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

net/publication/322529809

Particle Swarm Optimization from Theory to Applications

Article in International Journal of Rough Sets and Data Analysis · January 2018
DOI: 10.4018/IJRSDA.2018040101

CITATIONS READS

41 9,638

2 authors:

M. A. El-Shorbagy Aboul Ella Hassanien

Prince Sattam bin Abdulaziz University Cairo University
108 PUBLICATIONS 1,182 CITATIONS 1,182 PUBLICATIONS 19,989 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CERTIFICATES View project

Blockchain in space industry View project

All content following this page was uploaded by M. A. El-Shorbagy on 12 December 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Particle Swarm Optimization

from Theory to Applications
M.A. El-Shorbagy, Department of Basic Engineering Science, Faculty of Engineering, Menoufia University, Shebin El-
Kom, Egypt
Aboul Ella Hassanien, Department of Information Technology, Faculty of Computers and Information, Cairo University,
Giza, Egypt

ABSTRACT

Particle swarm optimization (PSO) is considered one of the most important methods in swarm
intelligence. PSO is related to the study of swarms; where it is a simulation of bird flocks. It can be
used to solve a wide variety of optimization problems such as unconstrained optimization problems,
constrained optimization problems, nonlinear programming, multi-objective optimization, stochastic
programming and combinatorial optimization problems. PSO has been presented in the literature
and applied successfully in real life applications. In this paper, a comprehensive review of PSO as
a well-known population-based optimization technique. The review starts by a brief introduction to
the behavior of the PSO, then basic concepts and development of PSO are discussed, it’s followed
by the discussion of PSO inertia weight and constriction factor as well as issues related to parameter
setting, selection and tuning, dynamic environments, and hybridization. Also, we introduced the
other representation, convergence properties and the applications of PSO. Finally, conclusions and
discussion are presented. Limitations to be addressed and the directions of research in the future are
identified, and an extensive bibliography is also included.

Keywords
Optimization Problems, Parameter Setting, Particle Swarm Optimization, PSO, Swarm Intelligence

INTRODUCTION

Particle swarm optimization (PSO) is considered one of the evolutionary computational algorithm
which is depend on intelligence of the swarm. It is proposed by Kennedy and Eberhart (1995a; 1995b)
where it has been simulated from the artificial livings research. Also, it is a population based optimizer.
The PSO mechanism is started by randomly initializing a set of potential solutions, then the search
for the optimum is performed repetitively. In PSO algorithm, the optimal position is found by follow
the best particles. By Comparison with evolutionary algorithms (EAs), PSO has profound intelligent
background which can be performed by more easily. Due to PSO advantages, it is very suitable for
research of science, and also in real life problems such as: evolutionary computing, optimization
and many other applications. PSO belongs to the swarm intelligence group (Kennedy, Ebherhart,
& Shi, 2001) to solve global optimization problems. It was proposed originally as a model of social
behavior; where it was firstly proposed as a method of optimization in 1995 (Kennedy & Ebherhart,
1995a). In other words, we can say that PSO is associated with artificial life especially to theories of
swarming, simulation of social behaviors and also to EAs.

DOI: 10.4018/IJRSDA.2018040101

Copyright © 2018, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.


1
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

PSO can be implemented easily, inexpensive computationally, required low memory and low
speed for CPU (Eberhart, Simpson, & Dobbins, 1996). Furthermore, it does not need any information
about the objective function such as gradient, but only needs its value. It was proved that PSO is an
effective method for solving many global optimization problems and, in some cases, it does not face
the difficulties faced by other EA.
In this paper, a comprehensive review of PSO is presented. The review starts by a brief introduction
to the behavior of the PSO, then basic concepts and development of PSO is discussed, it’s followed
by discussion of PSO inertia weight and constriction factor as well as issues related to parameter
setting, selection and tuning, dynamic environments, and hybridization. In addition, we introduced
the other representation, convergence properties and the applications of PSO. In addition, conclusions
and discussion are presented. Finally, we identify the limitations to be addressed, the directions of
research in the future and an extensive bibliography is also included.

THE BASIC OF PSO

To understand the PSO, we can imagine a swarm of bees passing through an open field that contains
wildflowers. The swarm has a natural desire to locate the highest density position of the flowers in
this field. The swarm hasn’t any previous knowledge for the field. So, the bees start their search and
spread out in random locations. Every bee can remember its locations that have most flowers and
this information can be transmitted by communication to the rest of the swarm. With the passage of
time, the bees are torn between returning to its previously successful position in finding flowers and
heading toward the location (having the most flowers) that reported by the rest of the swarm. The
hesitating bee moves in both directions, converting its path to fly somewhere between the two points
based on whether social influence dominates its decision or not. Occasionally, a bee may fly over a
place in the field that has more flowers than had been discovered by any bee in the swarm. Then, the
whole swarm would be drawn as a part in the direction of that location (see Figure 1).
In the figure, dashed lines trace the paths of our imaginary bees, and the solid arrows show their
two velocity vector components. The bee No.2 was found the global best position. While, the bee
No.1 has found one component of his velocity vector (its personal best location), while the other
component is the global best position. The bee No. 3 shows that an example of a particle that have
not found a good personal best, it is still drawn to the same position of global best.
By this manner, the field is explored by the bees by updating its speed and direction of travel
depending on its successful at finding flowers in comparison to the rest of the swarm. continuously,
they are checked the places to move away from that give lowest concentration of flowers, expecting to
find the places that give highest concentration. Finally, all places in the field will be explored by the
bees and they will end up by swarming around the position which has the highest flowers concentration.
Kennedy and Eberhart (1995a) gave a model describing the behavior of PSO in a computer program.
After short time, they realized that their model can be developed as an optimization algorithm and
they called it PSO. The term “particle” is used to represent fish, bees, birds, or any other type of
natural agents that have the behavior of swarm.

PSO ALGORITHM (PSOA)

The PSOA details are presented as follows: if there is a swarm of p birds or particles, every particle
representing a possible point (solution) in the problem space S. Kennedy and Eberhart (2001) originally
proposed that the position x i , per each particle i, is evolved as:

x ki +1 = x ki + vki +1 (1)

2
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Figure 1. Simulation of PSO by Swarm of Bees Searching for Flowers

and its velocity v i is calculated as:

( ) (
vki +1 = vki + c1 × r1 × pki − x ki + c2 × r2 × pkg − x ki ) (2)

The subscript k, indicates the increment of a time. pki represents the best position of the particle
i at time k so far, while pkg is the global best position in all particles in the swarm at time k. r1 and r2
are random numbers between 0 and 1. c1 and c2 represent the cognitive and social scaling parameters
respectively which are selected such that c1 = c2 = 2 to give a mean equal 1 when they are multiplied
by r1 and r2 (Kennedy, Eberhart, & Shi 2001). The using of these values make the particles overshooting
the target in the half of time. Equation (2) is used to determine the new velocity vki +1 of the i-th
particle’s, at time k, while equation (1) gives the i-th particle’s new position x ki +1 by adding its vki +1
to its current position x ki . Figure 2 shows the updating of velocity and position of PSO particle’s in
two-dimensional space.
Let us denote the so far best fitness value of the particle i at pki as fbest
i
and the so far best fitness
value of the particle i at pkg as fbest
g
. The general PSOA pseudo code can be described in Figure 3.

3
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Figure 2. Updates Velocity and Position of PSO in Two-Dimensional Space

EVOLUTION OF PSOA

The evolution of the original PSOA of Kennedy and Eberhart (2001; Suganthan, 1999) are detailed
in this section.

Introduction of Constant Inertia Weight

The first significant variation on the original PSOA is introduced by Shi and Eberhart (1998a); where
they added an inertia term w into the velocity Equation (2) as:

kmax (3)

The parameter w performs as scaling process on the particle velocity x0i ∈ S in Rn for i = 1,..., p. ,
i.e. give the particle a momentum. High values of w result in straight paths for particle with significant
overflying at the target. This leads to global search with good characteristic. While, low values of w
make the trajectories of particle erratic with a decreasing in overshoot. This leads to a good local
search characteristic. These properties are desirable for refined local search.

4
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Figure 3. The General PSOA Pseudo Code

Reduction of Linear Inertia

Also, Shi and Eberhart (1998a; 1998b) proposed reduction of linear inertia which is a new version
of the constant inertia weight. This variation falls under the linear scaling of the inertia parameter w
during the search process, usually between an initial value of 0.8 and a final value of 0.4 distributed
on a specified number of iterations.

Limitation of Maximum Velocity

This variation is proposed by Shi and Eberhart (1998b; 2000); where the velocity of each particle is
limited to a specified maximum velocity 0 ≤ v0i ≤ v max
i
for i = 1,..., p. . This situation is used to
reduce the large step size in the position Equation (1). fki is computed as a fixed fraction fki ≤ fbest
i

of the distance between the boundaries of the search domain:

i
fbest = fki (4)

where pki = x ki and fki ≤ fbest

g
respectively represent the upper and lower bounds of the domain D.

5
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Constriction Factor
A constriction factor fbest
g
= fki was introduced by Clerc (1999) into the velocity Equation (2); where
it reduces the effect of the particles velocity as the search progresses. The constriction factor pkg = x ki
value is computed as a function of the two parameters c1 and c2:

vki for i = 1,..., p (5)

x ki for i = 1,..., p (6)

Dynamic Inertia with Maximum Velocity Reduction

This evolution, introduced by Fourie and Groenwold (2000, 2002), aims to diminish the sensitivity
dependence of the problem on the parameters, which is related to the previous implementations of
inertia (Shi & Eberhart, 1998b; 2000). The approach is summarized as follows: Firstly, w0 is determined
( ) ( )
and the initial vki +1 = w × vki + c1 × r1 × pki − x ki + c2 × r2 × pkg − x ki is calculated. If no
improvement occurs after a fixed number of iterations h in the fitness values ( vk and v max ) of the
swarm i.e. v max , The domain of the swarm is decreased by reducing the inertia and maximum velocity
by two fractions γ and v max = γ (xUB − x LB ) ; where xUB as:

x LB (7)

PARAMETER SELECTION

There are a quite many of parameters that must be considered in the PSO algorithm: C F and the
population size (P). So, it might be a good idea to know how these parameters is controlled in the
model of PSOA. The aim of setting the PSO parameters is to determine how them optimize the search
domain.
The optimization of the search space is depending on setting of the PSO parameters. For instance,
the general setting, that gives sensible results on most optimization problems, can be applied, but

 ( ) ( )
rarely is optimal. C F and vki +1 = C F × vki + c1 × r1 × pki − x ki + c2 × r2 × pkg − x ki  is the useful

2
setting for the general search. While, C F = ; where ϕ = c1 + c2 , ϕ > 4
2 − ϕ − ϕ 2 − 4ϕ
depends on the search domain size and properties of the optimization function. Practically, v max is
approximately set to 10% of the average dimension size of search domain.
By these settings, getting optimal is not guaranteed for many problems. So, we must know more
information about different settings effects of PSO parameters, to select a favorable value from problem
to the other. For example, if the optimization problem has a uni-modal fitness landscape, the PSO
must be acted as a hill-climber (HC), and the parameters is setted in a specific way to make PSO an
efficient as the behavior of HC. while, if the optimization function has many peaks, we can set the
parameters in a different way to adapt the behavior of multimodal problems.

6
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

The Control Parameters c1 and c2

When setting c1 and c2, there are two important facts must be considered. The first fact is that the
relation between the two values c1 and c2 determines the attraction point pkg . If pki , every particle
( ) ( )
attracted to the mean of  f pkg ≥ f pkg−h  and α . Most implementations of PSOA used a setting
 
with c1 almost equal to c2.
In the extreme case β all particles are converted to independent HCs; where the social part
0 < α, β < 1 is 0. In the iterated HC, the best point in the neighborhood is obtained by changing
the current point, if a better point is found. This step is repeated until access to the local optimal.
furthermore, with new starting points, the process can be repeated as many times. On the other hand,
with setting wk +1 = αwk , vkmax = βvkmax PSO is flying around the point w, C F , c1, c2 , v max and
searches a neighborhood. If the better solution is found by the particle, c1 = c2 = 2 will be updated
(if w = 0.8 is not the optimum). The particle will start flying around the new v max . After that, the
particle is moved, by repeating this process, until the optimum is found, according to the HC behavior.
But, the HC particle does not know exactly when finds the optimum solution, because it doesn’t
search systematically in the neighborhood.
c p i + c2 pkg
In the other extreme case v max , the part 1 k of the particle equals 0 and the particles
c1 + c2
are attracted to the single point c1 = c2 .
Now, the swarm turn into big HC as previously described; where all particles fly around pkg ,
and change their positions with every update of pki . The neighborhood is scanned by all particles at
the same time in a parallel way, as is the case in parallel stochastic HC. If c2 = 0 , the particle i is
( )
very attracted to c2 pkg − x ki , rather than c2 = 0, , and vice a versa if pki . However, in (2001), Carlisle
and Dozier obtained on good results with pki and pki
The second fact is the magnitudes of c1 and c2. The higher magnitudes of c1 and c2, the more
acceleration. In fact, at any time k the acceleration of particle’s is obtained by pki with neglect the
inertia and constriction. So, setting the variables (c1 and c2) high enables the swarm to change their
positions rapidly. While, setting their magnitudes low makes movement and reaction of the swarm
slow. Generally, they move far from the point of attraction and will not change their directions as in
high magnitudes.

The Inertia Weight (w)

The momentum of the particle is controlled by the inertia weight w. If (c1 = 0) , only a little momentum
(
is maintained from the previous time-step. So, the direction is possible changed quickly. If c1 pki − x ki , )
the velocity concept is fully lost and the particles move in every step without any information about
the previous velocity. While, if w >1, we get the same effect of setting c1 and c2 are low: Particles
hardly change their directions, which means that they explore a larger area with a low probability for
convergence to optimum. putting pkg must be used with care, because the velocities are biased more
to exponential growth. In short, settings w near 1 facilitates global search, and settings it in the range
(0.2 to 0.5) facilitates rapid local search (LS).
The inertia weight w has been studied by Eberhart and Shi (1998b). They found that setting w
equals 0.8 is a good choice when pkg . It is a good choice in many cases, although it is based on one
test function (the Schaffer function). Also, Eberhart and Shi applied an annealing scheme for the
setting of w; where w changes from 0.9 (start of run) to 0.4 (end of run) (1999). They compared their

7
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

model results with results obtained by Angeline at pkg (Angeline, 1998), and they concluded that the
performance is improved on 4 tested functions. The strategy of decreasing w is an optimal setting for
alot of problems; where the swarm explores the search domain in the starting of the run, and turns
into LS in the ending of the run.

The Constriction Factor (CF)

As mentioned in section 4, the constriction factor model has the same effect as w, except that it also
scales the contributions from c1 >> c2 and pki with pkg . Thus, any trying to direction (i.e. the
relationship between c1 << c2 and c1 = 2.8 ) must be done with c1 and c2 with this model. However,
c2 = 1.3. works as w: Low values give little exploration and rapid convergence, while high values
facilitate much exploration and slow convergence.
One of the few theoretical contributions of PSOA comes from the mathematician Clerc (2002),
( )
who has entered the c1 (pk − x k ) + c2 pkg − x k on PSO. He has studied the system of PSO by means
of second order differential equations. However, this analysis did not model the all particles in the
swarm, but only one single deterministic particle in a one-dimensional space. Furthermore, the author
assumed w << 1 and w = 0 are static values. Despite these simplifications, there is one outcome
from the analysis. In the model of constriction, w > 0 is set as a function in c1 and c2, to guaranteed
the convergence even without v max >> 3 . An additional parameter m is introduced to control the
convergence speed of the particles to the attraction point, instead of w = 1 introduced in Equation
6.

pki (8)

The main advantage of using from pkg instead of m, is that m clear and reliable to control the
behavior of the swarm: If m is near to 0, a fast convergence is obtained (HC behavior). While, if m
is close to 1, a slowest convergence is obtained but with a high exploration degree, which is useful
in multimodal problems.

The Maximum Velocity (vmax)

( ) (
Primarily, C F was introduced to avoid divergence and explosion. With c1 pki − x ki + c2 pkg − x ki )
or w, v has become unnecessary; where the convergence at least can be reached without it (Clerc
i
k

& Kennedy, 2002). Thus, for simplifications, C F do not use by some researchers. Despite this fact,
C F limitation can improve the search. For example, if pki and pkg in one end of the search domain
and the particle is positioned in the other end, the particle will be capable to obtain a velocity of four
times. In other words, if the distance, d, from C F to v max and C F is approximately the length of the
2m
search domain, and we assume C F = where ϕ = c1 + c2 , ϕ > 4, m ∈  0, 1 ,
2 − ϕ − ϕ 2 − 4ϕ
then from Equation (2) we find that C F i.e. four times the size of the search domain, which clearly
is nonsense.
Hence, after going to the other end of the search domain, the particle is spent a long time in
evaluations and slowing its velocity before to turn around. In this time, the search is stopped, and the
PSO still uses the CPU-time on evaluating the same boundary solutions again and again. Instead of
that, it could have approached the attraction point v max in a controlled and more efficient way with

8
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

suitable C F and gaining useful information about the fitness landscape between the two corners by
sampling various solutions on its way across the search domain.

The Neighborhood Topology

The neighborhood topology is one of the important factors in PSO. The PSO presented in section 3,
uses a fully connected neighborhood topology (or gbest). In other words, each particle is a neighbor
for all other particle. Other topologies are described by Kennedy (1999) as follows:

1. The n-best topology; where every particle is connected to its nearest particles n in the space of
topological (see Figure 4). With n = 2, this becomes a circle topology and with n = swarmsize-1
it becomes a gbest/neighborhood topology.
2. The wheel topology; where only one central particle is make a connection to the others. In
(Kennedy, 1999), many of topologies were studied by Kennedy, and he concluded that the topology
has an influence on results. However, it is difficult to find these effects, when the obtained results
are analyzed.

Size of Population
The population size selection is problem dependent. The most common ranges of the population sizes
are from 20 to 50 particles. It was observed that PSO not needed population greater than other EA to
reach high quality solutions. Actually, in most optimization problems 10 particles is enough to obtain
good results, while in some complicated or special optimization problems, 100 or 200 particles may
be enough to get well results.

OTHER REPRESENTATION OF PSO

One of the limitations of the PSO algorithm is that it was designed just for real optimization. When
PSO is compared with the family of other EA such as genetic algorithms (GAs), it loses some basics of
its applicability. Many of researchers tried to introduce an extension of PSO to other representations.
In the following, we introduce these extensions of PSO.

Binary/Discrete PSOA
PSOA have been extended to tackle one of the important problems, binary/discrete problem. Firstly,
to extend the original version of PSO to binary/discrete problem, there is a critical part must be
understood: the concepts of velocity and trajectory in the space of binary/discrete problem.

Figure 4. a) Gbest Topology b) Circle Topology, and c) Wheel Topology

9
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Kennedy and Eberhart (1997) are used the velocity value of the particle as a probability to know
whether the bit v max will be in one state or zero (0 or 1). Kennedy and Eberhart squashed the velocity
v max using a logistic map v max ; where the velocity is determined by Equation (2). According to a
generated random number within [0,1], if it is greater than pki , then pkg is set to be 0, otherwise it is
set to be 1. The binary version of PSO outperforms many versions of GAs in most tested problems
(Kennedy & Spears, 1998).

Tracking Moving Extrema

Recently, a modification was proposed by Carlisle and Dozier (2000) and Eberhart and Shi (2001),
whereby a moving extrema in a dynamic problem environment could be tracked. The dynamic tracking
problems considered one of the difficult problems for all EAs, including PSO, in terms of dealing
with it. In dynamic problem, the optimization functions change over time. So, the currently solutions
which is good, may be worse in the near future. There are three approaches to handle the dynamic
tracking problems by PSOA. The first approach to handle this problem is to apply PSO without any
changing based on the belief that PSO have ability to fast convergence. The second approach, when
a change occurs in the environment optimization, is to reset and reevaluate the previous best. But,
this is a convenient for problems which has slow-changing environment. The third approach, is to
apply some rules to discover and return to the dynamic functions, i.e. re-randomize particles when
change is discovered.

Parallel PSO (PPSO)

Parallel PSO (Chen et al., 2014) aims to reduce the run time by using multiple groups of swarm,
concurrently with produce the same results. The mathematical form of the parallel PSOA can be
written as:

x ki (9)

pki (10)

pkg (11)

(
where c1 ≈ c2 ≈ 2 , vki +1 ≈ 4d , c1 pk + c2 pkg ) (c 1
+ c2 ) is the groups number, m is a positive integer,
v is the size of particles in the j-th group, x is the position of the i-th particle in the j-th group
max i

at the k-th iteration, v i is the i-th particle velocity in the j-th group at the k-th iteration,
( ) (
S v i = 1 1 + exp −v i( )) is the position of the global best between all particles of the j-th group
from the first iteration to the k-th iteration and S (v ) is the best position among x from the first
i i

iteration to the k-th iteration. There are three possible strategies, for parallel PSO, as follows:

• Strategy 1: The best particle is migrated among all particles

 ( ) ( )
vki,+j 1 = C F × vki, j + c1 × r1 × pki, j − x ki, j + c2 × r2 × pkg , j − x ki, j  to each group and mutate

( ) ( )
x ki,+j 1 = x ki, j + vki,+j 1 to replace the worse particles in each group and update fitness p g ≤ fitness pkg , j
with i = 0,..., p j − 1 for each group for every R1 iterations.

10
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

• Strategy 2: The best particle position j = 0,..., S − 1 is migrated of the j-th group to the q-th
groups to substitute some worse particles in each receiving group for every R2 iterations. Here
(
S = 2m . )
• Strategy 3: The groups are separated into two subgroups. The strategy No.1 is applied to the
subgroup 1 after every R1 of iterations, while the strategy No.2 is applied to both subgroup No.1
and subgroup No.2 after every R2 of iterations.

Quantum PSO (QPSO)

The QPSO is presented in (Yang, Wang, & Jiao, 2004) as an algorithm with good performance-
complexity trade-off. According to the biological and social principles of the swarms, QPSO is
based on the behavior of group of various animals, associated to the quantum principle of mechanics
physical. As in the quantum theory, a qubit is defined as the smaller unit that carries information.
The i-th particle is defined with quantum energy by:

pj (12)

where L is the particle length, x ki, j and vki, j .

The term pkg , j means the probability of a bit to be “0”, and has to be converted to a discrete form.
Therefore, the discrete particle p g is obtained through the statement:

If pkg , j (13)

where p g is a random value modeled through p g

As in the PSOA, QPSO algorithm has memory to store the best position values that found for
every particle ( pkg , j ) and the best global position ( p g ). From these positions, the best global and
individual quantum energy values are determined to make changes in the positions of particle.

pkg , j (14)

q = j ⊗ 2m , j = 0,..., S − 1, m = 0,...,n − 1 and S = 2n (15)

where α and β are the parameters that control the step for the q function, with
T
qi k  = qi 1 k  qi 2 k  ... qiL k   . The i-th particle energy is updated as:
 

i = 1, 2,..., P (16)

where qiL k  ∈ (0, 1) and qiL k  are represent the weight for every component of the energy, with
T
x i t  = x i 1 t  x i 2 t   x iK t  
 

11
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Hybridizing with Other Types of Algorithms

Hybrid PSO is a combination between the PSOA and other algorithms such as the EA techniques. EA
operators, selection, crossover and mutation, have been merged inside the procedures of PSOA. The
combination between the selection operation and PSO help to copy the best performance particles into
the next offspring; thus, PSO keeps the particles that have best performance in its memory (Angeline,
1998). On the other hand, the combination between crossover operator and PSOA helps the particles
to fly in a new area in the search domain as in GAs; where the information can be transferred between
two particles (Løvbjerg, Rasmussen, & Krink, 2001). The mutation operator is considered one of
the commonly evolutionary operators that used in PSO. The aim of merging mutation operator into
PSO is to rise the diversity of the particles and give PSO the capability to go to the global minima
(Miranda & Fonseca, 2002).
There are many of hybrid algorithms between PSO and other techniques such as: hybrid PSO
and GA (Ali & Tawhid, 2016; Garg, 2016; Grosan, Abraham, Han, & Gelbukh, 2005; Premalatha,
& Natarajan, 2010; Kao & Zahara, 2008), hybrid PSO and ant colony optimization (ACO) (Mandloi
& Bhatia, 2016; Shelokar, Siarry, Jayaraman, & Kulkarni, 2007), hybrid PSO and firefly algorithm
(FA) (Kora & Sri Rama Krishna, 2016), hybrid PSO and artificial bee colony (ABC) (Li, Wang, Yan
& Li, 2015), hybrid PSO and bacterial foraging (Dhillon, Lather & Marwaha, 2016), hybrid PSO
and Levy flight (Hakli & Uğuz, 2014), hybrid PSO and gravitational search algorithm (GSA) Das,
Behera & Panigrahi, 2016; Mirjalili & Hashim, 2010, hybrid PSO and Simulated Annealing (SA)
(Idoumghar, Melkemi, Schott & Aouad, 2011), hybrid PSO with chaos (Liu et al., 2005; Alatas, Akin
& Ozer, 2009; Tan, 2012; Gandomi, Yun, Yang & Talatahari, 2013) and hybrid PSO with LS (Chen,
Qin, Liu & Lu, 2005; Qin, Yin & Ban, 2010; Qu, Liang & Suganthan, 2012) etc.

PSO CONVERGENCE PROPERTIES

There are many theoretical studies have been published about the PSO convergence properties. These
studies have concluded, as in EAs, that PSOA is very affected by the selection of parameters.
Most studies are applied to the simplified model of PSO; where the swarm is containing only
one particle in one dimension. In addition, pbest and gbest particles are considered constants during
the optimization process. While, the terms ρiL k  > qiL k  , then x iL k + 1 = 1, else x iL k + 1 = 0
are set as constant values. Under these assumptions, the trajectories of particle and its convergence
have been analyzed. The PSO convergence is defined as in the following definition:

Definition 1: According to the sequences of global best as ρiL k  , the PSOA is converged iff

U (0, 1) . (17)

where p is any position in the space domain. So, Definition No.1 don’t determine whether the point
p is global or local.
Ozcan and Mohan (1998) are proposed the first study about the PSO convergence properties.
But, the model of Ozcan and Mohan did not take into account the inertia weight. They showed that
if xbest
i k  , the trajectory shape of particle is a sine wave; where the frequency and amplitude of the
 
wave is determined by parameter choices and initial conditions. Also, they concluded that the nature
of trajectory may be make the particle searching repeatedly in a region it visited before, unless other
particle in the neighborhood finds a good solution.

12
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

On the other hand, Van den Bergh (2002) proposed the same model of Ozcan and Mohan (1998)
but take into account the inertia weight of particle. He proved that, if xbest
g k  , the particle converges
 
and go to the point g
qbest   (
best   )
k  = α ⋅ x g k  + β ⋅ 1 − x g k  . W h i l e , i f
best  
i     ( )  
qbest k  = α ⋅ xbest k  + β ⋅ 1 − xbest k  i.e. α + β = 1 , the particle converges and go to the point
i i

qi k + 1 = c1 ⋅ qi k  + c2 ⋅ qbest

i    
k  + c3 ⋅ qbest k  .
g

These results were proved under the condition of c1, c2 and c3 being constant values. In addition,
Van den Bergh circulated his model using the stochastic nature of c1 + c2 + c3 = 1. and
φ1 = c1r1, φ2 = c2r2 ; where he assumed uniform distributions. In this case, he concluded that the
∞
particle converges to the point {gbestk } ; where lim gbestk = p i.e. the particle converges to a
k =0 k →∞
point between its neighborhood best and its personal best position.
To guarantee convergence, the condition 0 < φ < 4, (φ = φ1 + φ2 ) must be hold. Maybe, we
can choose any values of c1, c2 and w such that this condition is violated and the swarm still converges
1
(Van den Bergh, 2002): The swarm has convergent behavior if w > (c1 + c2 ) − 1 is near to 1.0;
2
φ1 pbest + φ2gbest
where . This means that the particle trajectory will converge all time with taking
φ1 + φ2
divergent steps occasionally.
The studies in Ozcsn and Mohan (1998) and Van den Bergh (2002) assumed that the trajectories
are not constricted. Clerc and Kennedy, in (2002), were gave theoretical analysis of particle behavior;
where they were introduced the constriction factor which aims to prevent the velocity and maintain
the particle inside the search domain.
In (2006), Sierra and Carlos Coello introduced two necessary conditions to guarantee convergence
to the global or local optimum:

1. Monotonic condition: the gbestk+1 position can be better than the gbestk position.
2. PSOA must be able to generate a solution in the optimum neighborhood with probability (nonzero),
from any solution x in the search domain.

In (2002), Van den Bergh gives a proof that the standard PSO is not a local (neither global)
optimizer. Because, although PSO achieves the monotonic condition, once the algorithm reaches to
this state: x = pbest = gbest for all particles, no further progress will be achieved. The main problem
is the swarm may be reached to this state before gbest goes to a minimum (global or local). Therefore,
the PSOA is said to be converge prematurely. Finally, we can say that the standard PSOA is not a
local (global) search technique; where it has no ensured convergence to the global (local) minimum
from an arbitrary initial point.
Also, van den Bergh offers two methods of extending PSOA to be a global search technique. The
first one, is to the generate new random solutions to force PSOA to perform a random search in an
area around the global best position to prevent stagnation. While, the second method is to propose
a multi-start PSO.
From the discussion previously provided, Sierra and Carlos Coello (2006), proved that it is
possible to guarantee convergence, by determining the flight formula parameters correctly. But, in
multi-objective optimization (MOO), this property does not guarantee the convergence to the true curve
of Pareto front, as in single optimization. In MOO case, they concluded that to ensure convergence,
we still need the two conditions (1) and (2) with change the condition (1) to:

13
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

• The solutions added to the archive of solutions at iteration c1 = c2 = c should be nondominated

with the solutions obtained in all iterations so far.

The use of w > c − 1 -dominance archiving technique (Sierra & Coello, 2006) achieves this
condition, while the normal dominance strategies do not. To ensure this condition with the normal
dominance-based strategies (Sierra & Coello, 2006), we must sure that any solution leave the archive
is dominated by the added solution. By this way, any multi-objective particle swarm optimization
(MOPSO) approach may be satisfy the condition No.1, while we must prove that it achieves the
condition No.2, to guarantee the global convergence to the true curve of Pareto front.

APPLICATIONS OF PSO

PSO has many advantages such as: easy to implement, simple in concept and few parameters to
adjust. So, PSO is applied to solve too large number of applications. PSO, like most techniques of
EAs, can be used for solving most problems of optimization. At this time, PSOA is applied to all
types problems whether its search space variables are real or binary. There are a major number of
problems that its results obtained by PSO are very competitive.

Constrained Optimization Problems (COPs)

Constrained optimization problems (COPs) is considered one of the very important applications areas
for PSO. The major troubles which we face when solving COPs is that how the constraints is handled.
There are three approaches is used to handle the constraints in COPs. The first approach is to convert the
COP to a non-COP by adding the penalty function to violate the constraints (Parsopoulos & Vrahatis,
2002). While, the second approach is to repair the infeasible solutions and keep the feasible solutions
(Hu & Eberhart, 2002). In the third algorithm (called hybrid algorithms) some information decoding
strategies is employed. Ray and Liew (2001) used a constraint matrix to handle the constraints. By
this matrix, better performer list is generated; where it used to determine the search direction for the
residual particles. In the following, we will review some of algorithms that solving COP.
Yadav and Deep (2014) proposed a new co-swarm PSO, which is obtained by combining shrinking
hypersphere PSO (SHPSO) with the differential evolution (DE). The swarm was divided to two sub
swarms; where the first sub swarms uses SHPSO and the second sub swarms uses DE. The Tests
were performed on one of the state-of-the art problems proposed: IEEE CEC 2006. The results of
the CSHPSO are compared with SHPSO and DE in a variety of fashions. Experimental results prove
that CSHPSO is considered a good new algorithm; where it can be used to solve any real COP.
Sun, Zeng, and Pan (2011) proposed an improved vector particle swarm optimization (IVPSO)
algorithm; where the constraint is handling by a technique depending on the simple constraint
preserving method. Position and velocity of all particles and its changes are expressed by vectors in
order to present the procedures of optimization by a more understanding method. The authors proposed
multi-dimensional search methodology to search in the local region and obtain new feasible positions.
By This manner (multi-dimensional search) the authors avoided neglecting any good positions in the
feasible region and improve optimization efficiency. The performance of IVPSO is evaluated by a
well-known 13 benchmark problems. Results show that IVPSO is competitive, stable and simple stable.
Elsayed, Sarker, and Mezura-Montes (2014) proposed a new self-adaptive PSOA which can be
solve a variety of COPs efficiently. This approach is used different mix of PSO variants; where each
of which has different number of particles from the current population that used for evolving. In
every iteration, the algorithm sets more particles to the better-performing variants and fewer to the
worse-performing ones. In addition, new methodology for adapting PSO parameters is proposed. The
performance of self-adaptive PSOA was analyzed and tested by two sets problems, CEC2006 and
CEC2010. The result showed a significantly better performance than the other comparison algorithms.

14
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Abd-El-Wahed, Mousa, and El-Shorbagy (2011) presented a hybrid approach combining PSO
with GA for solving COPs; where it integrates the advantages of both techniques. At first the algorithm
is begun by initializing random particles which travel in the search region. During this, integrating
PSO and GA is used in the evolution process of these particles. Secondly, a modified constriction
factor is introduced to control the velocity of the particles and restrict it. The results of various tests
have showed the efficiency of the proposed algorithm for solving COPs

Min/Max Problems
One of the important problems that was solved by PSOA is min/max problems (Laskari, Parsopoulos,
& Vrahatis, 2002; Shi & Krohling, 2002). There are 2 approaches used for solving this type of
problems. The first approach is handle the min-max problem by the same method in the minimization
problem and the maximum part is embedded in the calculation of the objective (fitness values). By
this approximation, the obtained solutions by this approach can meet the requirements of the min/
max problem (Laskari, Parsopoulos, & Vrahatis, 2002). The second approach transforms the min/
max problem to two optimization problems (minimum problem and maximum problem). Two swarms
of PSO are used for solving the two problems, respectively, but with independent run. The swarm
of minimization problem is used as a changing environment of the other swarm of maximization
problem and vice versa. In addition, the two swarms cooperate through the calculation of fitness
(Shi & Krohling, 2002).

Multi-Objective Optimization Problems (MOOPs)

In MOOPs, there are multiple objectives need to be optimize at the same time. In most MOOPs, no
one solution is found to be optimal to all objectives. But, there are a huge of alternative solutions
which are known as a Pareto front. In addition, these solutions are equal in their preference. So, there
is a difficulty in choosing between them.
MOOP is one of the important application areas which has studied by PSOA. Many of algorithms
have been designed to handle MOOPs by using PSO. In the first algorithm, that was used to solve
MOOP, is converted MOOP to a single objective optimization problem (SOOP).
The first study used to solve MOOPs by PSOA has been proposed by Parsopoulos and Vrahatis
(2002); where the weighted aggregation method was merged in PSO to find the Pareto front solutions.
Many PSOAs were proposed to solve MOOPs. A comprehensive survey for the PSOAs that are solving
MOOPs can be found in (Sierra & Coello, 2006); where different techniques have been discussed and
categorized. In the following, we will review some of algorithms that solving MOOPs.
In (Hwang, Koo, & Lee 2006), Homogenous PSO (HPSO) for MOOP is proposed; where single
global archive concept is used for determining gbest and pbest. This means that the particles have been
giving up its special identity and dealt as a member of social set. To avoid premature convergence,
HPSO needs, in the first iteration, at least two or more Pareto solutions. HPSO have been tested to
illustrate its ability to find the Pareto optimal set for MOOPs.
Cagnina, Esquivel, and Coello (2005) proposed simple multi-objective particle swarm optimizer
(SMOPSO); where it combines an elitist policy, Pareto dominance, and two schemes to keep diversity.
The major modifications that used to handle MOOPs are uniform mutation operator, an elitist policy,
and new technicality to select the gbest and pbest particles.
Hu and Eberhart (2002) proposed a modified PSO; where they used a strategy of dynamic
neighborhood, new updating for particle memory and deal with multiple objectives by one-dimension
optimization. The dynamic neighborhood PSO was applied only on optimization problems that have
two objectives. The best position pbest is updated only when the current solution is dominated by
a new solution. Based on these setting, PSOA found multiple optimal solutions successfully and
determined the Pareto front.
Parsopulos, Tasoulis, and Vrahatis (2004) proposed a new parallel version: vector evaluated
particle swarm optimization (VEPSO) for solving MOOPs; where it is inspired from the parallel

15
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

version of GA. In VEPSO, each of the objective function is evaluated by using one swarm, and the
information is transferred between the swarms to exchange their best experience.
In (Mousa, El-Shorbagy, & Abd-El-Wahed, 2012), a hybrid multi-objective EA integrates the
merits of both PSO and GA is proposed. The algorithm is started by group of random particles that
fly in the problem domain, searching for optimization. Then GA is applied to evolve these particles.
In addition, LS technique is used to improve the quality of solution; where it aims to decrease the
distance between the obtained solutions of PSO/GA by generating more nondominated solutions
between them. Many benchmark problems including the set of benchmark functions provided for
CEC09 have been reported to show the importance of this technique in solving MOOPs.
El-Shorbagy (2015) proposed hybrid approach combining trust region algorithm (TRA) and
PSOA. Firstly, the weighted method is used to convert the MOOP to a SOOP. Secondly, the SOOP
is solved by TRA in order to get some points on the Pareto frontier which are used as positions of
particles for PSOA. Finally, homogeneous PSO is applied to obtain the all nondominated solutions.
Many kinds of benchmark problems have been tested and the results demonstrated that TRA/PSOA
has the ability to solve MOOPs and generate the Pareto optimal solutions.
Meza, Espitia, Montenegro, Giménez, and González-Crespo (2017) presented multi-objective
vortex PSO (MOVPSO) as a strategy based on the behavior of a particle swarm using rotational and
translational motions. The MOVPSO strategy is based upon the emulation of the emerging property
performed by a swarm (flock), achieving a successful motion with diversity control, via collaborative,
using linear and circular movements. MOVPSO is tested through several MOO functions and is
compared with standard multi-objective MOPSO. The qualitative results show that MOVPSO behave
as expected.

Other Applications
In addition to the previous three main applications of PSO, it has been implemented successfully
for solving many of others problem that containing a lot of real life problems. In 1995, PSO was
applied to the first real life application in the field of neural networks. In this application, PSOA
was capable of train and set the weights in a feed-forward neural network as effectively ‎(Kennedy &
Eberhart, 1995a). Since then, PSO has been applied to a lot of real life applications in many fields
due to its efficiency, simplicity and nature of fast convergence. Generally, we can say that PSO can be
applied for solving most of optimization problems and real-life applications which can be described
as optimization problems. In the following, we will review some of algorithms.
In (Li, Jiao, Zhao, Shang, & Gong, 2017), a quantum-behaved discrete multi-objective PSO
algorithm is proposed for solving complex network clustering as a MOOP. In addition, the non-
dominant sorting selection operation is employed for individual replacement. The experimental results
showed that the proposed algorithm performs effectively and achieves competitive performance with
the latest methods.
In (Sethanan & Neungmatcha, 2016), a new structure of PSO is proposed; where gbest, lbest
and nbest (neighbor best position) are combined for solving sugarcane mechanical harvester route
planning (MHRP) is presented. A new encoding/decoding scheme of the particle is designed for
merging the path planning with the accessibility and split harvesting constraints. Numerical results
on many networks with sugarcane field topologies, comparing with other algorithms, showed the
efficiency of the new structure of PSO for computation of MHRP.
In (Kesharaju & Nagarajah, 2016), a binary coded PSO technique is proposed to defect detection
in armour ceramics. It is investigated in the implementation of feature subset selection and to optimize
the classification error rate. In addition, the population data is used as input to an artificial neural
network (ANN) based classification system to get the error rate, as ANN serves as an evaluator of
PSO fitness function.
In (Mousa & El-Shorbagy, 2012), an enhanced PSOA was presented for solving the problem
of reactive power compensation; where it is a combination between GA and PSO. This algorithm

16
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

begins by initialize a set of random particles which roving in the search domain, then these particles
are evolved by PSO/GA to get approximate nondominated solution. In addition, the solution quality
is improved by dynamic search technique. This enhanced PSOA was evaluated by By applying it to
one of the standard power systems (IEEE 30—bus/6—generator). The numerical results demonstrated
ability of this algorithm to get Satisfactory solutions of reactive power compensation problem.
Das, Abraham, and Sarkar (2006) presented a hybrid rough-PSO technique for Image Pixel
Classification; where them dealt with the image segmentation as a clustering problem. Every cluster
was modeled as a rough set. PSOA was used to set the threshold and relative importance of lower
and upper approximations of the rough sets. Davies–Bouldin clustering validity index was used as
the fitness function, which is minimized while arriving at an optimal partitioning.
In (Chandramouli & Izquierdo, 2006), the chaotic modeling of PSO is presented with image
classification application. The performance of this model is compared with standard PSO. Numerical
results are performed of this comparative study on images binary classes of from the Corel dataset.
The simulation result shows a considerable increase in the image classification accuracy with the
proposed chaotic PSO algorithm.

CONCLUSION

In this paper, we presented the main concepts of the population based optimization technique PSO.
Firstly, the main concepts of PSO with its basic definitions are proposed. Secondly, evolutions of
PSO, rules of parameter selection, other representation are described with convergence properties.
Finally, different applications of PSO are introduced.
From the previous survey, we can learn that PSO has many common points with EAs; where it
starts with a randomly generated particles (population) and has fitness values to evaluate it. In addition,
the particles (population) is updated and searched for the optimum with a random technique. Finally,
PSO has many strengths and weakness which is described as follows:

Strengths
PSOA gained their popularity from its straightforward concept and simplicity in computation
(Kennedy, Eberhart, & Shi, 2001). Particle swarms fundamentally used two simple formulae to
search about the optimum effectively. Also, many researchers have compared between PSO and other
techniques on a variety set of problems (Boeringer & Werner, 2004; Salman, Ahmad & Al-Madani,
2002; Zhao & Cao, 2005; Ghoshal, 2004). It proved that PSO has better performance in some problems
and able to competitive in other. PSO is a good search technique. So, many of researches is aiming
to improve the standard PSO to solve various problems. In addition, PSO has great potential which
can do further. For example, due to its similarity with EAs, PSO is integrated with many ideas of
EAs to improve it. Like many EAs, PSO has a number of parameter to adjust. On one hand, this
is beneficial for implementing adaptive systems (Kennedy & Eberhart, 1995a) and also shows the
extensibility of PSO to other specifically designed algorithms although it may not perform as well as
those algorithms. On the other hand, tuning parameters for solving a particular problem or a range of
problems can be time-consuming and non-trivial. Compared with EAs, PSO has a few parameters that
adjust in order to get acceptable performance (Hu & Eberhart, 2002). In addition, PSO is applicable
for both unconstrained and constrained problems even without transforming any of the constraints
or the objectives of a problem (Hu & Eberhart, 2002).

Weaknesses
Researchers have found many limitations that avoid the generic PSO from effectively solving certain
types of optimization problems. Although, improvements have been working on to handle these
limitations, these improvements may be not useful when used it to solve other problems. Thus, we
should store these limitations in mind when developing new PSOA to solve other problems. For

17
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

example, although PSO has the ability to converge quickly, it tends to wander and slow down as it
approaches an optimum (Vesterstrom & Riget, 2002). Owing to the premature convergence, it gets
stuck quite easily and cannot be explore the domain of search enough. This problem appears when
solving multimodal problems where the optimal solutions is multiple. Particularly, if many of these
optimal solutions are local, PSOA may be trapping into local optimal. In addition, while there are
little parameters as previously mentioned (Hu & Eberhart, 2002), these parameters open up a potential
for evolving PSOA, some of these parameters depend on the problem. Some suggested values and
experimental settings are still at trial-and-error stage (Eberhart & Shi, 2001), and it can be non-trivial
to find the right settings for individual problems.

Future Directions of Research

There are many directions of research for PSOA which can be proposed in the following points:

• Deeply statistical analysis for PSO parameters must be studied to show how it affect the
convergence of swarm to optimal solution of the problem.
• Introducing new terms to the velocity equation for the PSOA which simulating the wind speed
and the vision factor in the search domain
• Testing different chaotic maps for controlling the PSO parameters.
• Hybridizing PSO with new optimization techniques such as: Glowworm Swarm Optimization
(Eberhart & Shi, 2001), Monkey Algorithm (MA) (Zhou, Chen & Zhou, 2016), Krill Herd
Algorithm (Bolaji, Al-Betar, Awadallah, Khader & Abualigah, 2016) and A Sine Cosine
Algorithm (Mirjalili, 2016).
• Solving real life problems that more complex/larger scale to discover new limitations in PSOA.

18
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

REFERENCES

Abd-El-Wahed, W. F., Mousa, A. A., & El-Shorbagy, M. A. (2011). Integrating particle swarm optimization
with genetic algorithms for solving nonlinear optimization problems. Journal of Computational and Applied
Mathematics, 235(5), 1446–1453. doi:10.1016/j.cam.2010.08.030
Alatas, B., Akin, E., & Ozer, A. B. (2009). Chaos embedded particle swarm optimization algorithms. Chaos,
Solitons, and Fractals, 40(4), 1715–1734. doi:10.1016/j.chaos.2007.09.063
Ali A.F., Tawhid M.A. (2016). A hybrid particle swarm optimization and genetic algorithm with population
partitioning for large scale optimization problems. Ain Shams Engineering Journal.
Angeline, P. J. (1998). Evolutionary Optimization versus Particle Swarm Optimization: Philosophy and
Performance Differences. In Evolutionary Programming VII, LNCS (Vol. 1447, pp. 601–610). Springer.
Angeline, P. J. (1998). Using selection to improve particle swarm optimization. In Proceedings of the IEEE
Congress on Evolutionary Computation (CEC 1998), Anchorage, USA. doi:10.1109/ICEC.1998.699327
Boeringer, D. W., & Werner, D. H. (2004). Particle swarm optimization versus genetic algorithms for phased
array synthesis. IEEE Transactions on Antennas and Propagation, 52(3), 771–779. doi:10.1109/TAP.2004.825102
Bolaji, A. L., Al-Betar, M. A., Awadallah, M. A., Khader, A. T., & Abualigah, L. M. (2016). A comprehensive
review: Krill Herd algorithm (KH) and its applications. Applied Soft Computing, 49, 437–446. doi:10.1016/j.
asoc.2016.08.041
Cagnina L., Esquivel S., Coello C.C. (2005). A Particle Swarm Optimizer for Multi-Objective Optimization. J.
Comput. Sci. Technol., 5(4), 204-210.
Carlisle, A., & Dozier, G. (2000). Adapting particle swarm optimization to dynamic environments. In Proceedings
of the International Conference on Artificial Intelligence, Las Vegas (Vol. 1, pp. 429–434).
Carlisle, A., & Dozier, G. (2001). An off-the-shelf PSO. In Proceedings of the workshop on particle swarm
optimization, Purdue School of engineering and technology, Indianapolis, USA.
Chandramouli, K., & Izquierdo, E. (2006). Image Classification using Chaotic Particle Swarm Optimization.
2006 International Conference on Image Processing, Atlanta, GA, 3001-3004. doi:10.1109/ICIP.2006.312968
Chen, H. L., Yang, B., Wang, S. J., Wang, G., Liu, D. Y., Li, H. Z., & Liu, W. B. (2014). Towards an optimal
support vector machine classifier using a parallel particle swarm optimization strategy. Applied Mathematics
and Computation, 239, 180–197. doi:10.1016/j.amc.2014.04.039
Chen, J., Qin, Z., Liu, Y., & Lu, J. (2005). Particle Swarm Optimization with Local Search. In Proceedings
of the 2005 International Conference on Neural Networks and Brain, Beijing (pp. 481-484). doi:10.1109/
ICNNB.2005.1614658
Clerc, M. (1999). The swarm and the queen: Towards a deterministic and adaptive particle swarm optimization.
In P.J. Angeline, Z. Michalewicz, & M. Schoenauer et al. (Eds.), Proceedings of the Congress of Evolutionary
Computation, Mayflower Hotel, Washington D.C. (Vol. 3, pp. 1951–1957). IEEE Press. doi:10.1109/
CEC.1999.785513
Clerc, M., & Kennedy, J. (2002). The Particle Swarm Explosion, Stability, and Convergence in a Multidimensional
Complex Space. IEEE Transactions on Evolutionary Computation, 6(1), 58–73. doi:10.1109/4235.985692
Das, P. K., Behera, H. S., & Panigrahi, B. K. (2016). A hybridization of an improved particle swarm optimization
and gravitational search algorithm for multi-robot path planning. Swarm and Evolutionary Computation, 28,
14–28. doi:10.1016/j.swevo.2015.10.011
Das, S., Abraham, A., & Sarkar, S. K. (2006). A Hybrid Rough Set--Particle Swarm Algorithm for Image Pixel
Classification. In Proceedings of the 2006 Sixth International Conference on Hybrid Intelligent Systems (HIS’06),
Rio de Janeiro, Brazil, 26-26. doi:10.1109/HIS.2006.264909
Dhillon, S. S., Lather, J. S., & Marwaha, S. (2016). Multi objective load frequency control using hybrid bacterial
foraging and particle swarm optimized PI controller. International Journal of Electrical Power & Energy Systems,
79, 196–209. doi:10.1016/j.ijepes.2016.01.012

19
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Eberhart, R., Simpson, P., & Dobbins, R. (1996). Computational Intelligence PC Tools. Boston, MA: Academic
Press Professional.
Eberhart, R. C., & Kennedy, J. (1995). A new optimizer using particle swarm theory. In Proceedings of the
1995 6th International Symposium on Micro Machine and Human Science (Vol. 6, pp. 39–43). doi:10.1109/
MHS.1995.494215
Eberhart, R. C., & Shi, Y. (2000). Comparing inertia weights and constriction factors in particle swarm
optimization. In Proc. of the 2000 Congress on Evolutionary Computation, Piscataway, NJ (pp. 84–88). IEEE.
doi:10.1109/CEC.2000.870279
Eberhart, R. C., & Shi, Y. (2001). Tracking and optimizing dynamic systems with particle swarms. In Proceedings
of the 2001 Congress on Evolutionary Computation CEC2001 (pp. 94–100). IEEE Press. doi:10.1109/
CEC.2001.934376
Eberhart, R. C., & Shi, Y. (2001). Particle Swarm Optimization: Developments, Applications and Resources. In
Proceeding of the 2001 congress on evolutionary computation (CEC2001), Piscataway, NJ (Vol. 1, pp. 81-86).
doi:10.1109/CEC.2001.934374
El-Shorbagy, M. A. (2015). Weighted Method Based Trust Region-Particle Swarm Optimization for Multi-
Objective Optimization. American Journal of Applied Mathematics, 3(3), 81–89. doi:10.11648/j.ajam.20150303.11
Elsayed, S. M., Sarker, R. A., & Mezura-Montes, E. (2014). Self-adaptive mix of particle swarm methodologies
for constrained optimization. Information Sciences, 277, 216–233. doi:10.1016/j.ins.2014.01.051
Fourie, P. C., & Groenwold, A. A. (2000). Particle swarms in size and shape optimization. In J.A. Snyman, & K.
Craig (Eds.), Proc. Workshop on Multidisciplinary Design Optimization, Pretoria, South Africa (pp. 97–106).
Fourie, P. C., & Groenwold, A. A. (2002). The particle swarm optimization algorithm in size and shape
optimization. Structural and Multidisciplinary Optimization, 23(4), 259–267. doi:10.1007/s00158-002-0188-0
Gandomi, A. H., Yun, G. J., Yang, X.-S., & Talatahari, S. (2013). Chaos-enhanced accelerated particle swarm
optimization. Communications in Nonlinear Science and Numerical Simulation, 18(2), 327–340. doi:10.1016/j.
cnsns.2012.07.017
Garg, H. (2016). A hybrid PSO-GA algorithm for constrained optimization problems. Applied Mathematics and
Computation, 274, 292–305. doi:10.1016/j.amc.2015.11.001
Ghoshal, S. P. (2004). Optimization of PID Gains by Particle Swarm Optimizations in Fuzzy Based Automatic
Generation Control. Electric Power Systems Research, 72(3), 203–212. doi:10.1016/j.epsr.2004.04.004
Grosan, C., Abraham, A., Han, S., & Gelbukh, A. (2005). Hybrid Particle Swarm – Evolutionary Algorithm for
Search and Optimization. In A. Gelbukh, Á. de Albornoz, & H. Terashima-Marín (Eds.), MICAI 2005: Advances
in Artificial Intelligence. MICAI 2005 (p. 3789). Berlin, Heidelberg: Springer. doi:10.1007/11579427_63
Haklı, H., & Uğuz, H. (2014). A novel particle swarm optimization algorithm with Levy flight. Applied Soft
Computing, 23, 333–345. doi:10.1016/j.asoc.2014.06.034
Hu, X., & Eberhart, R. C. (2002). Solving constrained nonlinear optimization problems with particle swarm
optimization. In Proceedings of the 6th World Multiconference on Systemics, Cybernetics and Informatics (SCI
2002), Orlando, FL.
Hu, X., & Eberhart, R. C. (2002). Multiobjective optimization using dynamic neighborhood particle swarm
optimization. In Proceedings of the 2002 Congress on Evolutionary Computation, Honolulu, Hawaii (pp. 12-17).
Hwang, S. K., Koo, K., & Lee, J. S. (2006). Homogeneous Particle Swarm Optimizer for Multi-objective
Optimization Problem. ICGST Journal of Artificial Intelligence and Machine Learning.
Idoumghar, L., Melkemi, M., Schott, R., & Aouad, M. I. (2011). Hybrid PSO-SA Type Algorithms for Multimodal
Function Optimization and Reducing Energy Consumption in Embedded Systems. Applied Computational
Intelligence and Soft Computing.
Kao, Y., & Zahara, E. (2008). A hybrid genetic algorithm and particle swarm optimization for multimodal
functions. Applied Soft Computing, 8(2), 849–857. doi:10.1016/j.asoc.2007.07.002

20
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Kennedy, J. (1999). Small Worlds and Mega-Minds: Effects of Neighborhood Topology on Particle Swarm
Performance. In Proceedings of the 1999 Congress of Evolutionary Computation (Vol. 3, pp. 1931-1938).
IEEE Press.
Kennedy, J., & Eberhart, R. C. (1995). Particle swarm optimization. In Proceedings of the 1995 IEEE International
Conference on Neural Networks (Vol. 4, pp. 1942–1948). IEEE Press. doi:10.1109/ICNN.1995.488968
Kennedy, J., & Eberhart, R. C. (1997). A discrete binary version of the particle swarm algorithm. In Proc. 1997
Conf. on Systems, Man, and Cybernetics (pp. 4104-4109). doi:10.1109/ICSMC.1997.637339
Kennedy, J., Eberhart, R. C., & Shi, Y. (2001). Swarm Intelligence. Morgan Kaufmann.
Kennedy, J., & Spears, W. M. (1998). Matching algorithms to problems: an experimental test of the particle
algorithms on the multimodal problem generator. In Proc. Int. Conf. on Evolutionary Computation (pp. 78-83).
Kesharaju, M., & Nagarajah, R. (2017). Particle Swarm Optimization Approach to Defect Detection in Armour
Ceramics. Ultrasonics, 75, 124–131. doi:10.1016/j.ultras.2016.07.008 PMID:27951501
Kora, P., & Sri Rama Krishna, K. (2016). Hybrid firefly and Particle Swarm Optimization algorithm for
the detection of Bundle Branch Block. International Journal of the Cardiovascular Academy, 2(1), 44–48.
doi:10.1016/j.ijcac.2015.12.001
Laskari, E. C., Parsopoulos, K. E., & Vrahatis, M. N. (2002). Particle swarm optimization for minimax problems.
In Proceedings of the 2002 Congress on Evolutionary Computation CEC ‘02, Honolulu, HI (pp. 1576-1581).
doi:10.1109/CEC.2002.1004477
Li, L., Jiao, L., Zhao, J., Shang, R., & Gong, M. (2017). Quantum-behaved Discrete Multi-objective Particle Swarm
Optimization for Complex Network Clustering. Pattern Recognition, 63, 1–14. doi:10.1016/j.patcog.2016.09.013
Li, Z., Wang, W., Yan, Y., & Li, Z. (2015). PS–ABC: A hybrid algorithm based on particle swarm and artificial
bee colony for high-dimensional optimization problems. Expert Systems with Applications, 42(22), 8881–8895.
doi:10.1016/j.eswa.2015.07.043
Liu, B., Wang, L., Jin, Y.-H., Tang, F., & Huang, D.-X. (2005). Improved particle swarm optimization combined
with chaos. Chaos, Solitons, and Fractals, 25(5), 1261–1271. doi:10.1016/j.chaos.2004.11.095
Løvbjerg, M., Rasmussen, T., & Krink, T. (2001). Hybrid particle swarm optimizer with breeding and
subpopulation. In Proceedings of the third Genetic and Evolutionary Computation Conference (Vol. 1, pp.
469-476).
Mandloi, M., & Bhatia, V. (2016). A low-complexity hybrid algorithm based on particle swarm and ant
colony optimization for large-MIMO detection. Expert Systems with Applications, 50, 66–74. doi:10.1016/j.
eswa.2015.12.008
Marinaki, M., & Marinakis, Y. (2016). A Glowworm Swarm Optimization algorithm for the Vehicle
Routing Problem with Stochastic Demands. Expert Systems with Applications, 46, 145–163. doi:10.1016/j.
eswa.2015.10.012
Meza, J., Espitia, H., Montenegro, C., Giménez, E., & González-Crespo, R. (2017). MOVPSO: Vortex Multi-
Objective Particle Swarm Optimization. Applied Soft Computing, 52, 1042–1057. doi:10.1016/j.asoc.2016.09.026
Miranda, V., & Fonseca, N. (2002). New evolutionary particle swarm algorithm (EPSO) applied to voltage/
VAR control. In Proceedings of the 14th Power Systems Computation Conference (PSCC’02), Seville, Spain.
Mirjalili, S. (2016). SCA: A Sine Cosine Algorithm for solving optimization problems. Knowledge-Based
Systems, 96, 120–133. doi:10.1016/j.knosys.2015.12.022
Mirjalili, S., & Hashim, S. Z. M. (2010). A new hybrid PSOGSA algorithm for function optimization. In
Proceedings of the 2010 International Conference on Computer and Information Application, Tianjin (pp.
374-377). doi:10.1109/ICCIA.2010.6141614
Mousa, A. A., & El-Shorbagy, M. A. (2012). Enhanced particle swarm optimization based local search for
reactive power compensation problem. Applications of Mathematics, 3, 1276–1284.

21
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Mousa, A. A., El-Shorbagy, M. A., & Abd-El-Wahed, W. F. (2012). Local search based hybrid particle swarm
optimization algorithm for multiobjective optimization. Swarm and Evolutionary Computation, 3, 1–13.
doi:10.1016/j.swevo.2011.11.005
Ozcan, E., & Mohan, C. K. (1998). Analysis of a simple particle swarm optimization system, In Intelligent
Engineering Systems Through Artificial. Neural Networks, 8, 253–258.
Parsopoulos, K. E., Tasoulis, D. K., & Vrahatis, M. N. (2004). Multiobjective optimization using parallel vector
evaluated particle swarm optimization. In Proceedings of the IASTED International Conference on Artificial
Intelligence and Applications, Innsbruck, Austria (Vol. 2, pp. 823–828).
Parsopoulos, K. E., & Vrahatis, M. N. (2002). Particle swarm optimization method for constrained optimization
problems. Intelligent Technologies–Theory and Application. New Trends in Intelligent Technologies, 76(1),
214–220.
Parsopoulos, K. E., & Vrahatis, M. N. (2002). Particle Swarm Optimization Method in Multiobjective
Problems. In Proceedings of the ACM 2002 Symposium on Applied Computing, Madrid, Spain (pp. 603-607).
doi:10.1145/508791.508907
Premalatha, K., & Natarajan, A. M. (2010). Hybrid PSO and GA models for Document Clustering. International
Journal of Advances in Soft Computing and Its Applications, 2(3), 302–320.
Qin, J., Yin, Y., & Ban, X. (2010). A Hybrid of Particle Swarm Optimization and Local Search for Multimodal
Functions. In Y. Tan, Y. Shi, & K. C. Tan (Eds.), Advances in Swarm Intelligence. ICSI 2010. Berlin, Heidelberg:
Springer. doi:10.1007/978-3-642-13495-1_72
Qu, B. Y., Liang, J. J., & Suganthan, P. N. (2012). Niching particle swarm optimization with local search for
multi-modal optimization. Information Sciences, 197, 131–143. doi:10.1016/j.ins.2012.02.011
Ray, T., & Liew, K. M. (2001). A swarm with an effective information sharing mechanism for unconstrained
and constrained single objective optimization problem. In Proc. congress on evolutionary computation 2001,
Seoul, Korea (pp. 75-80).
Salman, A., Ahmad, I., & Al-Madani, S. (2002). Particle Swarm Optimization for Task Assignment Problem.
Journal of Microprocessors and Microsystems, 26(8), 363–371. doi:10.1016/S0141-9331(02)00053-4
Sethanan, K., & Neungmatcha, W. (2016). Multi-objective particle swarm optimization for mechanical harvester
route planning of sugarcane field operations. European Journal of Operational Research, 252(3), 969–984.
doi:10.1016/j.ejor.2016.01.043
Shelokar, P. S., Siarry, P., Jayaraman, V. K., & Kulkarni, B. D. (2007). Particle swarm and ant colony algorithms
hybridized for improved continuous optimization. Applied Mathematics and Computation, 188(1), 129–142.
doi:10.1016/j.amc.2006.09.098
Shi, Y., & Eberhart, R. (1998). A modified particle swarm optimizer. In Proceedings of the IEEE International
Conference on Evolutionary Computation, Anchorage, AK (pp. 69-73). doi:10.1109/ICEC.1998.699146
Shi, Y., & Eberhart, R. C. (1998). Parameter Selection in Particle Swarm Optimization. In Evolutionary
Programming VII. LNCS, 1447, 591–600. doi:10.1007/BFb0040810
Shi, Y., & Eberhart, R. C. (1999). Empirical Study of Particle Swarm Optimization. In Proceedings of the 1999
Congress of Evolutionary Computation. IEEE Press. doi:10.1109/CEC.1999.785511
Shi, Y., & Krohling, R. A. (2002). Co-evolutionary particle swarm optimization to solve min-max problems.
In Proceedings of the 2002 Congress on Evolutionary Computation CEC ‘02, Honolulu, HI (pp. 1682-1687).
Sierra, M. A., & Coello, C. C. (2006). Multi-Objective Particle Swarm Optimizers: A Survey of the State-of-
the- Art. International Journal of Computational Intelligence Research, 2(3), 287–308.
Suganthan, P. N. (1999). Particle swarm optimiser with neighbourhood operator. In P.J. Angeline, Z. Michalewicz,
M. Schoenauer et al. (Eds.), Proceedings of the Congress of Evolutionary Computation, Mayflower Hotel,
Washington D.C. (Vol. 3, pp. 1958-1962). doi:10.1109/CEC.1999.785514
Sun, C.-L., Zeng, J.-C., & Pan, J.-S. (2011). An improved vector particle swarm optimization for constrained
optimization problems. Information Sciences, 181(6), 1153–1163. doi:10.1016/j.ins.2010.11.033

22
 International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

Tan, D. (2012). Chaos Particle Swarm Optimization Algorithm for Multi-Objective Constrained Optimization
Problems. In Y. Zhang (Ed.), Future Wireless Networks and Information Systems. Lecture Notes in Electrical
Engineering. Berlin, Heidelberg: Springer. doi:10.1007/978-3-642-27326-1_60
Van den Bergh, F. (2002). An Analysis of Particle Swarm Optimizers [Ph.D. thesis]. Department of Computer
Science, University of Pretoria, Pretoria, South Africa.
Vesterstrom, J., & Riget, J. (2002). particle swarms – extensions for improved local, multi-modal, and dynamic
search in numerical optimization [Master’s thesis]. Department of computer science, University of Aarhus.
Yadav, A., & Deep, K. (2014). An efficient co-swarm particle swarm optimization for non-linear constrained
optimization. Journal of Computational Science, 5(2), 258–268. doi:10.1016/j.jocs.2013.05.011
Yang, S., Wang, M., & Jiao, L. (2004). A Quantum Particle Swarm Optimization. In Proceedings of the IEEE
Congress on Evolutionary Computation, CEC’04 (Vol. 1, pp. 320-324).
Zhao, B., & Cao, Y. (2005). Multiple objective particle swarm optimization technique for economic load dispatch.
Journal of Zhejiang University-SCIENCE A, 6(5), 420–427. doi:10.1631/jzus.2005.A0420
Zhou, Y., Chen, X., & Zhou, G. (2016). An improved monkey algorithm for a 01 knapsack problem. Applied
Soft Computing, 38, 817–830. doi:10.1016/j.asoc.2015.10.043

23
International Journal of Rough Sets and Data Analysis
Volume 5 • Issue 2 • April-June 2018

APPENDIX

Basic Definitions of PSO

In the following, some definitions of many terms that was used in PSO:

• Swarm: The algorithm population.

• Particle: The swarm member; where it represents the solution of the problem of optimization
which we want to solve.
• Local Best (lbest): For a given particle, the best particle position in the neighborhood is called
lbest.
• Personal Best (pbest): Personal best position of a given particle so far (i.e. the particle position
that give the greatest success).
• Global Best (gbest): For all swarm, the best particle position is called gbest.
• Leader: The particle that guide other particles in the direction of better regions in the problem
search space.
• Velocity: It is a vector used to drive the process of optimization; where It give the path of
movement of the particle to evolve its current position.
• Inertia Weight (w): This parameter is used to decrease the effect of prior velocities on the
current velocity and position of particle.
• Learning Factor: There are two learning factors c1 and c2 (they are defined as constants). c1 is
the cognitive factor; where it represents the particle attraction in the direction of its own success,
while c2 is the social factor; where it represents the particle attraction toward the success of its
neighbors.
• Neighborhood Topology: Determine the particle’s set that used to calculate the value of lbest
for a given particle.

24
View publication stats