See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/312212400

Fuzzy Genetic Algorithms: Fuzzy Logic Controllers and Genetics Algorithms

Article · November 2016

CITATIONS READS

18 17,309

3 authors, including:

Antonia P. Plerou
Ionian University
43 PUBLICATIONS 165 CITATIONS

SEE PROFILE

All content following this page was uploaded by Antonia P. Plerou on 11 January 2017.

The user has requested enhancement of the downloaded file.

IF : 3.62 | IC Value 70.36 Volume-5, Issue-11, November - 2016 • ISSN No 2277 - 8160

Commerce Engineering
Original Research Paper
Fuzzy Genetic Algorithms:
Fuzzy Logic Controllers and Genetics Algorithms

Department of Informatics, Ionian University, Corfu, Greece.

Antonia Plerou
Bioinformatics and Human Electrophysiology Laboratory
Department of Civil Engineering, Democritus University of Thrace
Elena Vlamou
Xanthi, Greece
Vasil Professor, Department of Civil Engineering, Democritus University of
Papadopoulos Thrace, Xanthi, Greece
ABSTRACT The fusion of Genetic Algorithms and Fuzzy Logic Systems allows researchers to model real world problems through
the development of intelligent and adaptive systems. Fuzzy Genetic Algorithm (FGA) is a Genetic Algorithm that uses
fuzzy logic-based techniques. The objective of this blending is to adjust the system parameters to robust and optimize
the performance of the genetic algorithms. This paper provides the basic knowledge of fuzzy systems (FSs) and genetic algorithms (Gas) and
provides an overview of how these technologies are cooperatively combined and have been applied in the real world.
Summary This paper deals with the fuzzy logic inference systems and genetic algorithmic fusion and evaluate the applications of fuzzy genetic
algorithms.
KEYWORDS : Fuzzy sets, Fuzzy logic controllers, Genetic algorithms, Fuzzy genetic
algorithms

Introduction according to the Darwinian evolution rule. In addition, the chromo-

Fuzzy logic and genetic algorithms during the last few years were some which has higher fitness value is having a greater probability
rapidly progressed in the industrial world in order to solve effective- of being selected again in the next generation. After several genera-
ly real-world problems. Fuzzy logic is applied to several fields like tions, the chromosome value is to converge to a certain value which is
control theory or artificial intelligence The term “fuzzy logic” was the best solution for the problem (Grosan, C., & Abraham, 1996).
introduced with fuzzy set theory proposal by Lotfi A. Zadeh in 1965
(Sanchez, Shibata, & Zadeh, 1997) but the idea had been studied Genetic Algorithm Process Basic Steps
since the 1920s as infinite-valued logics notably by Łukasiewicz and The genetic algorithm differs from a classical, derivative-based, opti-
Tarski (Sanchez et al., 1997). A genetic algorithm (or GA) is a search mization algorithm in two essential ways. Namely, a classical algorithm
technique used in computing to estimate approximate solutions for generates a single point at each iteration and the sequence of points
optimization and search problems. Genetic algorithms are evolution- approaches an optimal solution. On the other hand, a genetic algo-
ary algorithms that use techniques inspired by evolutionary biologies rithm generates a population of points at each iteration and the best
such as inheritance, mutation, selection, and crossover (which is also point in the population approaches an optimal solution. Furthermore,
known as recombination) (Vlahavas, 2008). The new population creat- a classical algorithm selects the next point in the sequence by a deter-
ed is thereafter used in the next iteration of the algorithm. In most of ministic computation while a genetic algorithm selects the next popu-
the cases, the algorithm terminates when either a maximum number lation by a computation which uses random number generators (Daras
of generations has been produced, or in the case that the population & Rassias, 2015)assessment, and data management are core competen-
reached a sufficient fitness level. In the case that the algorithm has cies for operation research analysts. This volume addresses a number of
terminated due to a maximum number of generations, it not certain issues and developed methods for improving those skills. It is an out-
that a satisfactory solution has been reached (Sumathi & Kumar, 2016) growth of a conference held in April 2013 at the Hellenic Military Acad-
emy, and brings together a broad variety of mathematical methods and
Genetic Algorithm Description theories with several applications. It discusses directions and pursuits
The basic genetic algorithm philosophy developed by Goldberg was of scientists that pertain to engineering sciences. It is also presents the
inspired by Darwin’s theory of evolution. This theory suggests that theoretical background required for algorithms and techniques applied
the survival of an organism is affected by rule “the strongest species to a large variety of concrete problems. A number of open questions as
that survives”. Darwin’s theory also suggests that the survival of an well as new future areas are also highlighted.This book will appeal to
organism can be maintained through the process of reproduction, operations research analysts, engineers, community decision makers,
crossover, and mutation (Hermawanto, 2013) Darwin’s concept of academics, the military community, practitioners sharing the current
evolution is also used to the computational algorithm in order to find state-of-the-art, and analysts from coalition partners. Topics covered
a solution to a problem known as an objective function in a natural include Operations Research, Games and Control Theory, Computation-
fashion. A solution generated by the genetic algorithm is called a al Number Theory and Information Security, Scientific Computing and
chromosome while as a population the collection of the chromosome Applications, Statistical Modeling and Applications, Systems of Moni-
is referred. A chromosome is composed of genes and its value can be toring and Spatial Analysis. (Daras & Rassias, 2015. The basic steps of a
either numerical, binary, symbols or characters depending on the giv- genetic algorithm are namely the initialization, the evaluation, the se-
en problem. These chromosomes are to undergo a process called fit- lection, the crossover, the mutation and the repetition. These steps are
ness function to measure the suitability of solution generated by the described in details below:
genetic algorithm. Some chromosomes in the population are to mate
through a process called crossover. The procedure named offspring is Initialization is used to create an initial population. This population is
the production of new chromosomes which genes composition are usually randomly generated and consists of a few individuals to thou-
the combination of their parent. In a generation, a few chromosomes sands (Shankar, Mishra, Dehuri, Kim, & Wang, 2016).
are about to go through mutation in their gene. The number of chro-
mosomes which is to undergo crossover and mutation is controlled Evaluation means that each member of the population is evaluat-
by crossover rate and mutation rate value. Chromosome in the pop- ed and a ‘fitness’ for that individual is calculated. The fitness value is
ulation that are to maintain for the next generation are to be selected calculated according to the problems’ desired requirements. These re-
GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS X 497
 Volume-5, Issue-11, November - 2016 • ISSN No 2277 - 8160 IF : 3.62 | IC Value 70.36

quirements could be simple due to the fact that faster algorithms are dle the concept of partial truth, where the value is ranged between
better, or more complex because stronger materials are better as long completely true and completely false. Compared to traditional bina-
as they are not too heavy (Garg, 2015). ry sets (where variables only take on true or false values), fuzzy logic
variables have a truth value that ranges in degree between 0 and 1.
The selection means that individuals are selected according to spe- Fuzzy systems suggest a mathematic model to translate the real pro-
cific selection rules, which are called parents and contribute to the cesses of human knowledge (Plerou, Vlamou, & Papadopoulos, 2016).
population at the next generation. The basic concept of selection
operator is to give preference to improved individuals, allowing them Fuzzy Theory
to pass on their genes to the next generation. The goodness of each Essential research fields in fuzzy theory are fuzzy sets, fuzzy logic,
individual depends on its fitness. Fitness could be determined by an and fuzzy measure. Fuzzy reasoning is an application of fuzzy logic
objective function or by a subjective judgment. Populations overall to knowledge processing. Fuzzy systems have the ability to realize a
fitness should be improving continuously. Selection enhances this complex nonlinear input–output relation as a synthesis of multiple
procedure by discarding the bad designs and solely keeping the best simple inputs–output relations, similar to neural networks function.
individuals in the population. There are several selection methods The input–output relation is described in each rule, nevertheless, the
though the basic idea is not altered, therefore, it is more likely that boundary of the rule areas is fuzzy. The system output from one rule
proper individuals are to be selected for the next generation (Vijaya- area to the next rule area gradually changes. This is the essential idea
rani & Vinupriya, 2015). of fuzzy systems and the origin of the term ‘fuzzy.’ Fuzzy control is an
application of fuzzy reasoning to control. It is worth mentioning that
Crossover is a process of taking more than one parent solutions and even that the majority of fuzzy theory applications were related most-
producing a child solution from them. Crossover is a genetic opera- ly to engineering, they recently address towards other disciplines,
tor used to vary the programming of a chromosome or chromosomes such as medical diagnostics, psychology, education, economy, man-
from one generation to the next one. This is similar to reproduction agement, sociology, etc (Ruan, 1997).
and biological crossover, upon which genetic algorithms are based.
(Shankar et al., 2016) With crossover operator use factor of GA is Fuzzy Sets
prime distinguished from other optimization techniques. Especially, Fuzzy sets are the basic concept of fuzzy logic theory. Let’s consider
two individuals are chosen from the population using the selection X as a nonempty set. A fuzzy set A in X is characterized by its mem-
operator and a crossover site along the bit strings chosen in a ran- bership function µA: X → [0, 1] and µA (x) is interpreted as the degree
dom way. The values of the two strings are exchanged up to and the of membership of element x in fuzzy set A for each x є X. The A set is
two new offspring generated from this mixture are put into the next determined as follows: A = {(u, µA (u))/u є X}. The family of all fuzzy
generation of the population. The procedure recombining portions of sets in X is denoted by F(X) function. When X = {x1, ……, xn} is a finite
good individuals lead to the possible creation of improved individuals set and A is a fuzzy set in X, the notation A = µ1 / x1 + …… + µn / xn is
(Ruan, 2013). used, where the term µi/xi , i = 1,…, n signifies that µi is the grade of
membership of xi in A and the plus sign represents the union of the
Mutation is a genetic operator used to maintain genetic diversity elements. Each element in the universe of discourse is a member of
from one generation of a population of genetic algorithm chromo- the fuzzy set to some extent. The set of elements that have a non-zero
somes to the next one. It is a similar procedure to biological muta- membership is called the support of the fuzzy set. The function that
tion. Mutation alters one or more gene values in a chromosome from ties a number to every element of the universe is known as the mem-
its initial state. In mutation, the solution probably alters completely bership function µ(x) (Ephzibah, 2011).
comparing with the previous solution. Hence, GA can come to a bet-
ter solution by using mutation. The mutation occurs during evolution Fuzzy Logic Controllers
according to a user-definable mutation probability. This probability A Fuzzy Logic Controller is a system which consists of a knowledge
is set low but in the opposite case, the search is about to turn into a base, that includes the information given in the form of linguistic con-
primitive random search (Shankar et al., 2016). trol rules, and a fuzzification interface, which has the effect of trans-
forming crisp data into fuzzy sets. Additionally, an Inference System is
The final step is repetition which means that the algorithm starts included which works together with the knowledge base to provide
again and repeats constantly in order to reach a termination condi- inference with the use of a reasoning method, and a defuzzification
tion. interface, which translates the fuzzy control action obtained to a real
control action by means of a defuzzification method (Abraham, Has-
Genetic Algorithm Applications sanie, Siarry, & Engelbrecht, 2009).
Genetic algorithms are inspired and based on the process of evolu-
tion by natural selection in order to provide solutions to real-world
problems. Specifically, the genetic algorithm is applied in order to
solve several optimization problems, like problems where the objec-
tive function is discontinuous, non-differentiable, stochastic, or highly
nonlinear. The genetic algorithm can address problems of mixed in-
teger programming, where several components are restricted to be
integer-valued (Almeida, Oliveira, & Pinto, 2015). Furthermore, genet-
ic algorithm (GA) are used solving complex search problems such as
engineering to create incredibly high-quality products due to their
ability to search a through a huge combination of parameters to find
the best match. For instance, GA searches through different combina-
tions of materials and designs to find the better combination in order Figure 1: Structure of a Fuzzy Logic Controller
to result in an overall enhanced result. Additionally, they are used to The knowledge base encodes the expert knowledge with the use of
design computer algorithms, to schedule tasks, and to solve several a set of fuzzy control rules. A fuzzy control rule is a conditional state-
optimization problems. ment which is formed as following: “In the case that a set of condi-
tions are satisfied then a set of consequences could be inferred”. The
Fuzzy Logic antecedent is a condition in its application domain, the consequent is
Fuzzy logic provides an inference morphology in order to enable a control action to be applied to the controlled system or else notion
approximate human reasoning capabilities to be applied to knowl- of control rule and both antecedent and consequent are related with
edge-based systems. The conventional approaches to knowledge fuzzy concepts, namely the linguistic terms or else the notion of the
representation lack the means for representation of fuzzy concepts. fuzzy rule. The knowledge base consists of two components. Namely:
The fuzzy logic is an approach to computing based on “degrees of 1) a database, which contains the definitions of the fuzzy control rules
truth” rather than the “true or false” (1 or 0) Boolean logic on which linguistic labels, meaning the membership functions of the fuzzy sets
the modern computer function is based. Fuzzy logic includes 0 and specifying the linguistic terms meaning, and 2) a rule base, which is
1 as extreme cases of truth, nevertheless, the various states of truth consisting of a collection of fuzzy control rules which represent the
in between are also included. Fuzzy logic has been extended to han- expert knowledge (Witold Pedrycz, 2012).
GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS X 498
IF : 3.62 | IC Value 70.36 Volume-5, Issue-11, November - 2016 • ISSN No 2277 - 8160

Fuzzy Genetic Algorithms prerequisite is basic undergraduate mathematics. This third edition
Essential expertise and knowledge on GAs have resulted from em- has been substantially revised and extended by three new chapters
pirical studies conducted over a number of years. This suggests the and by additional appendices containing working material to cover
use of fuzzy logic-based tools for dealing with several problems. An recent developments and a change in the perception of evolution-
essential application of fuzzy logic that is useful for controlling genet- ary computation. (Michalewicz, 2013. An evaluation function to rate
ic algorithms within the control strategies underlying based on the solutions in terms of their “fitness” is also necessary. Therefore, genetic
human experience and knowledge are fuzzy logic controllers (FLCs). operators change the composition of the children in order to make a
Fuzzy Logic Controllers implement an expert operator’s approximate successful run of a GA. In addition, the values for the parameters of
reasoning procedures in the selection of a control action. A Fuzzy the GA have to be defined as the population size and the parameters
Logic Controller allows one to qualitatively express the control strat- for the genetic operators and the terminating condition is evaluated
egies based on experience. These control strategies may be expressed (Michalewicz, 2013)i.e., survival of the fittest. Hence evolution pro-
in a form that permits both computers and humans to share them ef- gramming techniques, based on genetic algorithms, are applicable to
ficiently (Herrera, Lozano, & Verdegay, 1995). many hard optimization problems, such as optimization of functions
with linear and nonlinear constraints, the traveling salesman problem,
Fuzzy Genetic Algorithms (FGA), are Genetic Algorithms integrated and problems of scheduling, partitioning, and control. The impor-
with Fuzzy Logic Controllers. Fuzzy Genetic Algorithms may 1) choose tance of these techniques is still growing, since evolution programs
control parameters before GA’s run, 2) adjust the control parameters are parallel in nature, and parallelism is one of the most promising di-
on-line to adapt to new situations in a dynamic way and 3) assist the rections in computer science.The book is self-contained and the only
user to access, design, implement and validate the Genetic Algorithm prerequisite is basic undergraduate mathematics. This third edition
for a specific task (Herrera & Lozano, 1996). The idea is to use a Fuzzy has been substantially revised and extended by three new chapters
Logic Controllers whose inputs are any combination of Genetic Al- and by additional appendices containing working material to cover
gorithm performance measures or current control parameters and recent developments and a change in the perception of evolutionary
whose outputs are Genetic Algorithm control parameters. Current computation.”, “author” : [ { “dropping-particle” (Michalewicz, 2013. The
performance measures of the Genetic Algorithm are sent to the Fuzzy Fuzzy Genetic Algorithms are mentioned to be most efficient algo-
Logic Controller, which computes new control parameters values that rithm comparing to standard Genetic Algorithms in solving the trav-
will be used by the Genetic Algorithm. eling salesman and other optimization problems (Herrera & Lozano,
1996).

Limitations of Fuzzy Genetic Algorithm

Considering the enhanced degree of nonlinearity of the output of a
fuzzy system, traditional linear optimization tools are noticed to have
their limitations. Genetic algorithms are an enhanced and very pow-
erful method to perform tasks namely the generation of the fuzzy
rule base, optimization of fuzzy rule bases, generation of member-
ship functions, and adjusting membership functions. These tasks are
considered as optimization or search processes within large solution
spaces (Bastian & Hayashi, 1995).

Although genetic algorithms are effective tools to identify the fuzzy

membership functions of a pre-defined rule base, they demonstrate
limitation especially in the case that the input and output variables of
a fuzzy system from a given set of data need to be identified. Genet-
Figure 2: Structure of an Adaptive GA based on Fuzzy ic programming is used to specify the input variables, the rule base
Logic Controller and the involved membership functions of a fuzzy model (Bastian &
The use of Fuzzy Logic-based techniques for either improving GA Hayashi, 1995).
behavior and modeling GA components, the results obtained have
been called fuzzy genetic algorithms (FGAs). Genetic algorithms are The use of Fuzzy Logic Controllers for control Genetic Algorithms
applied in various optimization and search problems involving fuzzy is considered for solving two problems to which a standard Genetic
systems. A Fuzzy Genetic Algorithm is defined as an ordering se- Algorithm could expose limited search speed and premature conver-
quence of instructions in which some of the instructions or algorithm gence. This due to the fact that 1) control parameters are not well
components designed with the use of fuzzy logic based tools. A fuzzy chosen initially for a given task. 2) parameters always being fixed
fitness finding mechanism guides the GA through the search space, even though the environment in which the GA operates could be var-
combining the contributions of various criteria identified as the gov- iable and 3) problems resulting from the selection of several parame-
erning factors for the formation of the clusters (Pedrycz, 2012). ters like population size and in perceiving their influence.

Fuzzy Genetic Algorithms Optimization Discussion

A single objective optimization model is not able to serve the pur- The use of genetic algorithms in determining the optimal rules of the
pose of a fitness measuring index due to the fact that multiple criteria fuzzy logic system has been efficient in providing accurate results. Via
could be responsible for stringing together data items into clusters. the operations of selection, crossover, and mutation the GA will con-
This is valid for the clustering problem and for each problem solving verge over successive generations towards the optimum one. These
using GA involving multiple criteria. In multi-criteria optimization, the operations provide a fast, useful and robust method because genetic
notion of optimality is vaguely defined. The algorithm has two com- algorithms combine an effective and efficient solution. Since popula-
putational elements that work together i.e. the Genetic Algorithm tion implicitly contain much more information than solely the individ-
(GA) and the Fuzzy Fitness Finder (FFF). In the simple cases, there is ual fitness scores, Genetic Algorithm combines the good information
only one criterion for optimization, for instance, maximization or hidden in a solution with good information from another solution in
minimization of profit or cost respectively (Jaganathan & Karthikey- order to produce new solutions with good information inherited from
an, 2014). Nevertheless, in many real-world decision-making prob- both parents, inevitably leading towards optimality (Sivanandam &
lems, the multiple objectives simultaneous optimization is essential Deepa, 2007). The ability of the algorithm to explore and exploit si-
(Michalewicz, 2013)i.e., survival of the fittest. Hence evolution pro- multaneously, a growing amount of theoretical justification, and the
gramming techniques, based on genetic algorithms, are applicable to essential application to real-world problems strengthens the conclu-
many hard optimization problems, such as optimization of functions sion that GAs are an enhanced optimization technique.
with linear and nonlinear constraints, the traveling salesman problem,
and problems of scheduling, partitioning, and control. The impor- References
tance of these techniques is still growing, since evolution programs 1. Abraham, A., Hassanie, A.-E., Siarry, P., & Engelbrecht, A. (2009). Foundations of Com-
are parallel in nature, and parallelism is one of the most promising di- putational Intelligence Volume 3: Global Optimization. Springer Science & Business
rections in computer science.The book is self-contained and the only Media.

GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS X 499

 Volume-5, Issue-11, November - 2016 • ISSN No 2277 - 8160 IF : 3.62 | IC Value 70.36

2. Almeida, J. P., Oliveira, J. F., & Pinto, A. A. (2015). Operational Research: IO 2013 - XVI
Congress of APDIO, Bragança, Portugal, June 3-5, 2013. Springer.
3. Bastian, A., & Hayashi, I. (1995). An Anticipating Hybrid Genetic Algorithm for Fuzzy
Modeling. Journal of Japan Society for Fuzzy Theory and Systems, 10, 801–810.
4. Daras, N. J., & Rassias, M. T. (2015). Computation, Cryptography, and Network Secu-
rity. Springer.
5. Ephzibah, E. P. (2011). Time complexity analysis of genetic-fuzzy system for disease
diagnosis. Advanced Computing: An International Journal, 2(4), 23–31. http://doi.
org/10.5121/acij.2011.2403 23
6. Garg, P. B. (2015). Study of Different Selection Strategies for Implementing Optimi-
zation Problem. International Journal for Scientific Research & Development, 3(2).
7. Grosan, C., & Abraham, A. (1996). Solving nonlinear equation systems using evolu-
tionary algorithms. In Proceedings of genetic & evolutionary computation confer-
ence, Seattle, USA, Proceedings on CD.
8. Hermawanto, D. (2013). G. algorithm for solving simple mathematical equality prob-
lem. arXiv preprint arXiv:1308. 4675. (2013). Genetic algorithm for solving simple
mathematical equality problem. arXiv Preprint arXiv:1308.4675.
9. Herrera, F., & Lozano, M. (1996). Adaptation of genetic algorithm parameters based
on fuzzy logic controllers. Genetic Algorithms and Soft Computing, 8, 95–125.
10. Herrera, F., Lozano, M., & Verdegay, J. L. (1995). Tuning Fuzzy Logic Controllers by Ge-
netic Algorithms. International Journal of Approximate Reasoning, 12, 299–315.
11. Jaganathan, P., & Karthikeyan, T. (2014). An Evolving Approach on Efficient Web
Crawler using Fuzzy Genetic Algorithm. International Journal of Science and Re-
search (IJSR), 3(10).
12. Michalewicz, Z. (2013). Genetic Algorithms + Data Structures = Evolution Programs.
Springer Science & Business Media.
13. Pedrycz, W. (2012). Fuzzy Evolutionary Computation. Springer Science & Business
Media.
14. Plerou, A., Vlamou, E., & Papadopoulos, B. (2016). Visualization of Neuro-Fuzzy Net-
works Training Algorithms: The Backpropagation Algorithm Approach. In I. Deliyannis,
P. Kostagiolas, & C. Banou (Eds.), Experimental Multimedia Systems for Interactivity
and Strategic Innovation (pp. 222–257). IGI Global. http://doi.org/10.4018/978-1-
4666-8659-5
15. Ruan, D. (Ed.). (1997). Intelligent Hybrid Systems. Boston, MA: Springer US. http://doi.
org/10.1007/978-1-4615-6191-0
16. Ruan, D. (2013). Fuzzy Systems and Soft Computing in Nuclear Engineering (Vol.
21). Physica.
17. Sanchez, E., Shibata, T., & Zadeh, L. A. (1997). Genetic Algorithms and Fuzzy Logic
Systems: Soft Computing Perspectives. World Scientific.
18. Shankar, B., Mishra, P., Dehuri, S., Kim, E., & Wang, G.-N. (2016). Techniques and En-
vironments for Big Data Analysis: Parallel, Cloud, and Grid Computing. Springer.
19. Sivanandam, S. N., & Deepa, S. N. (2007). Introduction to Genetic Algorithms. Spring-
er Science & Business Media.
20. Sumathi, S., & Kumar, L. A. (2016). Computational Intelligence Paradigms for Opti-
mization Problems Using MATLAB®/SIMULINK®. CRC Press.
21. Vijayarani, S., & Vinupriya, M. (2015). An Efficient Algorithm for Facial Image Clas-
siffication. International Journal of Signal Processing, Image Processing and
Pattern Recognition, 8(11), 121–134. http://doi.org/http://dx.doi.org/10.14257/
ijsip.2015.8.12.13
22. Vlahavas, I. (Ed. ). (2008). Artificial Intelligence for Advanced Problem Solving Tech-
niques. In IGI Global. (p. 388). IGI Global.
23. Witold Pedrycz. (2012). Fuzzy Evolutionary Computation. Springer Science & Busi-
ness Media.

GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS X 500

View publication stats