.C.: The distribution of Cs-137 in the litter and upper 10
centimeters of soil under different cover types in northern Mississippi: Health Physics, vol.
22, p. 197 (1972)
Rogowski, A.S., Tamura, T.: Movement of 137Cs by Runoff, Erosion and Infiltration on the
Alluvial Captina Silt Loam. Health Phys. Soc. 11(12), December 1965
Warren, S.D., Mitasova, H., Hohmann, M.G., Landsberger, S., Iskander, F.Y., Ruzycki, T.S.,
Senseman, G.M.: Validation of a 3-D enhancement of the universal soil loss equation for
prediction of soil erosion and sediment deposition. CATENA – Interdisc. J. Soil Sci. Hydrol.
Geomorphol. Focusing Geoecology Landscape Evol. 64, 281–296 (2005)
Okumura, T.: The material flow of radioactive cesium-137 in the U.S. 2000, United States
Environmental Protection Agency (2003)
Yasunari, T.J. Stohl, A., Hayano, R.S., Burkhart, J.F., Eckjardt, S., Yasunari, T.: Cesium-137
deposition and contamination of Japanese soils due to the Fukushima nuclear accident. Proc.
Natl. Aced. Sci. USA (PNAS) 108(49), 6 December 2011
Debo, T.N., Reese, A.J.: Municipal Stormwater Management, 2nd edn. CRC Press, Boca Raton,

㬼ᮌၨ஧ ከ⏣኱Ⴙ ୰ᮧ࿴ṇ ụ⏣ᬕᙪ

FL (2003)

⣽ᕝ༤᫂ ⏿ᆅὶᇦ࡟࠾ࡅࡿ ࡟ࡼࡿᅵ◁ὶฟ㔞 ᥎ᐃ ➨ ᅇ

Unoki, K ( )., Tada, D ( )., Nakamura, K ( )., Ikeda, H ( ).,

ᖹᡂ ᖺᗘ ໭ᾏ㐨㛤Ⓨᢏ⾡◊✲Ⓨ⾲఍
Hosokawa, H ( ).: USLE の . 53
( 21 ) , February 2010. (in Japanese)
Walton, A.: The distribution in soils of radioactivity from weapons tests. J. Geophys. Res. 68(5),
1485–1496 (1963)
Wischmeier, W.H., Smith, D.D.: Predicting rainfall erosion losses – a guide to conservation

㜿Ṋ㝰ᕝᅪᇦ⥲ྜὶᇦ㜵⅏༠㆟఍ 㜿Ṋ
planning. U.S. Department of Agriculture, Agriculture handbook No. 537 (1978)

㝰ᕝᅪᇦ Ỉᐖ࣭ᅵ◁⅏ᐖᑐ⟇࡟࠾ࡅࡿㄢ㢟࡜ᙜ㠃 㐍ࡵ᪉ࠊᖹᡂ ᖺ ᭶ ᪥

Abukuma River Basin Risk Reduction Committee ( ):
の の 18 9 5
(2006). (in Japanese)
Radioactive Waster Management Center: Environmental Parameters Series 1: Transfer Factors
of Radionuclides from Soils to Agricultural Products (1988)
Japan Meteorological Agency: AMeDAS; Previous data. http://www.data.jma.go.jp/obd/stats/
etrn/index.php
Mizugaki, S., Onda, Y., Fukuyama, T., Koga, S., Asai, H., Hisamatsu, S.: Estimation of suspended
sediment sources using 137Cs and 210 Pbex un unmanaged Japanese cypress plantation
watersheds in sourthern Japan. Hydrol. Process. 22, 4519–4531 (2008)

ከ፹య⎔ቃࣔࢹࣜࣥࢢ
National Institute for Environmental Studies of Japan (NIES), G-CIEMS(Grid-Catchment
Integrated Modeling System ( ) (2012). (in Japanese)
 Charging and Discharging Characteristics
of a Quantum Well with an Initially Lorentzian
Wave Packet Incident on a DTM Type Potential

Youichirou Kubo and Norifumi Yamada(B)

Information Science, Graduate School of Engineering,

University of Fukui, 3-9-1 Bunkyo, Fukui, Fukui 910-8507, Japan
[email protected]

Abstract. We study how non-interacting electrons accumulate in the

quantum well region of a potential structure (simplified potential struc-
ture of Direct Tunneling Memory) after they start moving toward the
potential as a quantum mechanical wave packet. The probability PL (t) of
finding an electron in the well region obtained with an initially Lorentzian
wave packet behaves differently in several respects from the probabil-
ity PG (t) obtained with an initially Gaussian wave packet. Surpris-
ingly, PL (t) can increase rather than decrease in the decay (discharging)
process, implying that the electrons leaking from the well periodically
change the direction of motion and move backward to the well. This
counterintuitive “backflow effect” is caused by the quantum mechanical
interference between the waves with two wave numbers k0 and kr , where
k0 is the most dominant wave number of the initial wave packet, and kr
is the resonance wave number of the well. We also discuss similarity and
difference between PL (t) and PG (t) in the buildup (charging) process.

Keywords: Quantum dynamics · Non-exponential decay · Direct

tunneling memory

1 Introduction
An important piece that governs the operating speed of quantum devices that use
resonant tunneling phenomenon in quantum well structures is the time response
characteristics of the quantum well. Direct tunneling memory (DTM) [1,2] is
an example of such a quantum device. The memory’s 0 and 1 states correspond
respectively to the absence and the presence of electrons in its quantum well
region. The writing time and the retention time of the memory are related,
respectively, to the buildup characteristics and the decay characteristics of P (t),
which is the probability to find an electron in the quantum well region of DTM at
time t. The behavior of P (t) was studied when the electron incident on the DTM
type potential is characterized as a Gaussian wave packet [3]. For related but
different potential structures such as double barrier structures, many references
exist which study P (t) with Gaussian incident packets [4–6].

c Springer Science+Business Media Singapore 2016
S.Y. Ohn and S.D. Chi (Eds.): AsiaSim 2015, CCIS 603, pp. 118–128, 2016.
DOI: 10.1007/978-981-10-2158-9 11
 Charging and Discharging Characteristics of a Quantum Well 119

An incident wave packet is not necessarily a Gaussian. Little is known about

P (t) when a non-Gaussian wave packet is incident on a DTM type potential.
In this paper, we consider a wave packet of Lorentzian shape as an example
of non-Gaussian incident packets, and study the resultant P (t) in detail. We
will show the following: The obtained P (t) for a Lorentzian incident packet
(hereafter, referred to as PL (t)) shows a rapid increase and a slow decay as in
the case of PG (t) for a Gaussian incident packet. However, the functional forms
are quite different between PL (t) and PG (t). In the buildup (charging) process
where both PG (t) and PL (t) show a rapid increase, log PL (t) is a convex function,
while log PG (t) is a concave function. In the decay (discharging) process, PG (t)
decreases exponentially, whereas PL (t) shows a complicated behavior where the
exponential decay is modulated by an oscillation. Surprisingly, PL (t) increases
periodically in the decay process as the result of the oscillation. This implies that
the electrons leaking from the well periodically change the direction of motion
and move backward to the well to re-charge it.

Fig. 1. A simplified potential profile of DTM and an incident wave packet.

Although an actual DTM is a three dimensional object and the electrons

interact each other, we consider a simplified model where non-interacting elec-
trons described as a single particle wave packet move in one-dimensional space
and the wave packet is scattered by the simplified potential V (x) of a DTM
structure as shown in Fig. 1, where the value of V (x) is 0, V0 , 0, and ∞ for
x < 0, 0 < x < d, d < x < L, and x > L, respectively. If the readers not familiar
with quantum mechanics find it difficult to understand the electron’s accumula-
tion in the well in terms of P (t), simply multiply a large number N with P (t)
to convert the probability to the number of electrons in the well at time t.

2 Probability P (t) to Find an Electron in Quantum Well

We consider the case where a wave packet is incident on the simplified potential
from the left as shown in Fig. 1. The probability to find an electron in the well
region at time t is given by
 L
P (t) = |ψ(x, t)|2 dx, (1)
d
120 Y. Kubo and N. Yamada

where ψ(x, t) is the electron’s wave function, which is the solution to the time
dependent Schrödinger equation

∂ψ(x, t) 2 ∂ 2 ψ(x, t)
i =− + V (x)ψ(x, t) (2)
∂t 2m ∂x2
with a given initial condition ψ(x, 0), where m is the effective mass of the electron
and  is the reduced Planck constant (or Dirac’s constant). The wave function
in the well region can be expressed in the following form as the superposition of
plane waves:
 ∞
1 k2
ψ(x, t) = √ C(k)f (k) sin(k(x − L))e−i 2m t dk, (3)
2π −∞

where f (k), the weighting function for superposition, is the Fourier transform of
the initial wave function, i.e.,
 ∞
1
f (k) = √ ψ(x, 0)e−ikx dx. (4)
2π −∞

The function C(k) is determined from the appropriate boundary conditions

imposed on the wave functions in respective regions, and its modulus |C(k)|2
represents the transparency (transmittivity) of the barrier for the plane wave of
wave number k. The graph of |C(k)|2 has many resonance peaks. We consider a
situation where only one sharp resonance peak appears in the range of k where
f (k) takes non-negligible values (this is physically the most relevant situation).
In this case, only one resonance peak contributes to the k integral in (3). We
can then approximate C(k) in the vicinity of k = kr as follows:

iΔr 1
C(k)  C(kr ), Δr = , (5)
k − kr + iΔr θ (kr )

where θ is the phase of C(k) (= |C(k)|eiθ(k) ), the prime represents a k differen-

tiation, and Δr is the half width at half maximum of the resonance.
The steps of obtaining P (t) is as follows: We give an initial wave func-
tion ψ(x, 0), and use (4) to obtain f (k). We also solve the time independent
Schrödinger equation to obtain the steady state solutions, and determine the
coefficient C(k). We choose a resonance wave number kr at which |C(k)|2 takes
a maximal value. We calculate θ (kr ) numerically by using
  
C (kr )
θ (kr ) = Im . (6)
C(kr )

In order for the requirement stated above (5) to be filled, we must choose the
width δx of the incident packet in such a way that the following relation holds.
1
L  δx  (7)
Δr
 Charging and Discharging Characteristics of a Quantum Well 121

With thus determined f (k) and C(k), we use (3) to calculate ψ(x, t), and then use
(1) to obtain P (t), where the k and the x integrations are calculated numerically.
Let us, for now, consider the initial wave packet given by
 
1 (x − x0 )2
ψG (x, 0) =  exp − + ik0 x . (8)
4
2π(δx)2 4(δx)2

This is called a Gaussian wave packet, because |ψG (x, 0)|2 is a normal distribu-
tion. Equations (4) with (8) gives

2

4 2(δx) 2
fG (k) = exp −δx2 (k − k0 ) − i(k − k0 )x0 , (9)
π
where x0 and k0 are, respectively, the peak position and the central wave num-
ber of the initial wave packet, and δx is the standard deviation of the normal
distribution. Figure 2 shows a typical example of PG (t) which we obtained by fol-
lowing the steps described above. The parameters used are: x0 = −10.0 μm, δx =
0 .5 μm, k0 = kr  4 .9482915 nm−1 , d = 1.0 nm, V0 = 3.3 eV, L = 295.984 nm,
and m = 0.33 m0 (m0 is the bare electron mass). The use of these parameter
values ensures that the requirement stated above (5) is met (see Fig. 3). The
value of Δr is 0.029532489 pm−1 .

0.14

0.12

0.1

0.08

0.06

0.04

0.02

0
0 5 10 15 20 25 30
4.93 4.935 4.94 4.945 4.95 4.955 4.96 4.965

Fig. 2. The probability of finding an Fig. 3. Only one sharp resonance peak is
electron in the well region of the DTM covered by the weighting function fG (k).
type potential when the initial wave
packet is given by (8).

3 P (t) in the Case of Lorentzian Incident Wave Packet

3.1 Lorentzian Incident Wave Packet and an Example of PL (t)
We consider the initial wave packet given by
√
1 δx
ψL (x, 0) = √ eik0 x , (10)
π x − x0 + iδx
122 Y. Kubo and N. Yamada

where we call ψL (x, 0) given above a Lorentzian wave packet, because |ψL (x, 0)|2
is a Lorentz function δx/π{(x − x0 )2 + δx2 }. Substituting (10) into (4), we have
√
2δx exp{−δx(k − k0 ) − i(k − k0 )x0 } (k ≥ k0 )
fL (k) = (11)
0 (k < k0 ).

Not only for the Gaussian incident packet given by (8), but also for the
Lorentzian incident packet given by (10), the wave number k0 determines the
velocity v0 of the packet peak such that v0 = k0 /m, and δx represents the
width of the packet. It should be mentioned here that, although δx in (8) and
that in (10) are both a measure of the “initial packet width,” the two δx have
slightly different meanings; δx in (8) is the standard deviation of the normal
distribution |ψG (x, 0)|2 , while δx in (10) is the half width at half maximum of
the Lorentz function |ψL (x, 0)|2 .
To make the comparison between PL (t) and PG (t) meaningful, we use the
same parameter values as were used in Sect. 2. The choice of the value of k0 is,
however, exceptional. If we use the same value of k0 as was used in Sect. 2, only
a half of the resonance peak is covered by the weighting function fL (k). To avoid
this, we use a slightly different value of k0 (4.9475501 nm−1 ) to make sure that
the resonance peak is fully covered by fL (k) as shown in Fig. 4. With this careful
choice of the parameter values, we obtain PL (t) shown in Fig. 5 by following the
numerical steps explained in Sect. 2.

0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
4.947 4.948 4.949 4.95 4.951 4.952 4.953 4.954 0 5 10 15 20 25 30

Fig. 4. Only one sharp resonance peak is Fig. 5. The probability of finding an
covered by the weighting function fL (k). electron in the well region of the DTM
type potential when the initial wave
packet is given by (10).

3.2 Comparison Between PL (t) and PG (t)

In Fig. 6, we compare PL (t) (solid line) and PG (t) (dashed line). The two curves
share the same characteristics, i.e., they both increase rapidly and decay slowly.
In details, however, there are several differences between them as pointed out
below.
 Charging and Discharging Characteristics of a Quantum Well 123

Figure 7 shows log PL (t) and log PG (t). It is clear from the figure that PG (t)
shows an exponential decrease in the decay process. We have confirmed that
the exponent of the exponential decay agrees with theory; the slope of log PG (t)
obtained from the numerical data, −0.1025284 ps−1 , is very close to the theo-
retical value −0.1025318 ps−1 (= −2k0 Δr /m). From Figs. 6 and 7, we may say
that PL (t) shows a modulated exponential decay, where an exponential decay
with the same negative exponent as in the case of PG (t) is modulated by an
oscillation of a small amplitude.
Figure 8 is an enlarged view of the modulated exponential decay of PL (t).
Surprisingly, PL (t) increases between t = 11.56 ps and t = 11.85 ps. This implies
that the electron moves to the right in this time interval. The same thing happens
at later times as shown in Fig. 9. The temporary increase of the probability in
the well region means that the electron leaking from the well temporarily moves
backward to the well to re-charge it. This view is supplemented by Fig. 11,
where the flux (probability current density) is shown. The electron is moving
to the right when the flux takes a positive value, while it is moving to the left
when it takes a negative value. Figure 11 shows that, in the time interval where
the temporary increase is observed, the flux is positive, and thus the leaking
electron is moving backward to the well. Although the temporary increase of the
probability observed here is small, it can become much larger depending on the
system parameter values as we will see in the next subsection. Note that a large
temporary increase is of practical significance, because it would cause a memory
error of DTM. The cause of the temporary increase is explored in Sect. 3.4.

0.18 0.14 0
0.16 0.12
0.14 -4
0.1
0.12
0.08 -8
0.1
0.08 0.06 -12
0.06
0.04
0.04 -16
0.02
0.02
0 0 -20
0 5 10 15 20 25 30 0 5 10 15 20 25 30

Fig. 6. The comparison between PL (t) in Fig. 7. Natural log plot of PL (t) and
Fig. 5 and PG (t) in Fig. 2. PG (t).

Let us return to Fig. 7 to see the buildup process where the two curves show
rapid increase. We see there that log PL (t) is a convex function, while log PG (t) is
a concave function. This implies that the functional forms in the buildup process
are quite different between PG (t) and PL (t), although both curves in the buildup
process do not look so different in Fig. 6. We consider that the functional form
of P (t) in the buildup process is sensitive to the functional form of the incident
124 Y. Kubo and N. Yamada

0.094205 0.0007
0.0942
0.0006
0.094195
0.0005
0.09419
0.0004
0.094185
0.0003
0.09418

0.094175 0.0002

0.09417 0.0001
0.094165
0
11.5 11.6 11.7 11.8 11.9 12 80
55 60 65 70 75

Fig. 8. In the decay process, a temporary Fig. 9. Temporary increase of PL (t) at

increase of PL (t) is observed between t = later times.
11.56 ps and t = 11.85 ps.

160 0.2

0
120
-0.2
80 -0.4

-0.6
40
-0.8
0
-1
-1.2
-40
0 5 10 15 20 25 30 11.5 11.6 11.7 11.8 11.9 12 12.1

Fig. 10. The flux (the probability den- Fig. 11. The flux is positive during the
sity current) as the function of time. time interval when the backflow effect is
observed.

packet. From this viewpoint, it is not surprising that the functional form of PL (t)
seems very different from that of PG (t) in the buildup process, because the shape
of a Lorentzian packet and that of a Gaussian packet are very different.

3.3 How PL (t) Depends on the Initial Packet Width

The probability P (t) of finding an electron in the quantum well region at time
t depends on many system parameters. Here, among them, let us change the
“width” δx of the initial wave packet to see how P (t) is altered by the packet
size, where δx represents the half width at half maximum of the initial probability
density for a Lorentzian packet, while it represents the standard deviation of the
initial probability density for a Gaussian packet. Figures 12, 13, and 14 show
PL (t) for the case of δx = 500 nm, 1500 nm, and 3000 nm, respectively, where
other parameter values are the same as those used in Fig. 5 except for x0 , which
is −100 μm in the present examples.
 Charging and Discharging Characteristics of a Quantum Well 125

0.18 0.14
0.16
0.12

0.12
0.08
0.08

0.04 0.04

0
0
40 60 80 100 120
40 60 80 100

Fig. 12. PL (t) with δx = 500 nm. Fig. 13. PL (t) with δx = 1500 nm.

10
9
0.06
8
7
0.04 6
5
4
0.02
3
2

0 1
40 60 80 100 120 500 1000 1500 2000 2500 3000

Fig. 14. PL (t) with δx = 3000 nm. Fig. 15. The dependence of the buildup
time on the initial packet width.

We first find that the temporary increase of PL (t) becomes more distinct as
the packet width becomes large. This means that the experimental detection of
the backflow effect would be easier if energetically monochromatic electrons are
used as incident electrons (due to the position-momentum uncertainty princi-
ple in quantum mechanics, a large packet width means a small uncertainty in
momentum, and thus a small uncertainty in energy of the electrons). As men-
tioned earlier, a large temporary increase of the probability is harmful to DTM.
The numerical examples shown here warn that non-Gaussian incident packets
could cause a fatal memory error; one would not notice this in numerical simu-
lations if only Gaussian initial packets are used in the simulations.
The Figs. 12, 13, and 14 also show that the peak value of PL (t) becomes
small as the packet width becomes large, which means that less electrons enter
the well region as the incident electrons become monochromatic in energy (thus
the experimental detection of the backflow effect becomes difficult if the packet
size is too large). We also notice that PL (t) stands up sharply in the build-up
process when the initial packet width is small. To characterize how sharply it
stands up (i.e., how fast the well is filled with electrons), let us define the buildup
time (charging time) by t2 − t1 , where t2 is the time at which PL (t) reaches its
maximum value and t1 is the time at which it starts to take an appreciable value.
The buildup time has a practical meaning for DTM, because it corresponds to
126 Y. Kubo and N. Yamada

the memory’s writing time. Putting t1 to be the time at which the value of
PL (t) reaches 1 % of its maximum value PL (t2 ), we have calculated the buildup
time for some values of packet width. The result is shown in Fig. 15. The figure
shows that, both in the Gaussian case and in the Lorentzian case, the buildup
time increases as the initial packet size increases. This is reasonable because
a longer object takes a longer time to enter the “room” (the well region). It is
interesting to note that the relationship between the buildup time and the packet
size is almost linear in the Gaussian case, while it is apparently non-linear in the
Lorentzian case.

3.4 The Cause of the Backflow Effect

The backflow effect found in Sect. 3.2 (the temporary increase of PL (t)) is coun-
terintuitive, and has a practical significance as explained in Sect. 3.3. It is there-
fore important to clarify the cause of the phenomenon. In quantum mechanics,
an oscillation of probability is often caused by the interference of waves. In our
problem, the incoming wave packet is made up of plane waves with various
wave numbers. Function |fL (k)|2 (see Fig. 4) is the distribution of wave numbers
contained in the wave packet. As shown in Fig. 4, the wave number distribution
|fL (k)|2 starts to take non-zero value at k0 , so that |fL (k)|2 has a peak at k = k0 .
This means that the plane wave with wave number k0 is heavily contained in
the incoming wave packet. The potential structure, on the other hand, allows
the plane wave with wave number kr to be preferably transmitted into the well,
because |C(k)|2 has a resonance peak at k = kr . As the result, the overall weight-
ing factor C(k) f (k) on the right-hand side of (3) is doubly peaked at k = k0
and kr . Therefore, the wave in the well region has two dominant plane wave
components, that is, those with wave number k0 and kr . This view leads us to

|aeik0 x−iω0 t + beikr x−iωr t |2 (12)

as a rough expression for the probability density in the well, where ω0 =

k02 /(2m) and ωr = kr2 /(2m), and a and b are constants. Equation (12) gives an
interference term ei(ωr −ω0 )t . Therefore, (12) and hence |ψL (x, t)|2 would oscillate
in time with a period Tos given by
2π
Tos = . (13)
ωr − ω0
When |ψL (x, t)|2 oscillates in time with period Tos , its integration over position
x, namely, PL (t) would also oscillate in time with period Tos .
We have numerically confirmed that the above view is correct. Figures 16, 17,
18, and 19 show the oscillating part of PL (t), which we obtained by changing the
system parameter values. Equation (13) gives 4.88 ps, 3.45 ps, 2.88 ps, and 0.74 ps
as the oscillation periods expected for Figs. 16, 17, 18, and 19, respectively. It
is clear that the actual oscillation periods obtained from the numerical data
agree very well with the theoretical values given by (13). We may thus conclude
that the oscillations observed in PL (t) is caused by the quantum mechanical
 Charging and Discharging Characteristics of a Quantum Well 127

0.02 0.006

0.016 0.005

0.004
0.012
0.003
0.008
0.002

0.004
0.001

0 0
65 70 75 80 85 90 60 62 64 66 68 70 72 74

Fig. 16. Oscillatory behavior PL (t) in Fig. 17. Oscillatory behavior PL (t) in
the decay process. the decay process.

0.005 0.0009

0.004 0.0008

0.003 0.0007

0.002 0.0006

0.001 0.0005

0 0.0004
60 62 64 66 68 70 72 74 65 65.5 66 66.5 67 67.5 68

Fig. 18. Oscillatory behavior PL (t) in Fig. 19. Oscillatory behavior PL (t) in
the decay process. the decay process.

interference between the waves with wave numbers k0 and kr . The cause of the
counterintuitive backflow effect is, therefore, a purely quantum mechanical one.
The parameter values we used are as follows: As to Figs. 16 and 17:
δx = 3000 nm and kr = 4.94829150000208 nm−1 for both figures; k0 =
4.9475501 nm−1 for Fig. 16 and k0 = 4.94724240001316 nm−1 for Fig. 17. As
to Figs. 18 and 19: δx = 500 nm for both figures; kr = 4.714319700047028 nm−1
and k0 = 4.7130 nm−1 for Fig. 18; kr = 5.160935000005426 nm−1 and k0 =
5.156214700132405 nm−1 for Fig. 19. For Figs. 16, 17, 18, and 19, m = 0.33 m0 ,
x0 = −100000 nm, V0 = 3.3 eV, d = 1 nm, L = 295.984 nm.
Incidentally, although not in the context of DTM, the quantum backflow
effect has also been known for superposition of Gaussian wave functions (see, for
example, Ref. [7]), where the size of the effect is relatively small. The present
study shows that the backflow effect can be made more distinct with a Lorentzian
initial wave function. It also shows that the quantum backflow effect, which is
often studied in a purely academic context, can also be a practical issue in device
applications.
128 Y. Kubo and N. Yamada

4 Summary
We have studied the probability PL (t) to find an electron in the well region of a
DTM type potential when the incident packet is a Lorentzian, and compare the
result with the well-studied case of a Gaussian incident packet.
The obtained PL (t) shows a rapid increase and a slow decay as in the case
of PG (t) for the Gaussian incident packet. In details, however, PL (t) differs from
PG (t) both in the buildup process and in the decay process. In the buildup
process, log PL (t) is a convex function, while log PG (t) is a concave function,
implying that the functional form of PL (t) is very different from that of PG (t).
The buildup time shows a linear dependence on the packet width in the Gaussian
case, but a non-linear dependence in the Lorentzian case. In the decay process,
PL (t) shows a modulated exponential decay. The modulation amplitude increases
as the initial packet width becomes large, and PL (t) eventually begins to oscil-
lates when the packet width exceeds a certain value. The oscillation of PL (t)
periodically causes a temporary increase of the probability. A physical view of
this phenomenon is that the electron leaking from the well periodically changes
its direction of motion and moves backward to the well to re-charge it. The
backflow effect has a practical significance, because it could cause a fatal mem-
ory error of DTM. The backflow effect is counterintuitive and is caused by the
quantum mechanical interference between the waves with wave numbers k0 and
kr , where k0 is the most dominant wave number of the initial wave packet, and
kr is the resonance wave number of the well.

References
1. Kuo, C., King, T.J., Hu, C.: Direct tunneling RAM (DT-RAM) for high density
memory applications. IEEE Electron Dev. Lett. 24, 475–477 (2003)
2. Tsunoda, K., Sato, A., Tashiro, H., Nakanishi, T., Tanaka, H.: Improvement in
retention/program time ratio of direct tunneling memory (DTM) for low power
SoC applications. IEICE Trans. Electron. 88-C(4), 608–613 (2005)
3. Kubo, Y., Yamada, N.: Novel method for simulating quantum buildup process in the
potential well of a DTM potential. In: Proceedings of 33rd JSST Annual Conference
(JSST 2014), pp. 70–71 (2014)
4. Inaba, H., Nakagawa, J., Kurosawa, K., Okuda, M.: Dynamics of resonant tunneling
in double-barrier structures with trapezoidal potential profile. Jpn. J. Appl. Phys.
30, L544–L546 (1991)
5. Hauge, E.H., Støvneng, J.A.: Time-dependent resonant tunneling of wave packets
in the tight-binding model. Phys. Rev. B 44, 13582–13594 (1991)
6. Gong, J., Liang, X.X., Ban, S.L.: Tunneling time of electronic wave packet through
a parabolic quantum well with double barrier. Phys. Stat. Sol. (B) 244, 2064–2071
(2007)
7. Yearsley, J.M., Halliwell, J.J.: An introduction to the quantum backflow effect. J.
Phys.: Conf. Ser. 442, 012055 (2013)
 Author Index

Baek, Seonwha 49 Okuda, Hiroshi 102

Chen, Bin 17 Qiu, Xiaogang 17

Choi, Changbeom 40, 49
Choi, Jaewoong 49 Sakamoto, Naohisa 79
Sheng, Kai 17
Fujihara, Masayuki 91 Shim, Yonghyun 49
Song, Xiao 28
Gong, Na 65 Song, Zhichao 17

Hashimoto, Gaku 102 Takami, Marohito 79

Huang, Xin 65 Triadis, Dimetre 3

Kang, Onyu 49 Uchiyama, Fujio 102

Kim, Bitnal 49 Unami, Koichi 91
Kim, Young Min 40
Kim, Yun Jong 40 Wu, Yulin 28
Konishi, Katsumi 79
Koyamada, Koji 79 Yaegashi, Yuta 91
Kubo, Youichirou 118 Yamada, Norifumi 118
Yang, Ji Yong 40
Lee, Joongsup 40 Yi, Carine J. 110
Li, Zhen 17 Yoshioka, Hidekazu 3, 91

Ma, Yaofei 28 Zhai, Gang 28

Mitsuhashi, Yuta 102 Zhang, Peng 17