Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Overview of Selection Schemes In Real-Coded
Genetic Algorithms and Their Applications
Anil S. Mokhade and Omprakash G. Kakde Visvesvaraya National Institute of Technology /Computer Science & Engineering, Nagpur, India
Email: {asmokhade, ogkakde}@cse.vnit.ac.in
Abstract— The real-coded genetic algorithms are being used
by the research community for diverse applications. This
diversity is further extended into the selection schemes of
different types and combination of them have been tried by
the community. This paper is the overview of the selection
schemes in genetic algorithms with particular reference to
real-coded genetic algorithms. It discusses the evolution of
genetic algorithms, their principles of working and
advantages of real-coded genetic algorithms. It brings out
the important properties of the genetic algorithms and then
discusses the applications of real-coded genetic algorithms
with selection focus. It also points to the further research in
combining the real-coded genetic algorithm using 80:20
principles in selection involving intuition.
Index Terms—real-coded genetic algorithms, selection,
80:20 principle
I. INTRODUCTION
Its more than 50 years that biologically named
‘Evolutionary Algorithms’ are being used as
computational models depicting the evolutionary process
for problem-solving using computers. The three main
techniques used for these evolutionary computations are
‘Evolution Strategies’, ‘Evolutionary Programming’, and
‘Genetic Algorithms’. They are variedly used as a tool for
optimization in engineering problems. Common to all
these methods is the evolution of a population of
candidate solutions of the problem under consideration.
In turn, these solutions are obtained by ‘selection’,
‘crossover’, and ‘mutation’ operators resembling the
natural evolution with genetic variations [1], [2].
A. Evolutionary Techniques
Evolution Strategies: The ‘Evolution Strategies’ are
used for real-valued optimization with self-adaptation of
parameters standard. It is more applied to numerical
optimization with used to solve technical optimization
problems. Rechenberg was pioneer in using this method
that was further developed by Schwefel. These were more
rule based for the variables to change them all at once,
inconsiderably and randomly. If new set is found to be
good then it is to be retained otherwise old status of the
variable can be called for [3].
Evolutionary Programming: The ‘Evolutionary
Programming’ involves representation of the solution by
a finite-state machine. These involved stochastic
optimization with mutation operations on state-transition
diagrams. Thus the emphasis was more on the behavioral
linkages between parents and their offspring. Typically it
involved three steps process. First the initial population
for trial solution was chosen randomly. Then each
solution was replicated into new population and the
offspring solution was subject to mutation based on the
functional change imposed on the parent solution. Fogel
initially worked with the technique of evolutionary
programming, which later was further developed by
Owens, Walsh, and Atmar [1].
Genetic Algorithms: The ‘Genetic Algorithms’ were
propounded by Holland with his students and colleagues
in 60s and 70s. It involved shifting from one population
of solutions to a new population. It was done through
‘natural selection’ combined with crossover, mutation,
and inversion. Here again the inspiration was genetics
and operators involved in it. The living beings have their
chromosomes made of particular genes. These genes, in
turn, take up many different allele values. The
chromosome in Genetic Algorithm is represented by
string length or vector of parameters, each bit of the
string is gene, and the real value of the object is the allele
of the phenotypes [4]. It was a major innovative idea that
gave firm footing to theoretical foundations for the family
of Evolutionary Algorithms.
The basic purpose of any Genetic Algorithm is to
employ the method of searching amongst number of
possible solutions. Thus Genetic Algorithms is a method
of examining a small fraction of possible solutions called
candidate solutions and arrive at optimal or good
solutions. ‘Searching’ unlike the efficient retrieval of
information stored in computer memory, it is to
efficiently find set of actions that will move from a given
initial state to a given goal. In short one can say ‘search
for paths to goal’ [1].
There are two types of Genetic Algorithms, namely,
fixed-length binary coded, and real number
representations or real-coded Genetic Algorithms.
Though fixed length binary coded representations have
dominated research, it is the real coded Genetic
Algorithms, which have shown good properties for
different application problems [5].
A good number of research papers are dealing with the
real-coded genetic algorithms and their applications with
diverse application in engineering and business domains.
Though basic research in all the operator of genetic
algorithm is carried out all over the world the application
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 71doi: 10.12720/jiii.2.1.71-77
Manuscript received July 20, 2013; revised September 21, 2013
specific research is mainly confined with all the operators
combined together and their performance as genetic
algorithm as a whole. In this paper attempt is made to
take an overview of ‘selection’ operator of real-coded
genetic algorithms so as to build the base for performance
analysis of real-coded genetic algorithms.
B. Flow of the Paper
The main concern of this paper is to deal with
‘selection’ operator in real-coded genetic algorithms. In
this dealing the general working of the genetic algorithm
is first discussed. Then real coded-genetic algorithms are
placed vis-à-vis binary coded algorithms. It further
discusses the various research attempts as far as
‘selection’ operator is concerned. Here also the main
concern of the paper is to bring out the applications of
real-coded genetic algorithms with focus on ‘selection’
operator. Further the analytical aspects of research in
‘selection’ operator are discussed. The paper is divided
into six sections. The first section was Introduction. The
second section deals with general working of the Genetic
Algorithms. The third section discusses about ‘selection’
operator and their various types used by different
researchers. The fourth section deals with analysis of the
‘selection’ operator with respect to other operators in
performance of the Genetic Algorithms with particular
reference with real-coded genetic algorithms. The fifth
section deals with 80:20 principle and genetic algorithms.
The last section deals with conclusion and future research
dimensions.
II. GENERAL WORKING OF GENETIC ALGORITHMS
A. Principles of Working
Genetic Algorithms work on the principles of
probabilistic search with stochastic distribution of the
probability. They accumulate the information about initial
unknown search space. Then this information is further
exploited, to favor the subsequent search, resulting into
useful subspaces. This adaptability favors their usage in
large complex, and poorly understood search spaces,
which otherwise classical search methods like
enumerative, heuristic etc. fail to solve satisfactorily [5].
Properties: One of the important properties of Genetic
Algorithms is noise tolerance. Since Genetic Algorithms
work using sampling, populations may be sized to detect
a given degree of function difference with no more than a
specified amount of error [6].
Genetic Algorithms are robust, which is found to be
effective in solving global search problems [7]. This
property of robustness in Genetic Algorithms is due to the
ability to solve hard problems quickly and reliably;
ability to interface to existing simulations and models;
being extensible; and ability to hybridize ([6], [5]).
B. Representation of Genetic Algorithms
In a simple representation the Genetic Algorithms will
deliver the N – a set of number of potential solutions.
Each individual solution called Ji such that Ji € J where
possible individuals are represented in the space by J [8].
It may be mentioned here that all the solution in the set
are sampled simultaneously in the search space. In fact
that is distinguishing characteristics of Genetic
Algorithms. Thus a population P ={ J1, J2, …, JN } is
formed. This population is then subjected to natural
evolutionary process operators after initializing the
population. These operators are selection, crossover and
mutation. All these operations are run in a loop. One run
execution of loop is called generation and population at
the end of the loop denotes the population at that
generation. Figure 1 gives the generational model of
genetic algorithm [9].
procedure GA
begin
t = 0;
initialize P (t );
evaluate structures in P (t );
while termination condition not satisfied do
begin
t = t + 1;
select P (t ) from P (t !1);
apply genetic operators to structures in P (t );
evaluate structures in P (t );
end
end.
Figure 1. Generalized model for genetic algorithms
Thus genetic algorithm is a simple but robust
algorithm due to implicit parallelism it exhibits among
the subsets of coded parameter set within the population
under consideration using probabilistic transition of
operators without any domain specific information ([9],
[10]). These subsets are either binary-coded or real value
coded. Many theoretical studies have been done using
binary- coded genetic algorithms but many real life
applications have been dealt with using real-coded
genetic algorithms. The binary-coded genetic algorithms
can be studied in many other references including
Goldberg [11]. Here in this paper we are going to deal
with real-coded genetic algorithms.
III. REAL-CODED GENETIC ALGORITHMS
The binary-coded and real-coded genetic algorithms
differ in the usage of alphabets for representing genes or
parameters. Real-coded genetic algorithms have some
advantages over the binary-coded genetic algorithms,
which we will discuss later in this paper. It was most
natural to use real numbers to represent the genes directly.
This was particularly for problems of optimization
involving variable parameters in continuous domains [5].
Thus there is no difference between coding part and the
search space. A vector of floating point numbers
represents the chromosome. The vector length is for
solution to the problem. This makes each gene as a
variable to the problem. This forces gene values to
remain in the interval established by the variables they
represent. This requirement therefore must be fulfilled by
the genetic operators [12].
Going by the research and many studies on
comparative analysis of both these types of genetic
algorithms, the real-coded genetic algorithms are found to
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 72
be superior in following aspects [5],[13]. The precession
is achievable in real-coded genetic algorithms even after
increasing the domain. Graduality as defined by Herrea
and others, which means slight change in variables causes
slight changes in the function, can be very well exploited
by using real-coded genetic algorithms. When using real-
coded genetic algorithms, there is no difference between
coding (genotype) and search space (phenotype). This
gives solutions representation near to natural
formulations. According to Arumugam [14], the
implementations are carried in real-coded genetic
algorithms so as to move them closer to the problem
space. This makes the some specific characteristics of
real space operators more problem specific with some
specific characteristics of real space [14]. Thus one can
conclude that inadequacy of binary-coded genetic
algorithms to deal with continuous search spaces with
large dimensions and considerably higher numerical
precision have been dealt very well by real-coded genetic
algorithms [5].
Going by these advantages of real-coded genetic
algorithm, it will be greatly advantageous to study them
in detail. Number of researchers in several real-life
applications has tested them. The next sections take a
look at the existing status of research on selection
operator and its applications using real-coded genetic
algorithms.
IV. SELECTION OPERATOR
We have seen in the previous sections that selection
operator is the first operation acted upon on the
population of the solution after initialization. It selects the
chromosomes in the population for reproduction. If one is
able to get fitter chromosomes, they are most likely to be
selected for reproduction. It also will decide how many
offspring each chromosome will produce. The selection
operator should be chosen of such a nature that it
maintains the balance in crossover and mutation operators
mimicked in the genetic algorithms. If we have greater
emphasis on selection operator then population will have
highly fit solutions/population resulting reduced diversity,
which is very essential for further change and progress. If
one gives less emphasis on the selection operator, it may
result in slowing down the process of optimum solution
evolution [1]. In the ultimate analysis we must understand
that selection operator’s main purpose is to improve the
average quality of the population by increasing the
probability of high quality individuals so as to enable
them to be copied into the next generation [8].
A. Different Selection Schemes
Since Holland gave his schema theorem and used
proportional selection, many researchers have used
different selection schemes. Each selection scheme has
advantages and some limitations. It has also been debated
in the research community over these advantages and
limitations. The different schemes include [15].
Proportionate reproduction
Ranking selection
Tournament selection
Steady-state selection
Roulette wheel selection
Stochastic Universal Sampling
Local Selection
Truncation Selection etc.
Which selection schemes is best is still an open
question for the research community. The research
community is using different selection schemes but the
key idea is that a “survival-of-the-fittest” mechanism
biases the generation of the new individuals [16]. We will
first present the theoretical basis for understanding the
these selection schemes.
B. Theoretical Underpinnings for Selection Operator
Over the period different researchers have analyzed the
selection operator in particular and genetic algorithms
using various concepts and terms in general. They can be
called performance measures for genetic algorithms.
These include parameters like take over time, selection
intensity, average fitness, fitness variance, reproduction
rate, loss of diversity, and selection variance [8]. These
researchers have analyzed the selection schemes based on
its interaction on the distribution of fitness values. They
have dealt with mathematical foundations for these
properties which out of purview of this paper.
By take over time one means that the number of
generations required for a single best individual to fill up
the whole generation without using recombination. The
selection intensity one means the normalized increase of
the average fitness of the population after selection [17].
Fitness here means the solution, which are perfect not
only in its form and function but also able to contribute to
the future generation [18]. Here we look at fitness from
average and variance aspects. The Reproduction rate
denotes the ratio of the number of individuals with certain
fitness value after and before selection. Loss of diversity
we mean loss of bad individuals and replacement of them
with better individuals. Selection variance is same, as
selection intensity with new variance after the selection
operator has been applied.
V. SELECTION OPERATOR AS USED IN REAL-CODED
GENETIC ALGORITHMS
The selection operator as employed by real-coded
genetic algorithms differs with the applications. In this
section we will look into some of the applications using
real-coded genetic algorithms for the analysis of selection
operator.
Cuang uses a real-coded genetic algorithm with
ranking selection with direction-based crossover and
dynamic random mutation. The whole idea behind the use
of ranking selection is to eliminate bad solutions and
reproduce good solutions. This enables whole population
attains a better average fitness [19].
Baskar et. All [20] have use Roulette wheel selection
for their hybrid real coded genetic algorithm for
economic dispatch problem.
In Fossati’s Real-coded Extended Compact Genetic
Algorithm (rECGA) applies tournament selection and
restricted tournament replacement applied between
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 73
original real population and new real population. It is
observed that as the tournament size increases, the
convergence of rECGA becomes more easier [16].
Alberto et al. [21] suggest the use of two types of
selection operators: ranking selection- surviving and
death, elitist selection and roulette wheel selection
operators using Pareto optimality for multi-objective
evolutionary algorithm. The surviving selection first
assigns higher probability to smaller rank individuals
followed by roulette wheel selection. In death selection
gives probability of not surviving i.e. higher probability
to those individuals with greater ranking. Again, then
applying roulette wheel selection to eliminate spare
individuals. Lastly elitist selection is applied randomly to
eliminate spare individuals. The main aim was to improve
the behavior of the crossover operator with the these
selections.
Arumugam and others [14] used the Roulette-wheel
selection (RWS), Tournament selection (TS) and hybrid
of both selection procedures in conjunction with elitist
strategy. Elitist strategy ensures that the best individual in
each generation is passed to the next generation. Thus it
shows two kinds of hybrid selections, namely single level
and two level hybrid selection methods. First half is TS
and next half is RWS in the single level. In the two level
hybrid selection one fourth for TS and another one fourth
for RWS and again repeating the same sequence. The
results are better for the optimal control problem. Same
authors have further introduced hybrid crossover methods
and new hybrid selection methods for integrated
processes and optimal control in manufacturing
environment.
Subbaraj and others [22] have done enhancements of
self-adaptive real-coded genetic algorithm using Taguchi
method for economic dispatch problem. Here they have
employed tournament selection scheme using powerful
exploration capability. In this they bring out tournament
between two solutions to get the better one. This again
helps in better convergence and reduced computational
burden as claimed by authors.
Tang and others [23] proposed adaptive directed
mutation operator in real-coded genetic algorithm to
solve complex optimization problems using linear fitness
scaling mechanism and then selectively applying
selection through stochastic universal sampling. This is a
variant of roulette wheel selection. But unlike roulette
wheel selection it gives a starting position by employing a
single random number and then proceeding in equal-sized
steps around the wheel. Here the number of individuals to
be selected determines the step size. The results showed a
significant improvement in global optimum solution.
Sushil Kumar and others. [24] while dealing with non-
convex economic load dispatch have proposed use of
real-coded genetic algorithm using tournament selection
to get the selective pressure towards feasible region.
Their aim is to bring out the competition among
chromosomes and ultimately achieve better
computational efficiency, which is crucial for economic
operation of a power system.
Oyama and others [25] employ ranking selection in
real-coded adaptive range genetic algorithm (ARGA) so
that one-point crossover operation can be done so as to
achieve a practical three-dimensional shape and
optimization for aerodynamic design. They have included
the binary-coded ARGA using floating-point
representation.
These uses of selection schemes in real-coded genetic
algorithm indicate that there is wide differentiation in
approaches to application of real-coded genetic
algorithms in selection schemes. The ranking selection,
tournament and roulette wheel selection schemes are
mostly being used with some variations. These variations
are not in isolation but integrated with other operators.
The various performance measures are either explicitly or
implicitly mentioned in these research. With this look at
the applications using real-coded GAs we can take a look
at some the conceptualizations analytically which lie at
the bottom of these applications in general.
VI. ANALYTICAL ASPECTS FOR THE USE OF REAL-
CODED GENETIC ALGORITHMS
In this section attempt is take a look at the various
aspects, which favor the us of real-coded genetic
algorithms. The very first among these aspects is usage of
real-coded GAs instead of binary coded GAs itself.
Shortcomings in Binary Coded GAs: Research has
pointed out following shortcomings in the usage of
binary-coded GAs:
1) Difficulty in gradual search in continuous search
space: Hamming cliffs in a binary coding compel
deliberate hindrance to gradual search, as it
requires alteration of many bits for transition to
neighboring solution.
2) Not able to get any arbitrary precision in optimal
solution: Binary coded GAs have compulsion of
choosing precision before hand. More the
precision, larger the string length, and
therebyrequiring larger population. This
increases the computational complexity.
When real-coded genetic algorithms are used operators
crossover and mutation can be used directly to real
parameter values. Fitness values can be computed with
the direct use of decision variables. This helps in working
with selection operator. Only difficulty in real-coded
genetic algorithms is that of creation of new pairs of
offsprings from a pair of real-parameter decision
variables. Significant contribution towards the real-
parameter crossover and mutation operators is carried out
by [5], [26], [27].
Conceptual Tools for Selection Operator as used in
Real-coded Genetic Algorithms: One must understand
that though there are many papers published on the
application of real-coded genetic algorithms, there are
very few studies on theoretical understanding at the
conceptual level. This is reflected in the applied research
in the applications of the real-coded genetic algorithms.
Following are some of the studies highlighting selection
operator usage conceptualization. These studies focus on
two conceptual tools:
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 74
1) Search with blocking function
2) Distribution of the population using probability
density function.
Search with blocking function: The search in real
coded genetic algorithms causes retention of above
average solutions in the population. This is because real-
coded GAs starts treating continuous search spaces as
discrete search space problem there by behaving like
binary-coded GAs. This restricts the search to certain
blocks in search space. Only strong mutation can bring
some diffusion to create solutions in the outer domain.
The blocking function thus defined on the basis of virtual
alphabets, ensures global optimum within the block or
region. Otherwise it is difficult to find global optimum as
real-coded GAs will be converging at the intersection of
virtual characters of different decision variables [28].
Distribution of the population using probability density
function: The other conceptual tools used in
understanding the real-coded genetic algorithms are
probability density function denoted by πt. This gives
distribution of population at any generation t. This
probability density function is affected by the way the
selection and mutation operators are carried out on the
population. The combination of selection and mutation
operator helps to shift towards optimum in a unimodal
problem. The selection operator causes reduction in
variance of PIT whereas spread in distribution is caused
by mutation operator. This will approach Dirac function
for large number of generations. For areas above-average
fitness, selection operator alone increases the density. If
selection and crossover effect is seen it is observed that
the hyperplanes of higher fitness value are sampled more
frequently resulting in more frequent sampling of better
solutions. If schemata equivalence is to be considered
then intersecting hyperplanes means smaller regions or
areas for real-coded GAs giving higher-order schemata.
[29], [4].
The other aspect, which is to be looked into the
research on selection operator in real-coded genetic
algorithms, is the performance rating which takes us into
the domain of Software Engineering. In order to have the
performance rating of the real-coded genetic algorithms
we can further study it by applying a universally
practiced random scheme proposed through 80:20
principle.
VI. 80:20 PRINCIPLES AS APPLIED TO GENETIC
ALGORITHMS
80:20 principle is an empirically providing the test for
the nonlinearity [30]. It is also called Pareto Law. The
principle assertion of this law is that the majority of the
results, outputs or rewards are usually caused by very less
number of causes, inputs or efforts. Many of the
solutions are weaker and in genetic algorithm they are to
be disregarded. If the survival of the fittest strategy is to
be applied then it must be able to give us those solutions,
which are good even though their number is less. These
solutions need to be identified and multiplied. By
selecting 20% of these multiplied solutions and repeating
the different stages of algorithms the optimum solutions
can be achieved. It seems that this approach is special
case of proportionate selection method as applied in
genetic algorithms.
Some of the researchers have applied this principle in
their works ([31], [32]). They have claimed the avoidance
of premature convergence problem and finding most
critical paths respectively. In the ultimate analysis it is
observed that real-life problems involving practical
decision making, involves improvement in the directions
of effective constraints and keeping track of effective and
ineffective constraints if large number of constraints exist
[33]. If one applies real-coded genetic algorithm solution
using 80:20 rule, it has to incorporate these two issues
particularly in the selection scheme itself. The researchers
will have to work on the possibilities of quantifying the
heuristics to arrive at the decision parameters.
VIII. CONCLUSION AND FUTURE WORKS
The emphasis of this paper was on taking an overview
of applications of real-coded genetic algorithms with
emphasis on selection operator both at the application
level and performance analysis level using selection
operator. We have thus studied the selection operator
with its variations in genetic algorithm in general and
real-coded genetic algorithm in particular. We see the
diverse applications for which the real-coded genetic
algorithms are being used. Though, we could not find
much reference on selection operator comparisons at the
application level in real-coded genetic algorithm. The
experimental results, which are discussed in the paper,
are indicative of almost similar kind of output of research
in applications of real-coded genetic algorithm.
Discussing them will not add much differentiation as far
as applied research is concerned. Then we have seen the
analytical aspects for the use of real-coded genetic
algorithms with particular emphasis on selection operator.
We have also seen the research efforts in the performance
analysis using 80:20 rule.
From this study attempt it is humbly put that though
real-coded genetic algorithms are widely used in many
diverse applications, there is still large scope for further
research at the singleton ‘operator’ level. The
performance of the real-coded genetic algorithms with
distribution of the population using probability density
function with selection operator effect is discussed. This
line of research further can be utilized for performance
analysis in specific applications of the genetic algorithms
using real-coded alphabets. The author proposes to
undertake work in this regard and apply it to software
development effort and cost estimation that is open for
research due to changing modeling techniques including
agile process modeling.
ACKNOWLEDGMENT
This research is supported by the Cumulative
Professional Development Allowance given by the
Government of India through Ministry of Human
Resource Development.
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 75
REFERENCES
[1] M. Mitchell. An Introduction to Genetic Algorithms, A Bradfod Book, MIT Press,1998.
[2] W. M. Spears, K. A. De Jong, T. Bäck, D. B. Fogel, and H. de Garis, “An overview of evolutionary computation,” in Proc.
European Conference on Machine Learning, Vienna, Austria,
April 5-7, 1993. [3] H. Beyer and H. Schwefel, Evolution Strategies A
Comprehensive Introduction, Kluwer Academic Publishers,
Netherlands, 2002. [4] K. K. Deb, Multi-objective Optimization using Evolutionary
Algorithms, John Wiley & Sons, Ltd. Singapore ,2003.
[5] F. Herrera, M. Lozano, and J. L. Verdegay, “Tackling real-coded genetic algorithms: Operators and tools for behavioural analysis,”
Journal of Artificial Intelligence Review, Kulwer Academic
Publishers, Netherlands, vol. 12, no. 4, pp. 265-319 Aug 1998.
[6] D. Goldberg, “Genetic and evolutionary algorithms come of
age,” Communications of ACM, vol. 37, no. 3, pp. 113-119,
March 1994. [7] K. A. De Jong, Evolutionary Computation A Unified Approach,
The MIT Press – PHI Learning Pvt. Ltd. New Delhi 2009.
[8] T. Blickle and L. Thiele, A Comparison of Selection Schemes used in Genetic Algorithms, TIK-Report Version 2, Computer
Engineering and Communication Networks Lab, Zurich,
Switzerland, 1995. [9] J. Grefenstette and J. Baker, “How genetic algorithm work: A
critical look at implicit parallelism,” in Proc. Third International
Conference on Genetic Algorithms, San Mateo, CA, 20-27, 1989. [10] F. Chang and L. Chen, “Real-coded genetic algorithm for rule-
based flood control reservoir management,” Water Resource
Management, Kluwer Academic Publishers, Netherlands, vol. 12, no. 3, pp. 185-198, 1998.
[11] D. Goldberg, Genetic Algorithms in Search, Optimization and
Machine Learning, Addison-Wesley, MA, 1989.
[12] A. Blanco, M. Delgado, and M. Pegalajar, “A real-coded genetic
algorithm for training recurrent, neural networks,” Neural
Networks, vol.14, no. 1, pp. 93-105, Jan 2001. [13] A. W. Genetic, “Algorithms for real parameter optimization,” in
Foundations of Genetic Algorithms, G. J. E. Rawlins, Ed.,
Morgon Kaufmann, San Mateo, CA, 1991, pp. 205-218. [14] M. Arumugam, M. Rao, and R. Palaniapan, “New hybrid genetic
operators for real coded genetic algorithm to compute optimal
control of a class of hybrid systems,” Journal of Applied Soft Computing, vol. 6, no.1, pp. 38-52, January 2005.
[15] D. Goldberg and K. Deb, “A comparative analysis of selection
schemes used in genetic algorithms,” in Foundations of Genetic Algorithms, G. J. E. Rawlins, Ed., Morgon Kaufmann, San Mateo,
CA, 1991.
[16] I. Fossati, P. Lanzi, K. Sastry, D. Goldberg, and O. Gomez, “A simple real-coded extended compact genetic algorithm,” IEEE
Congress on Evolutionary Computation, Singapore, September
25-28, 2007.
[17] E. Cantu -Paz, “Markov chain models of parallel genetic
algorithms,” IEEE Transactions on Evolutionary Computation, vol. 4 no. 3, pp 216–226., 2000.
[18] Fitness and Adaptation. [Online]. Available:
http://biomed.brown.edu/Courses/BIO48/13.Fitness.Adapt.1.HTML
[19] Y. Chuang and C. Chen, “A study on real-coded genetic
algorithm for process optimization using ranking selection, direction-based crossover and dynamic mutation,” IEEE
Congress on Evolutionary Computation, New Orleans, LA, June
5-8, 2011. [20] S. Baskar, P. Subbaraj, and M. V. C. Rao, “Hybrid real coded
genetic algorithm solution to economic dispatch problem,”
Computers & Electrical Engineering, vol. 29, no. 3, pp. 407-419, May 2003.
[21] I. Alberto and P. Mateo, “A crossover operator that uses pareto
optimality in its definitions,” TOP Springer, vol. 19, no. 1, pp. 67-92, July 2011.
[22] P. Subbaraj, R. Rengaraj, and S. Salivahanan, “Enhancement of
Self-adaptive real-coded genetic algorithm using Taguchi method for Economic dispatch problem,” Journal of Applied Soft
Computing, vol. 11, no. 1, pp. 83-92, January 2011.
[23] P. Tang and M. Tseng, “Adaptive directed mutation for real-coded genetic algorithms,” Journal of Applied Soft Computing,
vol. 13, no. 1, pp. 600-614, January 2013.
[24] S. Kumar and R. Naresh, “Nonconvex economic load dispatch using efficient real-coded genetic algorithm,” Journal of Applied
Soft Computing, vol. 9, no. 1, pp. 321-329, January 2009.
[25] A. Oyama, S. Obayashi, and T. Nakamura, “Real-coded adaptive range genetic algorithm applied to transonic wing optimization,”
Journal of Applied Soft Computing, vol. 1, no. 3, pp. 179-187,
2001. [26] S. Bandaru, R. Tulshyan, and K. Deb, “Modified SBX and
adaptive mutation for real world single objective optimization,”
IEEE Congress on Evolutionary Computation, New Orleans, LA, June 2011.
[27] Y. Yoon and Y. Kim, “The roles of crossover and mutation in
real-coded genetic algorithms,” in Bio-Inspired Computational Algorithms and Their Applications, S. Gao, Ed., InTech Publisher,
2012, pp. 65 – 82.
[28] D. Goldberg, “Real-coded genetic algorithms, virtual alphabets, and blocking,” Complex Systems, vol. 5, no. 2, pp. 139–168,
February 1991.
[29] X. Qi and F. Palmieri, “Adaptive mutations in genetic algorithms,” in Proc. Second Annual Conference on Evolutionary
Programming, San Diego, CA, February 1993.
[30] R. Koch, The 80/20 Principle : The Secrete to Success by Achieving More with Lees. Currency Doubleday, New York,
1998.
[31] B. Li and W. Jiang. “Heuristic genetic algorithm using 80/20 rule,” in Proc. IEEE International Conference on Industrial
Technology, Shanghai, December 2-6, 1996.
[32] P. Srivastava and T. Kim, “Application of genetic algorithm in software testing,” International Journal of Software Engineering
and Its Applications, vol. 3, no. 4, pp. 87-95, October 2009.
[33] Y. Ho, “Heuristics, rules of thumb, and the 80/20 proposition,” IEEE Transactions on Automatic Control, vol. 39, no. 5, pp.
1025-1027, May 1994.
Anil S. Mokhade received Bachelor of Engineering degree in Mechanical Engineering
from Visvesvaraya National Institute of
Technology (formerly Visvesvaraya Regional College of Engineering), Nagpur, Maharashtra in
India in 1986 and Master of \ Technology degree in Computer Science and Engineering from Indian
Institute of Technology, Mumbai, India in 1988.
He has worked in industry for 2 years as Software Engineer and Scientist-B and for 3 years in government as Dy. CEO. He is currently
working as Asst. Professor in Dept. of Computer Science in VNIT, Nagpur. He has co-authored a book ‘Operating Systems’, New Delhi,
Macmillan Publishers India, 2008. His areas of interests are Software
Engineering, Software Architecture, Genetic Algorithms and Optimization, Operating Systems, Business Information Systems and
Business Analysis. He is currently perusing his doctoral research work in Genetic Algorithms.
Mr. Mokhade is Professional Member of ACM and Life Member of
Computer Society of India.
Omprakash G. Kakde received Bachelor of Engineering degree in Electronics and Power
Engineering from Visvesvaraya National Institute of Technology (formerly Visvesvaraya Regional
College of Engineering), Nagpur, India and
Master of Technology degree in Computer Science and Engineering from Indian Institute of
Technology, Mumbai, India in 1986 and 1989 respectively.
He received Ph.D. from Visvesvaraya National Institute of Technology,
Nagpur, India in 2004. His research interest includes theory of computer science, language processor, image processing, and genetic algorithms.
He is having over 22 years of teaching and research experience. Presently he is Director, Veermata Jijabai Technological Institute, India.
He is the author or co-author of more than thirty scientific publications
in international journals, international conferences, and national
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 76
conferences. He also authored five books on data structures, theory of computer science, and compilers. He also worked as the reviewer for
international and national journals, international conferences, and national conferences and seminars.
He is the life member of Institution of Engineers, India.
Journal of Industrial and Intelligent Information Vol. 2, No. 1, March 2014
©2014 Engineering and Technology Publishing 77