Chapter 5

Modern Optimization Techniques

5.0. Introduction

Nowadays, a variety of search algorithms are emerging in order to solve optimization

problems as discussed in last chapter, effectively and efficiently, which are commonly
known as modern optimization methods. Mathematical Optimization or programming is
basically a study for finding solutions with limited resources present for real world
problem [316]. To achieve certain objectives optimizing different parameters, viz. time,
quality, profit, money etc. under various constrains have always been a hard task.

5.1. Types of Optimization Problems

Most of the optimization problems are framed in the generic form [317] as follows;

 Minimization problem, i.e. suppose a function, represented as fi(x) has to be

minimized to a certain specified value, where i is the number of objectives and x
denotes the variables on which the function is dependent.

 Maximization problems, i.e. the function, fi(x) has to be maximized up to a desired

value, depending on the aim of the given problem.

Further, the functions, i.e. fi(x) pertaining to maximization or minimization can be

classified as shown below;

 Single Objective Function: Here, the function representing the maximization or

minimization problems consists of only one objective, i.e. i 1 for this type of
problem. This signifies that the minimal or maximum value can be achieved on
the basis of one objective.

 Multi-Objective Function: In these types of problems the function depends on

more than one objective. To achieve the required maximum/minimum value all
97
 the objectives should change accordingly. The function operates by considering
i  2,3,......n , where n is the number of objectives.

On the basis of the number of objectives related to problem, the function fi(x) can be
categorized as:

 Single Variable based Function: The function generally operates on the basis of
certain variables, called design or decision function variables. This type of
functions can be represented as fi(x), where x 1 ; i.e. the number of variable is 1.

 Multi Variable based Function: Here, the design or decision variables present in a
function is more than one. The functional value changes with the change in values
of more than one variables, i.e. fi(x,y,z,….m), where m is the number of variable
which is more than one.

The optimization problems can again be grouped into different types on the basis of the
constraints or limitations under which it operates. These are as follows;

 Unconstrained Optimization Problem: If a problem is solved without any types

of limitations or restrictions, then these types of problems are called as un-
constrained optimization problem.

 Constrained Optimization Problem: The problems, which are subjected to

certain constrains during search for their optimal solution, are generally
known as constrained optimization problem. Among the constrained
optimized problems, there are again two types of constraining factors or
ranges. These are classified as equality constraining function and inequality
constraining functions.

The optimization problems can be again identified as linear optimization problem and
non-linear optimization problem. The problems are defined as linear optimization
problem whose objective function is a linear one, and the constraining function may or
may not be linear. On the other hand, the non-linear functions or problems are defined as
those functions where both the objective function and the constraining functions are non-
linear in nature.

98
5.2. Process of Optimization

The search for an optimal solution for any kind of problem can be carried out in many
different ways [318]. These ways can be broadly classified as;

5.2.1 Deterministic Algorithm

The process follows the same search path, considering the same values of both decision
variables and objective functions. This makes the procedure highly rigorous and
repeatable in nature.

5.2.2 Stochastic Algorithm

The search process involved in this algorithm is mainly randomized. Thus the number of
solutions becomes different at each time for every run. Though the final results attained
are generally different but it doesn’t vary a lot with a wide margin. Moreover the paths of
each individual are not exactly repeatable.

5.2.3 Hybrid Deterministic-Stochastic Algorithm

Combining the advantages of both deterministic and stochastic methods, the hybrid
algorithms are developed on the basis of deterministic algorithms concept with different
initial points to start, similar to that of stochastic algorithm.

5.3. Introduction to Modern Optimization Techniques

Most conventional or deterministic methods are the classical algorithm, e.g. Newton-
Raphson methods, linear programming methods etc. Newton- Raphson methods generally
optimizes the function on gradient basis, i.e. using the functional values and their
respective derivatives. On the other hand some non-gradient methods, such as Hooke-
Jeeves pattern do not involve the derivative value, but only consider the functional value.
Hill climbing method is of such type where the search of optimal solution is performed
through a well known search path. As the replication and precise results achieved by the
deterministic methods is not always desirable, similarly the stochastic methods may lead
to some extreme cases which involves extensive and time-taking search following a
blind-fold path without any hint or guidance. To deal with such extreme cases, modern

99
optimization techniques have emerged which does a random search on basis of some
plausible hints. The modern optimization techniques [319] can be classified as heuristic
algorithm, where ‘heuristic’ means to find or discover certain value or solution by trial or
error. The trial and error process generally yields quality solutions in a reasonable amount
of time for any hard optimization problems. Though these methods can attain possible
solutions within a given finite time but it never assures its performance in achieving the
best solution. In focus to achieve more better and accurate solutions, meta-heuristic
algorithms have been developed by making some improvements over heuristic
algorithms. The Meta-heuristic algorithms are found to perform better than the heuristic
algorithms as its search process involves a higher level, i.e ‘meta’. The modern meta-
heuristic algorithms work on the basis of certain randomization of the search process. The
trial and error procedure based on randomization provides a good and flexible way to
move from local searched solutions towards global search scale.

The two major factors on which the modern optimization techniques operates are namely
exploitation or exaggeration and exploration or investigation [320]. The exploration
attribute deals in focusing the search for solutions over a local region by exploiting and
reforming the information gathered from the current solution of that region. On the other
hand, exploitation feature helps to generate diverse solutions for any optimization
problem. This helps to walk around the search space over a wide and global range
consequently helping the solutions for not getting trapped at a local optimal value. The
combination of these two factors makes the operating principle of modern optimization
technique efficient and effective in finding best optimal solutions.

5.4. History of Modern Optimization Techniques

Trial and error approach of problem solving has always been a very common approach by
human during early periods, i.e history. Most of the discoveries were done on the basis of
trial and error. Beside its universal nature, modern meta-heuristic methods are mainly
used to solve many problems nowadays. During Second World War, heuristic algorithm
was first used by Turning to break German Enigma chippers at Bletchley Park [321].
Finally in 1945, turning implemented his intelligent techniques for automatic computing
engine (ACE) in National Physic laboratory, UK [310]. Further in 1948 many algorithms

100
were implemented as intelligent machinery on the basis of heuristic or meta-heuristic
concepts. This resulted in further developments of modern methods on the basis of neural
networks and evolutionary algorithms. During early 60s and 70s development of
evolutionary algorithms emerged resulting to new methods of solutions for various real
life optimization problems [323]. Holland studied adaptive natures of evolutionary
algorithms, viz. genetic algorithm and implemented crossover and recombination
principles in improved algorithms in 1965. In 1975 Jong explored the power of
evolutionary algorithms over wide range of objective function including noisy and/or
discontinuous type [319].

Based on the concept of Darwinian evolution and natural biological selection systems, the
search based algorithms are serving many problems such as planning, data-fitting and
scheduling. In focus to solve problems in aerospace engineering, Rechenberg and
Schwefel developed certain modern methods in 1963 which was known as ‘evolutionary
strategy’. During early 1960s, Fogel studied artificial intelligence by using simulated
evolution as a learning tool [324]. Bienert further used only the rules of selection and
mutation to produce better solution. This algorithm which excludes the crossover
operation in the evolutionary process was named as simple trajectory style hill-climbing
algorithm. Fogel et al. doing further research developed many evolutionary algorithms or
techniques which were used to wider variety of problems. The concept of neural network
emerged around 1943. Artificial neurons were first proposed by Culloch et al. during this
period [325] while latter in 1948 it was implemented as intelligent machinery. A
trajectory-based search algorithm was development as a modern optimization technique
by Kirkpatrik et al. in 1983 [325]. The algorithm further works, i.e. finds better solutions
on the basis of acceptance of probability, this modern technique was named as simulated
annealing. During late 80s the improvement over meta-heuristic method was performed
by Glover, which later was recognized and implemented as Tabu search [327]. On the
concept of neural algorithm Dorigo developed as search method called as ant colony
optimization in 1992 [317]. In addition, Koza made sufficient progresses in computer
programming. During mid 90s i.e. 1995 another modern algorithm was developed on the
concept of swarm intelligence. This modern technique is based on multiple agents. The
nature of statistical learning theory was developed and proposed in1995 [328]. During
late 90s, vector based evolutionary algorithms were extended and developed by Storn et
al. The algorithm named as differential algorithm was applied to many problems [329]. In
101
1997, Wolpert et al. proposed a robust algorithm for optimization problems. This
algorithm named no free lunch theorem was specially developed for hard non-linear
programming problems [330]. Towards the beginning of 21st century Geem et.al proposed
harmony search algorithm [331]. It found its application in various problems such as
water distribution, scheduling, transport modeling and various other optimization
problems during 2001. In search for solutions regarding internet hosting, Nakrani
developed an algorithm, known as honey bee algorithm in 2004 [332]. Further
development over this algorithm was made by Pham et.al in 2005, which is commonly
known as novel bee algorithm [333]. During this period artificial bee colony (ABC) has
been introduced by Karaboga in 2005. Yang in 2008 further developed an algorithm,
called as firefly algorithm, which are widely used and appreciated in several applications
[334]. Further in 2009 the author proposed and exhibited an effective and efficient
algorithm known as cuckoo search algorithm [335]. Following this algorithm the author
again introduced another effective algorithm for continuous optimization problem on the
basis of bat’s characteristics. This was named as bat-inspired algorithm in 2010 [336].
During 1940s to 1990s most of the developments regarding modern optimization
techniques were carried out.

5.5 Basic Concept behind the Modern Optimization Techniques

The concept which decides the operational behavior of modern heuristic and meta-
heuristic algorithm algorithms is mainly based on randomization. The random search for
solution, which includes both exploration and exploitation, can be classified on the basis
of operation as follows;

5.5.1. Random Variables

An expression pertaining to the result of events associated with a random process is

generally recognized as a random variable. Random variables always generate real values.
Moreover, depending on nature of problems, the random variables are chosen
accordingly, which can be classified as discrete random variables, i.e. the values are real
discrete values, continuous random variable, i.e. when variables are generated in a given
interval; and mixed random variables, i.e. where both discrete and continuous random
variables are generated [337].
102
5.5.2. Random Walks

A random walk constitutes the occurrence of a series of consecutive random steps. This
random steps or walks are done on the basis of random distribution. This sometime
allows and helps to find the exact path or trajectory towards the desired optimal solutions.
The random walk generally follows the rule represented as St 1  St  wt , where St+1 is the

current position at time t, and wt is a step or random variable with a known distribution
[334]. The random walk can vary linearly in step size, as well as with situation or
position, i.e. 2D or 3D. Though the step size is usually considered to be fixed for simple
optimized problem, but it is dependent on following the distribution property. For
example the random walks obeying Gaussian distribution is generally termed as
Brownian motion.

5.5.3. Markov Chain Rule

The random walk which pursue or follow a particular method in choosing the step size or
step length for locating the optimal solutions. This rule depends on the sum of each
consecutive random step using a random distribution which is given as [338];

n 1
S n    xn  S n 1  xn (5.1)
i 1

Where Sn is the current state of the random walk, where xi is a random step, Sn-1 is the
previous state. Thus the next state or position depends on the current existing location.

5.5.4. Levy Distribution Rule

The Levy distribution rule is one of the important concept on the basis of which the
random walks in modern optimizing search techniques are performed. In this rule the
random walk operation is carried out on the basis of Levis distribution. This follows the
1 
simple power law formula [339], given as; L( s)  s , where β is an index whose value

lies between 0 to 2.

5.5.5. Markov Chain Monte Carlo

103
Further research over Markov chain emerged a new rule, named Markov Chain Monte
Carlo. This rule dictates the nature of the random walk. This law attempts to draw
samples directly from a highly complex multi-dimensional distribution using Markov
Chain using known transition probability. Markov Chain Monte Carlo guide the a
objective function, say f(x) where the function has to be minimized at x  x1 . This can be
performed by converting the function into a target distribution for a Markov chain, as
shown by [340]

Y ( x)  e  f ( x ) (5.2)

Where β > 0 is a parameter which acts as a normalized factor. The value of this parameter
β is set so that the probability is close to 1 when x tends to x1. The Markov chain Monte
carlo care rule formulates a general structure which shifts the objective function into a
large constant, say K > 0 in such a manner that if required f(x) tends to f ( x)  K .

5.6 Types of Modern Optimization Techniques

Currently, different heuristic methods are emerging for optimization problems. The
modern methods of optimization, includes genetic algorithms, simulated annealing,
particle swarm optimization, differential evolution, ant colony optimization, bee-colony
optimization, harmony search algorithm, bat-algorithm, Cuckoo search, neural network-
based optimization, and fuzzy optimization. These methods operate by considering the
issues such as imprecision, partial truth, uncertainty, and approximation which are often
present in few real world problems.

5.6.1 Genetic Algorithm

Genetic Algorithm was emerged during 1960s and 70s as a biological evolution based
model inspired by Charles Darwin’s theory of natural selection [341]. The operating
principle of genetic algorithm (GA) operates by starting an initial solution population of
random strings denoting several variables as vectors. Then each string or variable vectors
values are used to evaluated the objective function and find its fitness value. The
solutions are next updated through three probabilistic operators namely, reproduction,
crossover, and mutation and produces a new population vectors. The new population is

104
further evaluated to find the fitness values and tested for the convergence of the process.
If the convergence criterion is not satisfied, the population is iteratively operated by the
three operators and the resulting new population is evaluated for the fitness values.

5.6.2 Simulated Annealing

The simulated annealing method is based on the process of slow cooling of molten metal
to achieve the minimum function value of any type of problem [342]. The cooling
phenomenon of the molten metal is done by introducing a temperature-like parameter and
controlling it using the concept of Boltzmann’s probability distribution. This is because if
the temperature of a molten metal is reduced at a very fast rate, it may attain to a higher
energy state compared to that of the required crystalline state. Similarly rapid cooling in
the algorithm when applied to other engineering problems may introduce defects the
required results. Thus the temperature needs to be reduced at a slow and controlled rate to
ensure proper solidification with a low energy crystalline state that corresponds to the
best required result. Then, according to the Metropolis criterion, the probability of the
next design point (state) Xi+1 depends on the difference in the energy state or function
values at the two design points (states) the new state or design point can be produced
using the Boltzmann’s probability distribution until it matches the accepted function.

5.6.3. Particle Swarm Optimization

Particle swarm optimization, algorithm was inspired from the behavior of swarm of
insects, such as ants, bees, wasps, or a school of fish or a flock of birds. The algorithm
operates through the behavioral intelligence of each solution/particle or insect in a swarm/
function [331]. At first each particle is generally assumed to have two characteristics,
namely a position and a velocity. Depending on the intelligence one particle discovers the
good path/result in terms of food or objective function [344]. Next, remembering the best
position in the design space the other particles communicate information related to good
positions to each other and adjust their individual positions and velocities based on the
information received on the good positions. Thus the solutions reach towards best
optimized solution.

105
5.6.4. Differential Evolution Algorithm

Differential algorithm is a stochastic search algorithm with self organizing tendency and
does not use the information of derivatives. DE treats solutions as real number strings,
thus no encoding and decoding is needed. The differential algorithm works on the
optimization problem by means of the following steps. At first, a population size of n
solution vectors are initially generated for the optimization problem having d parameters.
Each solution generated for the specified problem consists of d components in d-
dimensional search space for at any generation time t. Next, mutation is performed
following the mutation scheme. As DE is a vector based method [345], so for mutation,
three distinct vectors are chosen at any instant time of generation. On the basis of a
mutation factor or perturbation factor, which is self organized and directed, a mutated
vector is produced from the three chosen initial vectors. Further the crossover operation is
executed by crossover factor, which is controlled by a crossover probability. Depending
on the binomial or exponential probability, the crossover function produces new solutions
from the mutated ones. Thus the differential evolution generates better solutions for many
generation times, until the desired value is not reached.

5.6.5. Ant Colony Optimization

Ant colony optimization (ACO) is based on the cooperative behavior of real ant colonies,
i.e. their characteristics of finding the shortest path from their nest to a food source. In the
early 1990s Dorigo and his associates first developed this method [346]. This technique
operates by initializing an equal amount of pherome, i.e. shortest from the first layer to
the last or final layer, and end at the destination node in each cycle or iteration with
respect to other paths (i.e. randomly selecting a destination node from home node by the
ants). The optimization process is terminated if either the pre-specified maximum number
of iterations is reached or no better solution is found in a pre-specified number of
successive cycles or iterations. Finally all ants travel along the same best (converged)
path [347], at the optimum solution.

5.6.6. Bee Colony Optimization

The artificial bee colony optimization algorithm was first developed by Karaboga in 2005
[348]. It works mainly on the basis of foraging behavior of the bees. The natures of bee,
i.e. forage which means hunt or explore, were again formulated on the basis of certain
106
subject, where the bees were considered as the object. One of such variants are honey bee
algorithm. The logic behind its operation is the way in which the honey bee lives in
colony and they store the honey in their constructed colony. The pherome, i.e. a chemical
release by one bee forces or attracts all other which makes the foraging nature of all the
bees. The bee which stimulates the attack response in other bees is known as alarming
bee. The alarming bee does so only when it find a good source of food and bring it back
to its hive. This communication path when followed by all other bees is known as wagle
dance, similar to a signal system for search of optimal solution. Multiple food sources
also construct a type of bee colony algorithm, where maximum intake of nectar is made
by proper allocation of foraging of bees. In virtual bee algorithm, virtual nectar
corresponds to the objective function and the solutions are the location of the nectars
[337]. The combination of type of food, i.e. nectar concentration and the search path, i.e
the foraging or wangle dance forms the fitness of the solution. An additional attribute of
this algorithm makes the method more efficient. This feature is based on the concept of
knowing the current best solutions by every bee without returning to their hives.

5.6.7 Harmony Search Algorithm

Harmony search algorithm was first developed by Geem et.al in 2001 [331]. This
algorithm was inspired by the nature of music, i.e. the search for a perfect state of
harmony. Optimization is performed on the basis of musician’s improvisation process
towards a more audio aesthetic and harmonious tune. To reach the harmonious state three
features are considered, i.e. the pitch or frequency, timber or sound quality and amplitude
or loudness. The search process is executed considering three components, namely
playing a piece of music from the harmony music memory, tuning or adjusting the pitch
of the music and composing a new random note of music. Initially, a set of solutions are
generated from the harmony memory. The best solutions from the harmony usage
memory were accepted on the basis a parameter, known as harmony memory accepting or
considering rate. A low acceptance rate leads to a very slow convergence of the result,
while with a high rate the results converge instantly to a wrong solution. Thus this
acceptance factor is generally chosen on a moderate range. Next, the pitch the accepted
solution is adjusted either linearly or non-linearly. This is made on the basis of a specified
bandwidth, which controls the local range of pitch of the accepted solutions. As low range
107
of bandwidth hardly makes any change over the accepted solutions, similarly for a chosen
very high range will never provide a converged solution. So, the pitch adjusting factor
generally set to lie within a moderate range. Finally, a randomization is performed over
the adjusted solutions in order to drive the solutions from local optima to global search
space [350]. Here the randomization process is performed by generating a random
number and multiplying it with the summation of solution related to the lower limit and
the constraining range of the solutions.

5.6.8. Bat Algorithm

The bat algorithm is one of the most recently developed modern optimization technique
which operate utilizing the behavior of both micro-bats, i.e. the tiny bats of about 1.5 to 2
kg and macro bats, i.e. the large ones which have a wing span of about 2 m [351]. The
operation logic utilizes the echolocation nature of the bats. Bats basically emit a loud
sound pulse and listen for the echo that bounces back from surrounding objects.
Depending on this echo voice they avoid the obstructions and find their rooting services
in search of prey. On the basis of certain properties of their pulses emitted and the time
for detecting the echo by their ears, the bats are able to detect the distance, orientation,
type of victims, viz. small insects, and their moving speed [352]. The optimization
process developed on the basis of echolocation characteristics operate through following
steps. At first, the bat senses a distance between the difference between the victim and
themselves. On this basis the bats fly randomly to a position with an initial velocity. This
velocity generally consist of three components, namely, a fixed frequency, suppose fmin, a
varying frequency, say f with a fixed wavelength λ, and a loudness factor in relation to the
sound through with they hunt the insects, generally the controlling factor which is varied
in finding the optimal result. Further, new solutions are generated on the basis of all these
factors represented as f min  ( f max  f min ) f  . The product f λ is generally known as the

velocity increment factor, which helps to find better solution depending on the type of
problems [353]. Again for exploring more solutions towards global search space from the
local one, the obtained solutions are added with the loudness factor.

5.6.9. Cuckoo Search Algorithm

Cuckoo search is the most recent nature-inspired metaheuristic algorithm developed by

Yang et al. in 2009 [334]. The algorithm operates on the brood parasitism of cuckoo

108
birds, i.e. the way of hatching their eggs. This phenomenon is based on levy flight
technology. The problems optimized by this rule follow an analogy with the brood
hatching features of cuckoo, as each egg in the nest represents each solution. Next, new
set of solutions are generated by levy flights, i.e. the new solutions are randomly
generated in a search space on the basis of levy distribution. In levy flight the random
walk is performed by the individual variables to find better solutions, whose random step
length is drawn by levy distribution [354]. Some set of new solutions are also generated
by levy distribution considering the best solutions. Further, the newly generated solutions
are then selected on the basis of the fitness function of the given problem. The potentially
better solutions then replace the not so good solutions after every generation time.

5.6.10. Fuzzy System

In many real-world problems, the design data, objective function, and constraints are
stated in vague and linguistic terms. Fuzzy theories can be used to model and design
systems involving vague and imprecise information [355]. The objective and constraint
functions are characterized by the membership functions in a fuzzy system, where
membership in a classical subset, of a classical crisp sets of objects. The membership
function can be viewed as valuation set obtained by multiplying a characteristic function
with the generic elements of the set of objects [356]. The fuzzy set is then characterized
by that set where all pairs of points are allowed to be the whole interval of valuation set.
Moreover, the closer the value of the points to 1, the more it belongs to the generic
elements.

5.6.11. Neural Network Based Optimization

A neural network is a massively parallel network of interconnected simple processors

(neurons) in which each neuron accepts a set of inputs from other neurons and computes
an output that is propagated to the output nodes [357]. In the presence of massive amount
of sensory data parallel computation of data is done, i.e. each neuron is considered as a
separate processor operating in parallel with other neurons. It mainly operates on the basis
of the network connectivity, the weights associated with the interconnections between
neurons, and the activation function of each neuron. The network connectivity is made by
mapping an input vector from one space to another. Considering a single neuron which
receives a set of n inputs from its neighboring neurons and a bias whose value is equal to

109
1, and the weight corresponding to each neuron wi, the weighted sum of the inputs are
calculated. This weighted sum (ws) of inputs determines the state or activity of a neuron.
Finally The output of a neuron is produced on the basis of the state activities along with

gain, i.e. basically the slope of a sigmoid function, i.e. f a  1 [358]. Finally, no
1  ews 
output will be further produced unless the activation level of the node exceeds a threshold
value.

From the above discussion of modern optimization techniques, it has been observed that
excluding the genetic algorithm and differential evolution almost all the other method
operates on the basis of the smart behavior and intelligent search in the state space.
However the optimal solutions are tried to be achieved in almost all types of optimization
technique. The behaviorally inspired algorithm such as PSO or Ant Colony optimization
algorithm, or Bee Colony optimization algorithm requires enough time to reach the
optimal solutions as it involves different paths without any updation in the values of
solutions. Moreover in fuzzy theory the membership function is formed by selecting a sub
set from the crisp set of values on the basis of some vague and linguistic factor which
might eliminate some possible good solutions from the crisp set. Other types of
optimization technique, like neural network, harmony search have very complex
operating principle which unnecessary makes search process complicated and slow.
Genetic algorithm or Differential Evolution is the algorithms which involve evolutionary
procedure within each iteration in search of better solutions. Furthermore, cultural
algorithm, as an improved version of genetic algorithm is expected to lead towards better
solutions.

Among many evolutionary methods, Genetic algorithm technique is found to be effective

on various optimization problems and is most appropriate and functionally excellent
[359-360] for the problems where locating the global optimum is a difficult task. It allows
the developer to explore and modify the control variables in a largely feasible variable
space, instead of having a precise domain expertise. Further, modifications to the basic
genetic algorithm can also be made to develop new methods for exploring better results in
such optimization problems.

Further, some fast evolutionary based optimization techniques are unable to reach the
global optimum results, so to incorporate such problems; several other optimization

110
techniques can be explored. Simulated annealing technique generally encompasses an
ability to avoid the trapping in local minima [361]. In context to get improved results
different characteristic of different optimizing techniques can be mixed.

Differential Evolution (DE) technique is also an evolutionary method which helps to

achieve the true global optimum solution using few control parameters regardless of
initial parameter values [362]. Moreover, DE is able to mitigate the drawback of long
computational time, so this algorithm seems to be a promising approach for engineering
optimization. Modifications to the basic DE algorithm can also be made to overcome
occasional problems of premature convergence and deterioration in the performance with
the growth of dimensionality of the search space for exploring better results in such
optimization problems. An improved performance can be predicted by tuning the control
parameters of basic DE as shown in Reference [363]. However, application of such
technique, where the initial parameter values do not affect much in control and optimizing
process, to solve economic problems that include microgrid and DER is sparse in the
literature.

In comparison with other evolutionary techniques, cultural algorithm [364] is a high level
searching technique. Cultural Algorithms support an additional mechanism for
information extraction during the execution of the algorithm, avoiding the need to encode
the information apriori. It reduces the need for every individual to waste energy for
acquiring domain or constraint knowledge by bypassing trial and error iterations. By
adding domain knowledge, it provides a mechanism that can extract knowledge from the
problem during the evolutionary process and thus accelerate its convergence. This process
of extraction and use of the information has been shown to be very effective in decreasing
computational cost while approximating global optima, in unconstrained, constrained and
dynamic optimization [365].

In focus to exploit the better features of evolutionary-based algorithm and consequently

exploring better results, TGA, GASA, MDE and RVCA have been identified as soft
computing techniques that can be applied to microgrid – DERs based optimization
problem. The relevant methods are discussed in detail in the next chapter. It also
describes the formulation for the present problem.

111