Archives of Computational Methods in Engineering (2022) 29:2531–2561

https://doi.org/10.1007/s11831-021-09694-4

ORIGINAL ARTICLE

Particle Swarm Optimization Algorithm and Its Applications:

A Systematic Review
Ahmed G. Gad1

Received: 11 November 2020 / Accepted: 21 April 2021 / Published online: 19 April 2022
© The Author(s) 2022

Abstract
Throughout the centuries, nature has been a source of inspiration, with much still to learn from and discover about. Among
many others, Swarm Intelligence (SI), a substantial branch of Artificial Intelligence, is built on the intelligent collective
behavior of social swarms in nature. One of the most popular SI paradigms, the Particle Swarm Optimization algorithm
(PSO), is presented in this work. Many changes have been made to PSO since its inception in the mid 1990s. Since their
learning about the technique, researchers and practitioners have developed new applications, derived new versions, and
published theoretical studies on the potential influence of various parameters and aspects of the algorithm. Various perspec-
tives are surveyed in this paper on existing and ongoing research, including algorithm methods, diverse application domains,
open issues, and future perspectives, based on the Systematic Review (SR) process. More specifically, this paper analyzes
the existing research on methods and applications published between 2017 and 2019 in a technical taxonomy of the picked
content, including hybridization, improvement, and variants of PSO, as well as real-world applications of the algorithm
categorized into: health-care, environmental, industrial, commercial, smart city, and general aspects applications. Some
technical characteristics, including accuracy, evaluation environments, and proposed case study are involved to investigate
the effectiveness of different PSO methods and applications. Each addressed study has some valuable advantages and una-
voidable drawbacks which are discussed and has accordingly yielded some hints presented for addressing the weaknesses
of those studies and highlighting the open issues and future research perspectives on the algorithm.

1 Introduction For working out those problems, the performance of candi-

date solutions is usually evaluated with high-fidelity numeri-
It has been proven in the literature beyond any doubt that cal analysis approaches (e.g., computational fluid dynamics
meta-heuristic optimization algorithms perform well by simulations or finite element analysis), which may deduct
optimally handling several versatile real-world optimization CPU time from minutes to hours, or even days [80, 222].
tasks, ranging from robotics [31], wireless networks [195, Therefore, within the paradigm of meta-heuristics, many
206], power systems [40, 76], job shop scheduling [109, Swarm Intelligence (SI) techniques and variants were pro-
137], to classification [35] and training of artificial neural posed to tackle complex/large-scale optimization problems.
networks [83]. While questing the global best (near-opti- SI, a substantial branch of optimization techniques, is the
mum) solution, numerous fitness evaluations are required by characteristic of a system where agents interact locally with
the meta-heuristic algorithms. This typically poses a serious their environment so that their collective behaviors render
barrier against the application of meta-heuristic algorithms the emergence of cohesive functional global patterns. Unlike
to high-computational optimization problems that exten- Evolutionary Algorithms (EAs), SI techniques are inspired
sively exist in computational fluid dynamic optimization by agents’ plain behaviors and self-organizing interactions,
[21] and structural optimization [79], among many others. such as fish schooling, honey bees, bacterial growth, ani-
mal herding, bird flocking, ant colonies foraging, and so on.
Indeed, Beni [17] was the first to use the SI term in cel-
* Ahmed G. Gad
[email protected] lular robotic systems, in which simple agents interact with
https://ahmedgad.com neighbors to organize themselves. SI was formally estab-
lished in [18, 19]. The rife SI algorithms include Ant Colony
1
Faculty of Computers and Information, Kafrelsheikh Optimization (ACO) [37] and Particle Swarm Optimization
University, Kafrelsheikh, Egypt

13
Vol.:(0123456789)
2532 A. G. Gad

(PSO) [91]. Less widespread SI algorithms are Bacterial tune. PSO obtains the best solution from particles’ inter-
Foraging Optimization (BFO) [141], Artificial Bee Colony action, but through high-dimensional search space, it con-
(ABC) [87], Firefly Algorithm (FA) [201], and many others. verges at a very slow speed towards the global optimum.
SI algorithms were primarily developed for steady optimi- Moreover, in regard to complex and large datasets, it shows
zation problems. However, dynamic environments involve poor-quality results. If there is a large number of dimen-
several real-world optimization problems [124]. sions in the problem at hand, PSO usually fails to discover
Typically, a swarm is defined as a vast number of simple, the global optimum solution. This phenomenon is caused
homogeneous agents interacting locally with their environ- not only by the presence of local optima trap, but also the
ment, as well as themselves, with decentralized control to potential fluctuation of the velocities of particles such that
authorize the emergence of a global important behavior. the successive range of trials is bounded within a sub-plain
Swarm-based techniques have recently arisen as a family of the whole search hyper-plain [178].
of swarm-based, nature-inspired algorithms that have the The local optima issue in PSO has been discussed and
ability to produce robust, fast, and low cost solutions to diverse variants of PSO algorithm were developed to tackle
numerous complex problems [50, 69]. Therefore, SI can that issue. For example, some of these variants have been
be figured out as a major category of Artificial Intelligence developed by incorporating the capabilities of EAs (e.g., the
(AI) that is utilized to model the collective behavior of natu- adaptation of PSO parameters, hybrid versions of PSO, etc.),
ral social swarms, such as honey bees, bird flocks, and ant thereby producing adaptive PSO versions. Other researchers
colonies. These agents (swarm individuals or insects) are have taken on incorporating evolutionary operators, such as
relatively gullible with simple own capabilities. However, crossover, mutation, selection, as well as the Differential
they perform cooperatively tasks substantial for their sur- Evolution algorithm (DE) itself, into PSO. Consequently,
vival through interacting together in particular behavioral hybrid versions of PSO have been tested and produced,
manners. Socially, swarm individuals can directly or indi- including the hybrid evolutionary PSO [128], Genetic Algo-
rectly interact among themselves [202]. Direct interaction rithm (GA) and PSO [48, 116], genetic programming-based
can be through audio or visual contact (e.g., a waggle dance adaptable evolutionary hybrid PSO [154], and many others
of honey bees), while indirect interaction is evident when [62]. Such improvements perform well with PSO and have
the environment is changed by one individual and the other the potential to avoid getting stuck at local optima. How-
individuals react to the new environment (e.g., pheromone ever, the problem of premature convergence in some high-
tracks of ants that look for food sources through depositing dimensional complex problems still exists, even if the local
on their way). This indirect pattern of interaction is known optima obstacle is absent. Hence, PSO does not always work
as “stigmergy”, which denotes communication through the properly for high-dimensional models [54].
environment [42]. The research area presented in this thor- To this point, the performance of PSO has generally been
ough paper focuses on SI. More specifically, this review improved by developing different variants of the algorithm.
strives to explore one of the most popular models of SI, However, few review papers and technical surveys have sys-
PSO, which is inspired by birds’ flocking behavior. tematically addressed the PSO literature [190, 216]. Conse-
PSO is a swarm-based stochastic algorithm proposed quently, the primary objective of this work is to present a
originally by Kennedy and Eberhart [44, 91], which exploits systematic survey by reviewing the PSO algorithm and its
the concepts of the social behavior of animals like fish different methods, as well as a wide variety of indicative
schooling and bird flocking. In PSO, each potential solution application domains. Moreover, some of the addressed stud-
to a given problem is viewed as a particle with a certain ies have highlighted the main approaches of PSO applica-
velocity flying through the space of the problem just like tions, including health-care, environmental, industrial, com-
a flock of birds. Each particle then combines – with some mercial, smart city, and general aspects. The key solicitude
random disturbances – some aspect of the record of its own of those PSO methods and applications is satisfying quality
historical best location and current location with those of standards: services arising from PSO methods and applica-
one or more agents of the swarm to determine its next move- tions should support user’s requirements that cover the qual-
ment through the search space. After all particles have been ity metrics of accuracy and efficiency, such as convergence
moved, the next iteration occurs. The swarm as a whole (e.g., rate, computational cost, consistency, stability, diversity, etc.
a flock of birds collectively searching for food) is probably In other words, this paper aims to present a timely, com-
to gradually approach the objective function optimum. PSO pendious, systematic and an in-depth overview of the PSO
has eventually gained prevalent vogue amongst researchers algorithm between 2017 and 2019 and the opportunities and
and emerged to provide high performance in an assortment challenges imposed during this period.
of application areas, with the potential to hybridize and spe- The structure of this study is organized as follows.
cialize and demonstrate some appealing emergent behaviors. Section 2 presents the related works. A brief foundation
PSO has a main advantage of having fewer parameters to of the PSO is presented in Sect. 3. Section 4 exploits the

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2533

Systematic Review (SR) procedure to provide the proposed evaluation parameters, such as convergence rate, diversity,
research approach and motivation. Section 5 demonstrates accuracy, and processing time, as quality factors in this area.
the various methods of PSO based on the SR method. Sec- Kulkarni et al. [96] outlined some issues regrading Wire-
tion 6 systematically outlines and categorizes the applica- less Sensor Networks (WSNs), introduced PSO, and dis-
tion approaches in PSO. Also, a technical classification and cussed the suitability of the algorithm for WSN applications.
differentiation of the approaches in the addressed papers is This study also presented a compendious survey of how PSO
presented in this section. Section 7 provides a discussion is tailored to tackle the issues of node localization, node
on the PSO approaches that have not yet been thoroughly deployment, data aggregation, and energy-aware clustering
analyzed. Finally, Sect. 9 concludes the paper along with as optimization problems. The strong point of this review is
its limitations. to comprehensively present open issues in WSNs. However,
the compatibility of high-speed real-time applications with
each approach was not considered.
2 Related Work Kothari et al. [94] reviewed the studies related to the
modifications of native PSO and its practical application
A swift explanation is presented in this section for the gen- in real-world problems. The rapid modification of PSO has
eral related studies in the PSO algorithm. been emerged in different examples, including the two-step
Poli et al. [143] presented an overview of the great efforts PSO and the PSO-Support Vector Machine (PSO-SVM).
which have given impetus and direction to research in parti- The integration of the PSO and the practical implementa-
cle swarms, as well as some important new applications and tion of PSO with the industry standard algorithm have also
directions. An analysis of IEEE Xplore and Google Scholar generated stunning results. The advantage of this survey was
citations and publications from 1995 to 2006 were presented to present recent diverse variations in PSO and analyze the
in this work, illuminating the sense meant by Kennedy and accuracy of PSO in different domains. The main deficiency
Eberhart [92]. The strength of this study was to present com- of this study is the lack of statistical information about the
prehensive challenges and open issues in the PSO algorithm. discussed standard PSO and its application in various speci-
However, this study did not consider the compatibility of fied contexts.
PSO application with each presented approach. Imran et al. [71] discussed diverse variants of PSO in
Banks et al. [13, 14] offered, in two parts, a timely and regards to swarm initialization, mutation operators, and
brief review of the field in general, alongside the opportuni- inertia weight. The main advantage of this overview was to
ties and challenges emanating from the versatile application highlight the importance of introducing the different muta-
of PSO. On the one hand, part I [13] has considered the tion operators and inertia weight parameter to improve the
history and background of PSO and its position within the performance of PSO. However, other promising variants of
broader paradigm of natural computing. The review then PSO were not considered.
continued to discuss different improvements to the native Alam et al. [7] surveyed systematically the evolution of
formulation of PSO both in discrete and continuous prob- clustering techniques based on PSO and presented the results
lems, swarm behavior analysis, and measures considered of rapidly increasing trends in the literature of SI, PSO para-
to address stagnation. Furthermore, the review focused on digm, and PSO-based data clustering approaches, proving
research regarding adaptations for parallel implementation, that there is a tremendous increase in the popularity of such
algorithm configuration, and dynamic environments. The approaches. This research affirmed that the approaches are
achievement of this study was identifying two significant novel and simple to implement and builds on communication
areas of challenge for future further development: swarm and collaboration. This study outlined different application
stagnation and dynamic environments. The shortcoming of domains of PSO relevant to clustering. However, applica-
this part is the insufficiency of the explanation the related tions to more complex problems are absent. Furthermore,
work. On the other hand, part II [14] has discussed recent variants of PSO for clustering high-dimensional data were
studies in some of the impressive areas of research: con- surveyed by Esmin et al. [49]. The main advantage of this
strained and multi-objective optimization, combinatorial survey was to present the attempts of different publica-
problems, and hybridization. In that review, a number of tions in reducing data dimensions. The weakness is still the
domains were briefly touched upon, including the optimi- absence of discussing more clustering-related applications.
zation of neural fuzzy networks, artificial neural networks, Marini and Walczak [123] described comprehensively the
computational biology, image processing and medical imag- PSO algorithm to show its ability to solve different kinds
ing, optimization of electricity generation, network routing, of optimization problems in chemometrics. The importance
and financial forecasting. Nevertheless, the study has a of the appropriate choice of PSO meta-parameters has
main defect of not analyzing the selected studies in terms of been highlighted in this work by means of selecting prac-
tical examples in the subject areas of variable selection,

13
2534 A. G. Gad

estimating robust Principal Component Analysis (PCA) • In existing studies, the methods and applications of PSO
solutions, and signal warping. This tutorial contributed to are not organized in a clear taxonomy, and do not have
presenting works subject to chemometrics impressively. analytical assessment from which one can learn some-
However, it lacked splashing other state-of-the-art fields. thing beneficial.
Zhang et al. [216] investigated comprehensively the PSO • Important assessment factors are not involved in the
algorithm. They provided advances in PSO, including its methods and applications of PSO in some current studies.
theoretical analysis (convergence analysis, parameter tuning, • The systematic arrangement is typically absent in the
etc.), hybridization (with GA, DE, ABC, ACO, biogeogra- structure of literature studies, and the paper selection
phy-based optimization, harmonic search, Tabu Search (TS), criterion is vague as well.
Artificial Immune System (AIS), and Simulated Annealing
(SA)), modifications (including fuzzy PSO, chaotic PSO,
bare-bones PSO, quantum-behaved PSO), population topol-
ogies (including star, ring, random, von Neumann, fully con-
nected, etc.), extensions (to binary, discrete, constrained, and 3 Particle Swarm Optimization: PSO
multi-objective optimization), and parallel implementation Mechanism
(in cloud computing, multiprocessor, multicore, and GPU
forms). Moreover, they introduced a survey on the applica- 3.1 Self-Organization Features
tions of PSO to the following eight fields: biology, chemis-
try, medicine, electrical and electronic engineering, fuel and SI system has a major feature, namely, self-organization,
energy, mechanical engineering, operation research, com- in which the components of an initially disordered system
munication theory, and automation control systems. Despite interact locally to produce a coordination or global order.
all that, the publication analysis demonstrated in this survey This process is characterized by spontaneousness; that is,
lacks the presentation of the annual exponential fluctuation no agent inside or outside of the system dominates the inter-
of publications for each variant and application field. action. The self-organization in swarms was interpreted
Wang et al. [190] presented the inception and background by Bonabeau et al. [25] through three key components as
of the PSO algorithm, and carried out a theoretical analysis follows:
of the algorithm. Then, they analyzed its current situation of
application and investigation in algorithm structure, topol- • Robust dynamical non-linearity (always comprising
ogy structure, parameter selection, multi-objective optimiza- positive and negative feedback) convenient structures
tion, discrete and parallel PSO algorithm, and engineering are promotionally being created with the help of positive
applications. This overview is characterized by suggesting feedback, while this positive feedback is counterbalanced
distinctive future research directions. However, no analytic and the collective pattern is stabilized with the help of
discussion is introduced in this work. negative feedback.
Summarizing, Table 1 outlines the recent review studies • Trade-off between exploration and exploitation A valu-
on PSO with respect to the general survey and review studies able mean creativity artificial approach is provided
discussed in this section. This table lists the key topics, pub- through a suitable balance that is identified by SI.
lication year, and covered years (if available) for each study. • Multiple interactions Information coming from neigh-
The deficiencies in the existing review papers propose that bor agents in the swarm are used by individual agents,
a comprehensive and systematic literature review should be allowing information to be disseminated throughout the
provided to address some of such common weaknesses as: network.

Table 1 Relevant studies in PSO

Study Main context Publication year Covered years

Poli et al. [143] PSO algorithm, new directions, and applications 2007 1995–2006
Banks et al. [13, 14] PSO field, challenges, and opportunities 2007 & 2008 N/A
Kulkarni et al. [96] PSO suitability for WSN applications 2011 N/A
Imran et al. [71] PSO variants 2013 N/A
Alam et al. [7] PSO-based data clustering 2014 2002–2012
Marini and Walczak [123] PSO in chemometrics fields 2015 N/A
Zhang et al. [216] PSO advances and applications 2015 2000–2013
Wang et al. [190] PSO theory and application 2018 N/A

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2535

3.2 SI Features where 𝐯 represents the velocity vector, 𝜔 is the inertia weight
utilized to balance the local exploitation and global explo-
Millonas [127] proposed that SI must satisfy five basic prin- ration, 𝐫𝟏 and 𝐫𝟐 are random vectors uniformly distributed
ciples: adaptability, diverse response, stability, quality, and within the range [0, 1]D (D being the search space dimen-
proximity. Table 2 lists their meanings. sionality or the size of the problem at hand), and c1 and c2 ,
called “acceleration coefficients”, are positive constants.
3.3 Standard PSO Algorithmic Structure An upper bound is commonly set for the velocity vector.
As a means to prevent particles from shaving off the search
A swarm of particles updates their relative positions from space and forcing them to take a proper step size to comb the
iteration to another, boosting the PSO algorithm to duly entire search domain, the “velocity clamping” method was
perform the search process. To get the optimum solution, used [163]. The “constriction coefficient” strategy is another
each particle moves towards its prior personal best position method, proposed by Clerc and Kennedy [36], in which the
(𝐩best ) and the global best position (𝐠best ) in the swarm [215]. velocities can be also constricted by theoretically observing
Assuming a minimization problem, one have and analyzing the swarm dynamics.
By scrutinizing Eq. (3), we can interpret that the first part,
𝐩tbest = 𝐱i∗ ∣ f 𝐱i∗ = min
( ) ({ ( k )})
i k=1,2,…,t
f 𝐱i , (1) known as “inertia component”, represents the prior velocity,
which provides the particles with appropriate momentum
where i ∈ {1, 2, … , N}, and to rove across the search space. The second part, the “cog-
nitive component”, denotes the own positiveness for every
𝐠tbest = 𝐱∗t ∣ f 𝐱∗t =
( ) ({ ( k )})
min f 𝐱i ,
particle. It motivates the particles to move towards their own
i = 1, 2, … , N (2)
k = 1, 2, … , t best positions found so far over subsequent iterations. The
third part, known as the “social component”, indicates the
where i denotes particle’s index, t is the current iteration’s collective effect of the particles to reach the global optimum
number, f is the objective function to be optimized (mini- solution [218].
mized), 𝐱 is the position vector (or a potential solution), and
N is the total number of particles in the swarm. The fol- 3.4 PSO Pseudocode
lowing equations update, at each current iteration t + 1, the
velocity 𝐯 and position 𝐱 of of each particle i as: Let f ∶ 𝐑N → 𝐑 be an objective function which needs to be
( ) simplified. Then, the function takes a vector of N real num-
𝐯t+1 t t t
+ c2 𝐫𝟐 𝐠tbest − 𝐱it , (3) bers denoting an N candidate solutions and outputs a real
( )
i
= 𝜔𝐯 i
+ c 𝐫
1 𝟏 𝐩 besti
− 𝐱 i
number that indicates the value of the objective function.
The f gradient is either hard( to calculate or often unknown.
𝐱it+1 = 𝐱it + 𝐯t+1
i
, (4) Then, the global minimum 𝐠best is pursued as exhibited in
)

Algorithm 1.

Table 2 SI basic principles

Principle Definition

Adaptability Swarm should have the potential to change its search behavior when the computational cost is high
Diverse response Swarm should not perform its activities along extremely narrow channels
Stability Swarm should not change its search behavior in response to the environment changes
Quality Swarm should should have the potential to respond to performance measures in the environment
Proximity Swarm should should have the potential to smoothly perform time- and space-consuming computations

13
2536 A. G. Gad

Algorithm 1 PSO pseudocode

Input:
N – Swarm size
D – Problem dimensionality
T – Maximum number of iterations
LB – Lower bound of the search space
U B – Upper bound of the search space
Output:
gbest
t
– the best position (solution) found so far
1: Start
2: Initialize the swarm randomly;
3: for i = 1 to N do Iterate through the swarm
4: vi0 ← a random vector within [LB, U B]D ; Initialize particles’ velocity using a uniform distribution
5: x0i ← a random vector within [LB, U B]D ; Initialize particles’ positions using a uniform distribution
6: p0besti ← x0i ; Initialize pbest to its initial position
7: end for
0
8: Apply Eq. (2) to find gbest ; Initialize gbest to position with the minimum fitness value
9: t ← 1; Initialize first iteration number
10: while t ≤ T do
11: for i = 1 to N do Iterate through the swarm
12: r1 , r2 ← two independent vectors randomly generated from [0, 1]D ;
13: Apply Eq. (3); Update particle’s velocity
14: Apply Eq. (4); Update particle’s position
15: if f (xti ) < f pt−1
besti then If new solution is better than current personal best
16: f ptbesti ← f (xti ); Update the best known position of the particle
17: end if
18: end for
19: Apply Eq. (2) to find gbest
t
; Update the swarm’s overall best known position
20: t ← t + 1;
21: end while Maximum iteration number is reached or termination criterion is satisfied
22: End

3.5 PSO Flowchart Elsevier, Springer, IEEE, ACM, MDPI, Taylor & Francis,
Wiley, Inderscience, MDPI, etc.
The flowchart to demonstrate the flux of PSO is depicted In relation to the objectives of this SR paper, comprehen-
in Fig. 1. sive answers are provided to the following four Analytical
Questions (AQs) [156]:

4 Proposed Research Approach • AQ1: Into which domains can PSO methods and applica-
tions be classified?
This section depends on the SR method to present, in a tax- • AQ2: Which main contexts can be considered for PSO
onomy, a review of the research studies related to the meth- methods and applications?
ods and applications of PSO [71, 123, 143, 190, 216]. Other • AQ3: What assessment environments are used for evalu-
synonyms and alternatives of the key essential components ating the PSO methods and applications?
are also considered to commit the final exploration string • AQ4: What are generally the PSO open perspectives and
as [49, 94, 96]: future research directions that can be drawn from the
work presented in the paper?
• (“Methods” OR “Variants” OR “Hybrid” OR “Improved”
OR “Software” OR “Application” OR “Application layer” After outlining the AQs, the ultimate research selection
OR “Application-based” OR “App”) AND (“PSO”) OR was conducted by applying the inclusion/exclusion criteria.
(“Particle Swarm Optimization”) With respect to the number of published research papers,
only conference papers and journal articles indexed in Web
Figure 2 demonstrates the distribution (from 2017 to 2019) of Science (WoS) or Scopus proceedings are selected and
of research studies published by the most popular ven- analyzed as peer-reviewed publications for the methods and
ues regarding the review methods and articles, including applications of PSO. Finally, I selected 2,140 peer-reviewed

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2537

Start
book chapters, non-peer-reviewed studies, and low-quality
researches (published in rapacious journals) that did not pro-
Initialize a swarm of vide any technical information or scientific discussion. The
N particles randomly ultimately selected studies are recognized by considering the
inclusion criteria as follows:
Evaluate the current
personal fitness value
f ptbesti for each particle i • Papers published online from 2017 to 2019.
• Papers on PSO topics, including methods and applica-
tions.
Yes Is f ptbesti better than No • Papers prepared in a technical quality method in PSO
previous f pt−1
besti ? methods and applications.
• Papers that follow the quality standards.
Recognize ptbesti as the
personal best position for i
Keep the previous pt−1
besti The ultimately selected studies are recognized by consider-
ing the exclusion criteria as follows:
Assign the best position’s fit-
ness value ptbest∗ to gbest
t
• Papers that present survey and review studies.
• Papers indexed in neither WoS nor Scopus.
• Papers written in a language other than English.
No Maximum epochs, or Yes
optimum reached?
• Papers not subject to peer-review.

Update the velocity of

each particle using Eq. (3)
5 Methods of PSO

Update the position of In fact, the literature is rich in many PSO techniques. To
each particle using Eq. (4) organize them, a proper classification criterion is suggested,
and various PSO methods are classified as per Fig. 4.
End

5.1 Hybridization
Fig. 1 The flowchart of PSO
Hybridization is a generic model of two or more algorithms,
which exploits their advantages while decreasing their
impediments. The combination of techniques does well for
tacking a specified problem, given that the obtained results
can be improved by these combined techniques on their own.
The exploitation and exploration of an algorithm can com-
pletely be enhanced through hybridization of algorithms [53,
177, 180]. For example, an algorithm can cater for the lack
of its preciseness and refine the results through synergy with
a local search method. The hybrid approach in the field of
optimization is growing in popularity and utilizes the pur-
port of hybridizing the components from leading optimiza-
tion techniques to improve the performance of traditional
optimization algorithms.
Fig. 2 Distribution of research papers by publisher
5.1.1 Meta-heuristic Algorithms

papers. Due to the length limitation of the paper, I addressed Many nature-inspired hybrid approaches have been devel-
curtly only 84 state-of-the-art research papers from reputed oped by numerous researchers/scholars for achieving sig-
journals to address the four AQs in detail in Sects. 5 and nificant performance in the exploitation and exploration of
6. Figure 3 demonstrates the selection criteria and evalua- existing algorithms. To optimally design a Labyrinth Spill-
tion method for the literature studies. The exclusion phase way (LS) having quarter-round or half-round crest shape, an
was carried out by omitting white papers, short papers, evolutionary Hybrid Algorithm (HA) combining the PSO

13
2538 A. G. Gad

Fig. 3 The selection principles and evaluation of research papers

algorithm with the Bat Algorithm (BA) was developed by In [12], Aydilek proposed a Hybrid algorithm combin-
Ferdowsi et al. [52]. This way, the best outcomes of one ing both FA and PSO (HFPSO). HFPSO checks the global
algorithm substitute the worst ones from the other. More- historically best fitness values in order to properly deter-
over, the proposed algorithm has a high convergence rate mine the proper start of the local search. The high accuracy
regarding the optimal absolute outcome and does not get of convergence and runtime was statistically approved by
stuck into the local minima. Responses and computational evaluating the method using expensive benchmarks from
time of this HA proved that LS with quarter-round crest the high-dimensional CEC’15 and CEC’17 functions.
requires more concrete, compared to the half-round one. Typically, a strong classifier can be created based on the
In another work, Melton [126] examined a strategy for strengths of penalization and, say, SVM, proving the effec-
hybridizing two heuristic algorithms, PSO and DE, to better tiveness of Penalized Support Vector Machine (PSVM). In
address the stagnation problem when solving slew-maneuver [6], Al-Thanoon et al. proposed a new hybrid PSO and FA
time-optimal problems having constraints in solution path. to find the tuned parameters of PSVM, showing a high abil-
The stagnation was handled very well by the combined PSO- ity to escape from being trapped into the local optima trap.
DE method, with a reduction of 40% in the computational In [223], a hybrid method of PSO and GA was proposed
time compared to just applying DE. In [34], a Dynamic by Zhu et al. to optimize an antenna array deployment for
Multi-Swarm Differential Learning Particle Swarm Opti- locating the sources of Partial Discharge (PD) in an entire
mizer (DMSDL-PSO) was proposed. In this study, the DE substation. First, an algorithm called Direction-of-Arrival
operators were incorporated into each one of the DMSDL- (DOA) estimation was presented for testing its applicabil-
PSO’s sub-swarms to form a novel method. DMSDL-PSO ity to arbitrary array configurations. In order to minimize
has a good capability of exploitation and exploration by the objective functions, the hybrid PSO-GA algorithm was
employing a local searcher like Quasi-Newton method to applied to optimize the array deployments for the localiza-
enhance the exploitation capability, making, at the same tion of both DOA and coordinates.
time, use of the capability of exploration existing in the dif- Genetic Learning PSO (GL-PSO) breeds eminent exem-
ferential mutation. On the other hand, the optimization per- plars to steer the motion of particles for the purpose of
formance of DE was improved by developing a novel self- improving the performance of PSO. However, the perfor-
adaptive mutation DE algorithm based on PSO (DEPSO) mance of GL-PSO is not satisfactory on complex optimi-
[194]. DEPSO can significantly utilize the fast convergence zation problems, in which a global topology is depicted
capability of PSO and the strong global exploration capa- for exemplary generation and efficient diversity cannot be
bility of an improved DE mutation strategy. As a result, the retained to boost exploration. For the betterment of the
diversity of the swarm was managed well throughout the adaptability and performance of GL-PSO, Lin et al. [106]
evolution, resulting in a higher convergence speed. modified two versions of the basic GL-PSO algorithm. In

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2539

Fig. 4 Categorization of diverse PSO methods

exemplar generation, a ring topology is depicted to boost resulting technique: In the bottom layer, the search agents
diversity and so exploration, while the algorithm’s adaptabil- of SCA explore the search space, and in the top layer, the
ity is improved by adopting a Global Learning Component region around the best result found by the layer at bottom
(GLC) with control parameters which are linearly adjusted. is exploited. Hence, there is a balance between exploitation
Nenavath et al. [135] proposed a novel hybrid algorithm and exploration in the proposed technique, which maintains
of Sine Cosine Algorithm (SCA) and PSO (SCA-PSO) to fast convergence and improves the solution quality.
overcome the SCA premature convergence at local minima In another research, Sanchez et al. [159] used a column
for the two major goals of object tracking and solving opti- generation framework to solve the Binary Cutting Stock
mization problems. Two parameters of PSO, 𝐩best and 𝐠best , (BCS) problem by developing a benchmark of different
were embedded into the conventional SCA to lead the pro- approaches. Master- and sub-problem are the main compo-
cess of pursuing the possible candidate solutions, so that the nents of this framework. Classical integer linear program-
𝐩best in SCA is then used to initialize PSO swarm to further ming was used to solve the master problem, while meta-
exploit the feasible search spaces. heuristic algorithms (PSO, SA, and GA) were used to solve
PSO usually experiences premature convergence, so it the sub-problem. The aim of this benchmark analysis was
is easily to be caught into a local optimum solution. Also, to compare the results of the hybrid meta-heuristics with an
it is ineffective in balancing exploration and exploitation, exact approach.
especially in complex multipeak search functions. To over- In [68], Huang et al. presented a methodology to utilize
come these shortcomings, a Hybrid Particle Swarm Opti- various loading-unloading curves to extract the mechani-
mizer with Sine Cosine Acceleration Coefficients (H-PSO- cal properties of its bulk materials using an Instrumented
SCAC) was proposed [32]. First, Sine Cosine Acceleration Indentation Test (IIT). Simulated Annealing PSO (SAPSO)
Coefficients (SCAC) were presented to efficiently control the coupled with finite iterative element simulation were used
convergence to the global optimum solution by guiding the to obtain the mechanical properties, including strain-hard-
local search. Second, the swarm is initialized by adopting ening rate (K) and strain-hardening exponent (n), from the
opposition-based learning method. Additionally, the inertia IIT data.
weight is adjusted by utilizing a sine chaotic map. Finally, a Ibrahim et al. [70] proposed a hybrid optimization
modified formula for updating position was proposed. approach for solving the feature selection problem. The Salp
The SCA is mainly stigmatized by its limitation to exploit Swarm Algorithm (SSA) was combined with the PSO as
only emerging regions, depending on the parameter settings. SSAPSO, in which the performance of the exploitation and
However, the SCA is well capable of exploration. Taking exploration stages in PSO was significantly amended. Based
advantage of the PSO capability of strong exploitation, Issa on 15 benchmark functions alongside different UCI datasets,
et al. [73] developed a new enhanced version of the SCA the original datasets were refined by removing the confus-
by hybridizing it with PSO. Two layers have formed the ing or redundant features while yielding or keeping a better

13
2540 A. G. Gad

efficiency, demonstrating the improvement in the SSAPSO advantage of the introduced so called quasi-entropy index
with respect to the prediction accuracy without affecting the for addressing its main issue (specifically, when to start the
computational cost. local search) by developing an adaptive local search start-
Laskar et al. [98] proposed a new hybrid algorithm named ing scheme.
as the Hybrid Whale PSO (HWPSO) algorithm. Since In most of the above-discussed studies, PSO was imple-
Whale Optimization Algorithm (WOA) is known for its very mented without coherence with other meta-heuristics. For
good capability of exploration, it is hybridized with PSO in instance, a good initial solution can be produced by utilizing
a novel way to overwhelm the limitations related to a PSO DE or GA as a starting point, and the PSO can take care of
phase (i.e., stagnation impact). The HWPSO has employed the rest of the search. In some cases, an initial solution can
two novel techniques during hybridization: “Forced” whale be produced by using PSO as a starting point, and the search
in the exploration phase so that the WOA can guide PSO can be conducted by others. Furthermore, the PSO, as a sup-
to better avoid getting stuck into local optima, and “Cap- plementary tool, may be hybridized with the deterministic
ping” phenomenon in the exploitation phase for restricting local search to boost the exploitation capability and produce
the search mechanism of WOA in order to converge to the high-quality solutions. Additionally, the search speed may
global optimum value more fast. be elevated by the PSO.
A new form of multi-swarm BBPSO, in which its updat-
ing distributions are reoriented by adaptively learning the 5.1.2 Artificial Neural Networks (ANNs)
optimal alignments, is presented by Vafashoar and Mey-
bodi [183]. The idea is that, along appropriate directions, Artificial Neural Networks (ANNs) can be, in a word, sum-
the prospect of generating new particles is maximized. Due marized as learning via training [187]. To achieve that, a
to the lack of a priori knowledge of these alignment direc- series of input and output vectors is adopted to handle a
tions, the proposed method adaptively learns suitable align- set of data that needs to be trained. At the training time,
ments by employing a learning mechanism. Cellular learning the ANN is applied to the training data iterative, so that
automata were adopted to build this learning mechanism. the desired input-output mapping emerges by balancing
For each particle in the presented method, multiple align- weights of the network several times in a while. After the
ment strategies are developed. Moreover, these strategies are training stage completes, a foreseeing model is ready and
adjusted so that, during the search process, the particles are the corresponding output vector can be then created from
attracted toward the most promising regions based on the input ones unrelated to the training pairs. An appropriate
cellular learning automata. set of weights and transferring functions should be carefully
Durán-Rosal et al. [43] proposed novel approaches chosen (via, say, meta-heuristic optimization) to ensure that
based on time series segmentation. The proposed methods the ANNs can present a quantifiable smooth function that
include the PSO algorithm adaptation to this problem, as would properly link future input and output vectors. Vari-
well as more advanced variants of PSO (e.g., BBPSO and its ous components, including prediction and system control
exploitation variant (BBePSO)). Furthermore, a new algo- and modelling, are inferred to express the broadly connected
rithm, called Dynamic exploitation BBPSO (DBBePSO), neural systems.
was derived, in which the importance of the cognitive and In this regard, Moayedi et al. [131] introduced a PSO-
social components are updated throughout the successive optimized ANN model to solve the prediction problem of
generations. A final local search step is considered to further Landslide Susceptibility Mapping (LSM). The focus of this
improve the solutions obtained from these algorithms based study was the prediction of landslide hazardous susceptibil-
on the incorporation of two popular traditional segmentation ity mapping by applying a hybrid model of PSO and ANN
algorithms (Top-Down and Bottom-Up). (PSO-ANN). Two statistical performance metrics, Root
In [134], Nagra et al. proposed an enhanced Self-Inertia Mean-Squared Error (RMSE) and coefficient of determi-
Weight Adaptive PSO algorithm with a gradient-based Local nation ( R2 ), were used to assess the predicted results from
Search strategy (SIW-APSO-LS) to solve the problem of both the original ANN and the PSO-ANN. Comparing the
premature convergence in PSO. This proposed algorithm two models, the PSO-ANN model was observed to be more
takes the advantages of the exploration ability of the adap- reliable in LSM estimation than the ANN.
tive PSO with enhanced inertia weight, and the exploitation In another work, Junior and Yen [84] proposed a novel
of the gradient-based local search strategy. algorithm based on PSO and Convolutional Neural Network
In [30], Cao et al. proposed a Comprehensive Learn- (CNN), namely, PSO-CNN. In comparison with other evo-
ing Particle Swarm Optimizer (CLPSO) embedded with lutionary approaches, the proposed algorithm is capable of
a local search utilizing the CLPSO’s capability of global fast convergence. In an application to image classification, it
search and the fast convergence capability of local search to was able to automatically find deep meaningful CNN archi-
achieve higher optimization performance. This work takes tectures. To allow the use of PSO optimization with CNN,

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2541

a novel strategy for direct encoding, as well as a velocity in recent decades, QPSO has drawn great attention from
operator were built. many scholars. In what follows, some of them are presented.
In [81], Jmal et al. proposed a QPSO method for solving
5.1.3 Support Vector Machine (SVM) the K-Traveling Repairman Problem (K-TRP). The proposed
approach incorporates a K-TRP-specific repair operator to
Support Vector Machine (SVM) [56] is an intelligence algo- ensure that a feasible solution space will be always there to
rithm lying under supervised machine learning [2]. In this guide the search process and that the quality of solutions is
algorithm, the data is analyzed and particular visible or hid- significantly improved as much as possible.
den patterns are quested with the major aim to perform two In [147], an optimal path planning was proposed for free-
main tasks of classification and regression. SVM is known floating two-wheel pendulum robot system based on its self-
for its potential to perform non-linear classification and balance. First, the corner trajectory of this pendulum robot is
linear separation of the data, to build another dimensional parametrized by QPSO. The native attitude and the control
space from existing training datasets by investigating the precision of the robot’s terminal attitude and position are
respective categories. The separation process is working by utilized to formulate the objective function. The addressed
increasing the chasm between too close points of various problem of optimal path planning was considered as a non-
categories (i.e., support vectors) to better explore the optimal linear optimization problem that is solved using the QPSO
separating hyper plane. It should also point out that SVM algorithm to achieve the demanded goal.
can perform other tasks like linear classification [41]. Motivated by the characteristics of PSO and quantum the-
In this framework, Hoang et al. [61] proposed a novel ory, Xu et al. [199] proposed a new method, namely, QPSO.
Differential PSO-based SVM (DPSO-SVM) classifier to The QPSO used fractional calculus to enhance its global
monitor the conditions of a surge arrester. In the proposed search ability. The discrete expression of the Grünwald-
method, the parameters of SVM classifiers were optimized Letnikov definition, one of the most popular differential
by investigating the DPSO-SVM technique configuration fractional definitions, was incorporated for better updating
that gives the best results. Input vectors composed of three of the particles’ positions in QPSO.
extracted features are used to assess five arrester conditions,
including degradation (D), tracking (T), pre-fault (A), nor- 5.1.5 Rough Set
mal (N), and abnormal (U). Meanwhile, a DPSO-based
ANN classifier was also involved, to carry out a compara- The theory of rough set [142] is an approach devised to deal
tive study on fault diagnosis. with uncertainty and ambiguity. The philosophy of this
In another work, Tharwat and Hassanien [179] optimized approach was built on the fact that every object in the uni-
the SVM parameters by employing quantum-behaved PSO verse has its own characteristics. The rough set theory is a
(QPSO) for reducing the classification error. The proposed mathematical tool utilized to extract knowledge from infor-
model, called QPSO-SVM, was evaluated by adopting 7 mation obtained from uncertain data [105]. In feature selec-
traditional classification datasets from the UCI data reposi- tion, rough set analysis is utilized because it only requires
tory. The experimental results revealed the capability of the the supplied basic data without the need for any supplemen-
proposed model to obtain the best values of the parameters tary information. Also, it is characterized by the suitability
of SVM. Moreover, the results showed lower classification for exploring the qualitative and quantitative properties of
error rates than those obtained by the standard versions of data.
PSO and GA. Fan et al. [51] proposed an algorithm called RoughPSO
for solving the convergence to a local optimum in PSO,
5.1.4 Quantum depending on the rough set theory. The RoughPSO uses
upper- and lower-approximation rough sets to get the mem-
40 years ago, quantum computer [20] was invented, and bership values. Then, these values were employed to refine
in the late 1980s, the quantum computer formal definition the velocity and position for each particle.
was given. Due to the potential of the quantum computer in In another work, Bhattacharya et al. [23] proposed a
various special problems [85], this field has witnessed many technique of Feature Selection based on an impromptu PSO
great efforts. Various popular algorithms are developed, and algorithm and Rough Set (PSORS-FS), to select the most
the most well-known one of them is Shor’s quantum factor- relevant features for detecting permission-based Android
ing algorithm [185]. In 2004, Sun et al. [170] firstly intro- malwares. This work mainly contributed to the recommen-
duced the Quantum-behaved PSO (QPSO) for improving the dation of a new method of random key encoding that con-
convergence rate of the native PSO. In quantum space, par- verts the conventional PSO algorithm in discrete space. It
ticles often have the guarantee to reach the real global opti- also reduced the issues related to the sigmoid function in
mum via searching throughout the full solution space. Later binary PSO, as well as the particles’ maximum velocity.

13
2542 A. G. Gad

5.2 Improved PSO velocity, social and cognitive factors, and inertia weight),
Fuzzy Logic (FL) can be first exploited to determine the
This section is competent of the improvements in PSO based best values of these parameters. So far, the PSO implementa-
on some third-party methods as learning strategy, fuzzy tions based on FL aim at calculating distinct settings for the
logic, mutation, Lévy flight, opposition-based learning, and swarm as a whole. Nobile et al. [138] proposed a new algo-
surrogate. rithm called Fuzzy Self-Tuning PSO (FST-PSO) which, for
every particle, independently calculates the minimum and
5.2.1 Learning Strategy maximum velocity, social and cognitive factors, and inertia
by exploiting FL, thereby creating a completely setting-free
In reality, increasing complexity in many optimization variant of PSO. The strength and novelty of FST-PSO was
problems calls for further research on the exploration and derived from the fact that there is no experience required to
improvement of diverse optimization algorithms. It has been formulate PSO, since the optimization process automatically
proved that the PSO algorithm is an efficient tool to solve adaptively adjusts the behavior of each particle.
different types of optimization tasks. However, for the base In [114], López et al. proposed a Fuzzy Logic Controller
PSO, the main aim of the updating strategy is to learn the (FLC) modified by PSO, namely, Fuzzy-PSO to increment
global optimum, and it always experiences too fast conver- the lifetime of power electronics with a faster response of
gence in addition to poor performance on various complex drive’s speed in a brushless DC electric motor. Furthermore,
optimization tasks, principally for multimodal problems. the reference temperature and the desired speed are proposed
The mechanism of learning in classical PSO is simple and as parameters to formulate an objective function.
easy to develop, but some likely issues, such as the phenom-
ena of “two steps forward, one step back” and “oscillation”, 5.2.3 Mutation
may disrupt it. Therefore, PSO researchers have managed to
design new efficacious learning strategies for avoiding the Salajegheh and Salajegheh [157] combined PSO with gradi-
two phenomena and improving the overall search perfor- ent directions of first and second order, thereby achieving a
mance. Xu et al. [198] presented a Dimensional Learning great increasing in the approach robustness. The literature
Strategy (DLS) following the best personal experience of was thoroughly searched to choose an appropriate set of
each particle to eventually discover and integrate the pro- design problems, and the results of the standard PSO and the
pitious information extracted from the optimal solution of counterparts are compared with each other. In the proposed
swarm. Thereafter, different learning strategies are used to approach, the overall optimization cost was trivial as a result
provide a Two-Swarm Learning PSO algorithm (TSLPSO). of decreasing the number of initial particles.
One of the two sub-swarms used the DLS to create learning A modified variant, namely, Repository and Mutation
exemplars for guiding the particles’ local search, while the based PSO (RMPSO) was proposed in [75]. In RMPSO,
other sub-swarm used the comprehensive learning strategy global and personal best solutions with the same fitness val-
to create learning exemplars for managing the global search ues are stored in two maintained extra repositories. And,
of particles. another proposed Enhanced Leader PSO (ELPSO) indicated
Wang and Liu [193] proposed a Heterogeneous Com- that the swarm optimum could be improved by applying five
prehensive Learning PSO (HCLPSO) algorithm to attain successive mutation strategies.
3D spatial trajectory tracking to realize a new saturated
approach to control a quadrotor. First, the quadrotor model 5.2.4 Lévy Flight (LF)
was divided into: an inner position control loop, inside cas-
caded control structure, and an outer attitude control loop. Lévy Flight (LF) is a particular category of random walk in
Second, the saturated control is applied to limit the quadro- which a heavy power tails law is applied for the distribution
tor’s thrust force in the outer attitude control loop. Next, of step lengths. A global search is done or conducted by an
the parameter adjustment difficulty for the quadrotor was algorithm with the occasional aid of larger steps. A better
alleviated by employing the HCLPSO algorithm to optimize trade-off between the two capabilities of exploration and
the control parameters. exploitation in an algorithm could be gained by applying
the LF trajectory [47], and the ability to avoid local optima
5.2.2 Fuzzy Logic (FL) gives plus points.
In [213], Zhang et al. established an optimization model
Since the performance of PSO is strongly dependent on for the task scheduling model in the Multiple-Input and Mul-
the selection of its settings (i.e., minimum and maximum tiple-Output (MIMO) radar, and proposed a hybrid Discrete
PSO (DPSO) algorithm with LF for solving the problem.

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2543

The internal structure of tasks, the features of the MIMO standard PSO algorithm itself with the goal of solving high-
radar task scheduling, and the three main principles of task dimensional problems. The cooperation between SL-PSO
scheduling were all considered in the optimization model. and standard PSO was proposed so that the search space
In another work, Tang et al. [175] employed the signal level can be thoroughly exploited and explored, and the surrogate
of Automatic Identification System (AIS) to build a new model global and local search performance is simultane-
method for estimating the atmospheric profile refractivity, ously improved.
specifically the parameters of surface-based duct in the mari- In [203], Yang et al. proposed an Improved Surrogate-
time environment. Due to the complexity and non-linearity Assisted PSO (ISAPSO) algorithm, in which global and
of this optimization problem, LF with QPSO algorithm (LF- local surrogates are integrated into a hybrid PSO. The global
QPSO) was proposed to find the optimal solution. surrogate uses an efficient global optimization algorithm to
In [100], Li et al. developed a novel Unsupervised LF work as a global searcher for speeding up the PSO global
with PSO method (ULPSO) for image classification, giving search process, and reduces the computational burden by
a good exploration-exploitation balance. ULPSO depends on predicting the fitness values. Meanwhile, the local surrogate
a new searching mechanism that targets the worst particle in explores the expected optimum solution so that it can later
the swarm at each iteration to update its position with LF. locally search through the neighborhood of this solution.

5.2.5 Opposition-Based Learning (OBL) 5.3 Variants of PSO

Tizhoosh [181] originally introduced the basic idea of Different variants of the PSO algorithm are addressed in
Opposition-Based Learning (OBL). Based on the concept of this section.
OBL, typically, the fitness values of current agents and their
opposites are evaluated [119]. Then, the suggested algorithm 5.3.1 Binary
selects N agents with the highest/lowest fitness values.
In [86], Kang et al. proposed a hybrid PSO algorithm to Continuous meta-heuristics cannot handle a variety of opti-
improve the PSO performance by maximizing PSO diver- mization problems. Therefore, combinatorial optimization
sity introduced by OBL. Probabilistic OBL was adapted for has emerged, which includes a type of optimization called
a swarm by the proposed hybrid algorithms. Unlike other binary optimization. The {0, 1} set elements are distinct
fusions of PSO and OBL, the fitness of the entire swarm was decision variables in binary optimization problems. Thus,
improved by selecting the fittest particles from the current the binary optimization problem can have 0 and 1 as poten-
swarm and its opposite one. tial values for each decision variable. In other words, the
In [189], Wang et al. addressed the learning strategy decision variables can be digitally expressed as 0 or 1 with
impact on a scheduling problem with two-stage assembly respect to the binary optimization problem at hand.
flow shop. Three different machines were used to, in turn, Sameer et al. [158] proposed a new Modified Binary PSO
process several different workpieces. The total completion with a Kernel Fuzzy Clustering Method (MBPSO-KFCM)
time was minimized by developing a branch-and-bound for selecting relevant features and determining the number
technique combined with a lower bound procedure and of clusters in fuzzy data clustering. To enhance the quality
several developed dominance rules to obtain the optimal of cluster analysis, Gustafson-Kessel (GK) algorithm was
scheduling solution. Besides, high-quality feasible solutions established and improved. The datasets Taiwanese, German,
were obtained by devising a new Dominance Rule-based and Australian credit were used to test the performance of
Opposition-based PSO algorithm (DR-OPSO). the three proposed algorithms. The results showed that fuzzy
partition (classification) is robust based on the fuzzy Rand
5.2.6 Surrogate validity metrics, which provides an opportunity to reduce
the risk associated with loans.
Surrogate-assisted optimization was established to handle In [89], Karbassi Yazdi et al. proposed a Binary PSO vari-
expensive and complex problems in real-world applications. ant (BPSO) for optimally solving the problem of ship routing
The methodology of this type of optimization depends on and scheduling in Liquefied Natural Gas (LNG) extraction,
the maximal exhaustion of the available information for transportation, and regasification.
minimizing the number of expensive evaluations required
for a given objective function, thereby reducing the related 5.3.2 Chaotic
costs, resources, and time [169].
Related to this, Yu et al. [207] proposed a hierarchical Generally, chaotic, adapted from the term “chaos”, formally
surrogate-assisted particle swarm optimizer comprising a refers to such a property that describes the unpredictable
Social Learning-based PSO algorithm (SL-PSO) and the behavior of a complicated system. In chaotic systems, a

13
2544 A. G. Gad

function is used to “map” or associate algorithm chaotic pairwise competitions which in turn are used to update the
behavior with some parameters, using two typical chaotic position particles. On the other hand, a bi-objective prob-
maps: logistic and sine. Chaos has distinctive properties of lem was formulated for the hand posture estimation problem
non-repeatability and periodicity. So, it has the ability to do [160]. To solve this problem, the concepts of Evolutionary
general searches at higher speeds compared to the probabil- Population Dynamics (EPD) were employed for developing
istic stochastic searches [72]. an improved version of MOPSO. This problem has different
In [200], Xu et al. proposed a new Chaotic Search PSO types of parameters. Therefore, some of the parameters were
algorithm (CS-PSO) which enhances the PSO algorithm significantly calibrated by the MOPSO algorithm.
with the CS method for solving combinatorial optimization In [219], Zheng et al. presented an adaptive neighbor-
problems. In the initialization phase, the particles were posi- hood function for developing a Multi-Objective Cellular
tioned depending on the advance knowledge of the combi- PSO (MOCPSO). Three objective functions were taken for
natorial optimization problems. Furthermore, in the chaos drilling a wellbore trajectory less costly, more quickly, and
perturbing phase, the positions and velocities of particles are safely, compared to other potential trajectories. Then, a set of
perturbed by introducing a brand-new range of rules for sat- Pareto optimal solutions is gained by applying the MOCPSO
isfying the adaptability and the capability of the ideal global algorithm to the three objective functions for designing a
search with the major aim to effectively avoid the fruitless cheaper and safer wellbore trajectory.
early convergence that haunts the standard PSO algorithm. Adhikari and Srirama [5] developed a new Container-
In another work, Wang et al. [188] presented an efficient based Energy-Efficient Scheduling method (CEES) that
multi-objective optimization approach that combines an responds effectively to different types of fast submitted
adaptive chaotic PSO and the IsoGeometric Analysis (IGA) Internet of Things (IoT) and non-IoT tasks. A Multi-Objec-
to optimize the Ceramic Volume Fraction (CVF) distribu- tive Accelerated PSO (MOAPSO) technique was used in
tion under eigenfrequencies in Functionally Graded Plates the proposed method to ensure a minimum delay by finding
(FGPs). And, the B-spline basis function was used to repre- the most appropriate container for executing each task. To
sent the CVF distribution. better utilize cloud resources, cloud environment involves
In [209], Yue et al. proposed a hybrid PSO algorithm other important missions of resource scheduling, which can
based on the chaos theory for solving the mobile loca- be explored in [64].
tion estimation problem. In the proposed location estima-
tion approach, it was crucial to consider the influence of
measurement error arising from the nonlinear optimization 6 PSO Applications
problem and Angle Of Arrival (AOA)/Time Difference Of
Arrival (TDOA) hybrid location method. The proposed In accordance with the SR process adopted in this study, this
algorithm recognized the AOA/TDOA hybrid location algo- section is to provide a technical review of the designated
rithm to track the object to significantly enhance the locating PSO applications in the actual literature. Figure 5 demon-
accuracy and the performance of the estimation process in strates a comprehensive taxonomy of the PSO applications
general. in different domains, including health-care, environmental,
industrial, commercial, smart city, and general aspects. Each
5.3.3 Multi-objective category of PSO applications is likely to face some issues
that should be highlighted to come up with thriving solu-
Multi-Objective Problems (MOPs) [108] is to be introduced tions, enabling further efficient and viable PSO implementa-
in this section. Minimizing or maximizing multiple conflict- tion in future real-world applications. Consequently, studies
ing objective functions are two main tasks often considered focusing on some pivotal issues are reviewed to prop PSO
in MOPs [148, 221]. Unlike single objective optimization applications in a particular context associated with these
ones, MOPs involve multiple contradictory objective func- issues. For example, in environmental applications, the main
tions. It is not easy to simultaneously optimize many func- contexts, such as economic emission dispatch, parameter
tions. Hence, for obtaining an optimal solution set, there identification of PhotoVoltaics (PV), pollution forecasting,
should be balance between the objective functions. Some segmentation and classification of plants, flood control and
constraints and at least two objectives must be included in a routing, water quality monitoring, and many other issues, are
multi-objective optimization problem. In a solution space, floated in different aspects of environmental PSO applica-
the objectives and constraints are satisfied through finding tions. Thus, this paper presents a taxonomy based on diverse
mutually acceptable solutions. categories of PSO applications in selected research studies
In [214], Zhang et al. proposed a multi-objective parti- in which special subjunctives are addressed and discussed.
cle swarm optimizer based on a competitive mechanism, Considering the concerns and challenges in various types
in which the current swarm performs, at each iteration, of PSO applications, I firstly addressed different categories

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2545

Fig. 5 The taxonomy of PSO applications

in the PSO applications and then reported the main subjects 6.3 Industrial Applications
which have been paid special attention in each category.
PSO applications have some general concerns, so a division Table 5 illustrates a summary of the recent researches and
in the taxonomy, namely “general aspects”, was adopted to important information to assess the industrial approaches in
refer to the studies that introduced a proposal to cope with a PSO applications. Some of the major contexts in the indus-
particular challenge in any general type of PSO applications. trial domain include economic dispatch problem in power
In other words, a new conceptual approach is introduced systems, optimizing Phasor Measurement Units (PMUs)
based on the studies shown as being of general aspects, thus placement, allocating daily electrical loads, deploying
promoting the development of any type of PSO application. WSNs, predicting the defection of products, and optimizing
The upcoming subsections present diverse approaches in the design and operation of microgrids.
PSO applications. In addition, various researches will be
compared from several sides, such as key subject, case stud- 6.4 Commercial Applications
ies, strengths, shortcomings, and special outputs.
Table 6 illustrates a summary of some seminal researches
6.1 Health-Care Applications and important information to assess the commercial
approaches in PSO applications. Some of the major contexts
Table 3 illustrates a summary of some seminal researches in the commercial domain include prediction of cost and
and important information to assess health-care approaches price, risk assessment, and profit calculation.
in PSO applications. Some of the major contexts in the
health-care domain are based on intelligent diagnosis, dis- 6.5 Smart City Applications
ease detection by medical robots, medical image segmenta-
tion, and disease classification. Table 7 illustrates a summary of some seminal researches
and important information to assess the smart city
6.2 Environmental Applications approaches in PSO applications. The approaches include
case studies, such as smart city, smart home, appliance
Table 4 illustrates a summary of the recent researches scheduling, and vehicular monitoring systems.
and important information to assess the environmental
approaches in PSO applications. The approaches include 6.6 General Aspects of PSO Applications
case studies, such as wild vegetation environmental moni-
toring, agriculture environmental monitoring, flood control Table 8 describes a summary of some seminal researches
and routing, water quality monitoring, and pollutant concen- and important information to assess the general approaches
tration monitoring systems. in PSO applications. Some of the major contexts in the

13
 2546

13
Table 3 Summary of some seminal researches in health-care applications
Research Main subject Strengths Shortcomings New finding(s)

Pashae et al. [140] Disease diagnostic system by identifying • Identifying biologically and statisti- • Not considering computational time • Hybrid meta-heuristic approach
the most beneficial genes for classifica- cally significant genes from the clinical
tion datasets
• A better performance of the fusion of
Binary Black Hole Algorithm (BBHA)
and BPSO (4-2)
Zeng et al. [212] Diagnosis of Alzheimer’s Disease • Outperforming several SVM models and • Experimenting with only one dataset, • Framework
two other state-of-the-art deep learning ADNI
methods
Jain et al. [74] Cancer diagnosis and classification using • Providing a serious solution to the inher- • Not considering computational time • Two-phase hybrid model
DNA microarray technology ent local optimum problem in traditional
BPSO
• Ensuring faster and more reliable gene
selection in classification
Li et al. [104] Medical image segmentation • Getting better exploited contextual • Limited variation of image segmentation • Algorithm
information problems
• Improving the performance of dynamic
context cooperative QPSO
Raj and Ray [151] ECG signal analysis • Positive predictivity • Not performing real-time analysis • Hybrid algorithm
• High sensitivity • Not incorporating many classes of
• High overall accuracy arrhythmia signals for analysis
• High computational time
Srisukkham et al. [168] Intelligent Leukaemia diagnosis • Escaping from the local optima trap • Not considering the velocity, as well as, • Two modified BBPSO algo-
• Accelerated chaotic search only updating the particles’ positions rithms
A. G. Gad
 Table 4 Summary of some seminal researches in environmental applications
Research Main subject Case study Strengths Shortcomings New finding(s)

Kumar et al. [97] Short-term temperature predic- Environmental monitoring • Improving the generalization • Not testing in a real environ- • Hybrid model
tion using ambient sensors performance ment
• Improving the accuracy
Zarei et al. [210] Multi-purpose water reservoir Shahid Dam Reservoir in Fars, • Low computational time • Not evaluating monetary cost • Hybrid evolutionary algorithm
southern Iran • High mean water release
values
• High reliability
Chen et al. [33] Forecasting short-term atmos- Temple of Heaven, Beijing • High forecasting accuracy • Considering temporary • Hybrid forecasting model
pheric pollutant concentration • Low runtime emergencies (such as major
based on PSO-SVM holidays) as influential factors
Rahgoshay et al. [149] Predicting daily suspended Royan and Veynakeh earth • Increasing accuracy • Structure is so simple and • Support vector method
sediment load dams in Semnan, Iran • Reducing the computational does not have complex
time and RMSE parameters
Kour and Arora [95] Segmentation and classification Agriculture environment • High accuracy • Not considering plants with • Automatic vision-based
of plants based on leaf images • Low computational time medicinal and scientific method
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review

using hybrid PSO-SVM significance

Cao et al. [29] Remote sensing to monitor Inland waters in Weishan Lake, • High prediction • Using hyperspectral sensors • Catastrophe strategy-modified
water quality based on HJ-1A China • High stability with low spectral resolution BPSO
HSI imagery • Fast convergence • Not testing on diverse lakes
Ali Ghorbani et al. [8] Forecasting daily pan evapora- Talesh meteorological station in • High level of accuracy • Not addressing non-stationar- • Forecasting & decision-support
tion based on hybrid ANN- Northern Iran ity in the climate datasets tool
QPSO embedded into a multi- • Uncertainty in the forecasted
layer perceptron method evaporation data
Ehteram et al. [45] Improving the Muskingum Wilson, Karahan, and Viess- • Increasing significance in • Building the Muskingum • Hybrid BA and PSO algorithm
flood routing method man and Lewis floods in the Obtaining the best solutions models with little parameters
USA and UK • Decreasing the computational
time
Camci et al. [28] Quality inspection for rice Rice farms in Longsheng, • High performance in han- • Ineffective quadcopter control • Control structure
farms by employing quad- China dling noise and uncertainties in natural landscapes
copters in the system

13
2547
2548 A. G. Gad

Table 5 Summary of some seminal researches in industrial applications

Research Main subject Strengths Shortcomings New finding(s)

Maiyar and Thakkar [120] Food grain transportation • Reducing food grain • Not considering perish- • Algorithm
problem wastages able food grain products • Decision support tool
• Economic and environ-
mental results
Alnaqi et al. [10] Prediction of energetic • High performance • Not considering scal- • Neural network model
performance of a building • High reliability ability • Hybrid algorithm
integrated photovoltaic/
thermal system
Mohebbi et al. [132] Optimal design of a • Integrated concurrent • Not evaluating on a • Algorithm
mechatronic quadrotor design multi-agent platform
system
Wang et al. [192] Recognizing and diagnos- • Improving the classifica- • Not evaluating actual • Hybrid fault diagnosis
ing potential faults for tion accuracy data method
nuclear power plants • Reducing the training • Hybrid algorithm
time
• Reducing noise
Liu et al. [111] Multi-objective optimi- • Reducing mass • Only three targets were • Hybrid algorithm
zation design of the • Improving stability considered in the multi-
airborne electro-optical • Reducing mechanical objective optimization
platform resonance problem
Ghorbani et al. [57] Optimal sizing of an • Increased reliability • Not presenting the total • Hybrid PV-WT generat-
off-grid house with PV • Minimizing the total accuracy evaluation ing unit
panels, Wind Turbines monetary cost • Hybrid algorithm
(WTs), and battery
Jiang et al. [77] Generating association • Low time-consuming • Inability to detect spam- • Algorithm
rules for supporting • Low monetary cost ming reviews • Decision support system
effective design based on • Affected by sentimental
online customer reviews reviews
Song et al. [167] Positioning a 3D wind • High power production • Lack of assessment on • Hybrid algorithm
turbine with multiple hub • Low monetary cost complicated and realistic
heights on flat terrain wind farm optimization
problems
Qi et al. [145] Predicting the unconfined • Low cost • Not foreseeing the long- • Algorithm
compressive strength of • Low response time term strength
cemented paste backfill • High accuracy • Omission of some influ-
encing variables
Lopes et al. [113] Distribution of electrical • Minimizing the cost of • Not evaluating computa- • Hybrid algorithm
loads throughout the day industry production tional time
in an industrial environ- • Better-quality solutions • Testing on small datasets
ment
Rahman and Zobaa [150] Optimizing PMUs place- • Reducing the computa- • Not considering scal- • Algorithm
ment tional cost ability
• Reducing the number of
PMUs needed for IEEE
300-bus system

general domain include service allocation, image segmen- competitiveness over, for example, EAs in terms of conver-
tation, scheduling, prediction, and security management. gence rates and accuracy in many cases. Table 9 lists some
of the impacting strengths and weaknesses of PSO.
Previous sections discussed the review process of the
7 Discussion and Comparisons selected studies regarding PSO methods and applications.
In turn, this section statistically analyzes those methods and
Similar to many other meta-heuristic algorithms, PSO has application areas. Furthermore, the proposed AQs proposed
distinctive advantages, as well as some unavoidable short- in Sect. 4 are called to present some analytical reports as
comings. Although there is no evidence of convergence for follows:
this algorithm, the results presented in this study indicate its

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2549

Table 6 Summary of some seminal researches in commercial applications

Research Main subject Strengths Shortcomings New finding(s)

Jiao et al. [78] Optimal location for an • Social benefits • Not including the multi- • Location estimation
electric business centre • Low cost objective concept model
• High transportation • Hybrid algorithm
convenience
Tang et al. [174] Forecasting building mate- • High scalability • Not evaluating computa- • Price prediction model
rials’ prices • Fast convergence rate tional time • Hybrid algorithm
• High prediction precision
Shen and Han [165] Profit calculation module • Getting real-time finan- • Not considering compu- • Accounting information
of financial accounting cial processing results tational cost system
information system
Yi et al. [205] Cost prediction of trans- • Strong practical signifi- • Not evaluating scalability • Intelligent cost prediction
mission line project cance model
• Improving the accuracy
Pradeepkumar and Ravi Forecasting volatility from • Yielding statistically • Not evaluating the overall • Neural network model
[144] financial time series significant results computational time

• AQ1: Into which domains can PSO methods and applica- • AQ3: What assessment environments are used for evalu-
tions be classified? ating the PSO methods and applications?

According to the taxonomy exhibited in Sect. 5, Fig. 6 According to Fig. 10, it has been observed that 47% of the
depicts a comparison of the percentages of PSO methods as research studies apply simulation tools to evaluate the pre-
of the date of this study. I considered three PSO methods that sented case studies in the PSO community. In addition, it has
span hybridization, improvement, and variants. As shown in been observed that 33% of the research papers have imple-
Fig. 7, PSO variants have the largest share of PSO methods mented approaches to develop PSO methods and applica-
in the literature at 42%. Of course, hybridization techniques tions. Moreover, 11% of the researches have not presented
have 32%, and improved PSO approaches have 26% usage any simulation or implementation for the indicated methods
of the PSO methods. and application domain. Finally, 9% of the existing studies
use datasets to apply analysis approaches (e.g., testing and
Similarly, according to the taxonomy exhibited in Sect. 6, prediction) for assessing the involved case studies.
Fig. 8 depicts a comparison of the portions of PSO appli-
cations as of the date of this study. Six PSO application
domains, including health-care, environmental, industrial, 8 Challenges and Future Opportunities
commercial, smart city, and general aspects applications,
are considered. If we look closely, the general approach has While PSO has been effectively applied in versatile domains,
the largest percentage of application areas in the literature challenges are required to be addressed and drawn as future
with a usage ratio of 27%. Of course, industrial applications research directions. In the past few years, PSO has gained an
have 20%, environmental applications have 17%, smart city increasing interest from researchers and has been utilized in
applications have 16%, health-care applications have 11%, different areas of applications. However, critical problems
and commercial applications have 9% usage out of the over- and issues remain. Thus, more research efforts should be
all PSO applications domain. done by scholars and researchers to conquer the challenges
and problems that may hinder the future application of PSO.
• AQ2: Which main contexts can be considered for PSO Moreover, further inspirations and more effective techniques
methods and applications? should be pursued to develop novel PSO approaches. For
example, the research community needs to address new
Within the framework of this study, the main contexts of methodologies for complex problems. Based on the litera-
PSO methods and applications are shown in Fig. 9. It has ture review, I discuss below different open issues and related
been observed that the electrical engineering field received topics for potential future research.
the most attention from scholars and practitioners with 13 I should point out that the assortment of PSO methods
studies, while prediction, image processing, and mechanical and applications mentioned in this study applied the SR pro-
engineering approaches were 6 studies. cess so that the challenges discussed in the following two
subsections are presented as the AQ4:

13
 2550

13
Table 7 Summary of some seminal researches in smart city applications
Research Main subject Case study Strengths Shortcomings New finding(s)

Abid et al. [4] Managing energy in smart homes Residential area of ten homes • Low cost • Not comparing the presented • Energy management
• Low power consumption method with other existing ones strategy
• Not evaluating accuracy • Algorithm
Zhang et al. [217] GIS-based placement of charging Changping, Beijing, China • High revenue • Considering only two scenarios • Algorithm
stations for electric vehicles • High coverage • Placement model
• Low cost
Li et al. [102] Forecasting Day-ahead traffic flow Highways • High stability • High time-consuming • Hybrid model
• High accuracy • The dataset is for a small fraction • Algorithm
of a highway
Jordehi [82] Scheduling shiftable appliances Smart home • Reducing consumers’ daily • Not considering scalability • Hybrid algorithm
electricity bill without affecting
their comfort
Le et al. [99] Estimating and controlling the Smart city • Robust technique • Not evaluating cost • Hybrid algorithm
heating load of buildings • High reliability • Low accuracy of some proposed
models
Ma et al. [118] Appliance scheduling Residential unit • Obtaining profits from selling • Uncertainty of renewable genera- • Appliance scheduling
electricity to the power grid tion model
• Good convergence performance
Hu et al. [65] Scheduling urban traffic light An area centering Xudong, • Elaborating traffic congestion • Not considering the case of self- • Hybrid algorithm
Wuhan, China • Controlling vehicle movements organized cities
• Not considering scalability
Not evaluating cost
Sato et al. [161] Multifaceted optimization of power Smart city • Improving the solution quality • Not evaluating monetary cost • Hybrid algorithm
grids • Low scalability
• Not verifying the robustness
Ramya et al. [152] Retrieval of deprived riot video Smart city • Can be used with other signal • Low accuracy when processing • Algorithm
data and image processing techniques large databases
• Identifying the previous criminal
records in a particular region of
the smart city
A. G. Gad
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2551

Table 8 Summary of some seminal researches in general aspects of PSO applications

Research Main subject Strengths Shortcomings New finding(s)

Bhattacharya et al. [24] Permission based detection of • High scalability • Considering only the per- • Hybrid algo-
Android malwares • High classification perfor- missions set rithm
mance • Not handling float datasets
Sivaranjani et al. [166] Speckle noise removal in • Considering reference and • Not considering scalability • Algorithm
SAR images no-reference metrics in • Framework
experiments
• Optimizing threshold values
Zarrouk et al. [211] Job shop scheduling problem • Low CPU time • Not considering the cost • Algorithm
• High-quality solutions
Mansouri et al. [122] Task scheduling in cloud • Low resource usage • Not trying in a real cloud • Algorithm
computing • Low execution time environment
• Not combining fault toler-
ance parameters in cloud
• Not considering the prec-
edence of tasks and load
balancing
Lin et al. [107] Set-union knapsack problem • High-quality solutions • High computational cost Hybrid algorithm
Zhong et al. [220] Travelling salesman problem • High balance between • Not comparing with a sub- • Algorithm
intensification and diversi- stantial number of previous
fication optimization techniques
Mokhtari and Noroozi [133] Flow shop scheduling • No job earliness or tardi- • Considering only com- • Algorithm
problem ness mercial optimization of
problem instances with
small size
Sun et al. [172] Locating attacks’ position in • Low energy consumption • Not locating the attack • Algorithm
WSNs Low task processing time source nodes
Thabit and Mohades [176] Path planning of multi-robots • Short, safe, and smooth • Not implementing real- • Algorithm
paths world case studies
Nouiri et al. [139] Job shop scheduling problem • Decentralizing decisions • High energy consumption • Architecture
• Effectiveness in directing • Not all entities participate
real production in the final solution
Alswaitti et al. [11] Data clustering • High cluster compactness • Lack of assessment on real- • Algorithm
• High classification accuracy world applications
Suresh and Lal [173] Segmentation of satellite • Robust and fast algorithm • Low quality metric values • Algorithm
images based on multilevel • High Stability • High computational time
thresholding
Sheikholeslami and Navimi- Service allocation in cloud • High revenue for cloud • No sensitivity analysis • Algorithm
pour [164] computing users and providers for different weighting
• Fast resource provisioning parameters
• High resource utilization • Not implementing in realis-
tic cloud environment
• Not evaluating customer
satisfaction

Table 9 Advantages and disadvantages of PSO

Advantages Disadvantages

• Simple implementation • Difficulty to initialize control parameters

• Relatively few parameters to tuned • Inability to work out the problem of swarm scattering
• Ability to run parallel computations • Premature convergence and trapping into the local minima especially
• Robustness on solving high-dimensional problems
• Higher efficiency and probability to find the global optima
• Fast convergence
• No overlapping or mutation
• Low computational time
• Ability to build accurate mathematical models for solving complex
problems

13
2552 A. G. Gad

Surrogate Learning Strategy

13% 19%

Opposi!on-Based
Learning
16%

Fuzzy Logic
23%

Lévy Flight
16%
Muta!on
13%

Fig. 6 Percentage of the presented PSO methods

• AQ4: What are generally the PSO open perspectives and

future research directions that can be drawn from the
work presented in the paper?

8.1 Open Issues

Premature convergence When applying the traditional PSO

algorithm, if the search gets closer to the local optimal solu-
tion through random initial conditions, the optimal solution
for a single particle (individual particle optimum) and the
optimal solution for all particles (the group/global optimum)
Fig. 7 Total percentage of the presented PSO methods would both converge towards the local optimal solution, and
a deceptive optimization result will therefore be obtained.
Thus, it is not possible to guarantee the capability of finding
the global optimal solution. As a result, the fast convergence
capability will be ineffective. Over the past few years, the
issue of stagnation (premature convergence) has been a sub-
ject of research in the PSO community [197, 208]. A road-
map for future research in this regard can be created based
on some topics, including particles’ stability analysis [26],
redistributing mechanism [146], and random sampling of
control parameters [171].
Convergence speed controller Despite the typical fast
convergence of PSO, it may be trapped into a local optimum
Fig. 8 Percentage of the presented PSO applications due to premature converge. Therefore, a convergence speed
controller was devised for the ultimate goal of solving this

Fig. 9 Percentage of the main

contexts for PSO methods and
applications

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2553

mainly because the explicit or implicit historical memory in

PSO improves the search process by storing promising solu-
tions and reusing them in later stages [101]. Furthermore,
a new inertia weight can be generated by using historical
memory through a parameter adaptation mechanism. There-
fore, improving PSO using historical memory is a pressing
challenge [60, 103]. And how to set the memory size adap-
tively can be also considered as an interesting topic to be
addressed in the future [117].
Parameter & topology selection In PSO-based algo-
rithms, the best performance can be achieved by elaborately
determining the control parameters. However, choosing
Fig. 10 Percentage of the PSO assessment environments offered in these parameters cannot be adequately guided. To address
the literature the job of parameter selection, future efforts in this topic
should include choosing the best parameters based on simu-
lations [38], parametric analysis in a computational envi-
problem. Two adaptive approaches were proposed to adjust ronment with limited resources [162], and hyper-parameter
the convergence speed. First, when the particle prematurely selection based on heuristics [115]. On the other hand, the
converges, the convergence speed is slowed down. Second, performance of PSO in engineering applications is signifi-
when the particle cannot update its best solution in the pre- cantly affected by topology selection, and each problem has
sent time, the convergence speed of PSO is accelerated. its appropriate optimal topology. However, this issue still
Thus, a key residual challenge is developing an effective lacks much study. As a future direction, topology selection
convergence speed controller [110]. In the same context, for for PSO can be better guided by taking into account factors
improving the convergence speed of PSO-based algorithms, affecting the optimality of algorithmic parameters (i.e., the
the social and cognitive components of the velocity update topological degree and the number of particles) with the aim
formula are modified based on the fitness of each particle of selecting a proper class of deterministic regular topolo-
[125]. In addition, the convergence speed is adaptively con- gies [112]. Other challenges including, studying and evaluat-
trolled to empower the PSO algorithm for solving large-scale ing the performance of tree topology [155] and suggesting
numerical optimization problems [66]. Hence, designing different topologies for the PSO optimizer in order to reach,
an appropriate convergence speed controller framework for for example, an optimal Proportional-Integral-Derivative
PSO has become a major challenge in this topic. (PID) controller design [3], are also pressing issues which
High-dimensional search space Because of the curse of calls for further research and audit.
dimensionality problem, the effectiveness of PSO applica-
tion to classification of high-dimensional data has recently 8.2 Future Perspectives
gained a major concern [9]. In a recent study, the effec-
tiveness of PSO approach for feature selection has been PSO was practically applied for the first time to the domain
demonstrated. However, due to the large search space, the of neural network training, which was indicated by the
challenge sill exists to apply PSO-based feature selection to algorithm itself [93]. Many more areas of PSO methods
high-dimensional datasets with tens of thousands of features and application have been explored ever since, including
[182]. In addition, the PSO algorithm can be adapted for signal processing, power systems, combinatorial optimiza-
the problem of high-dimensional feature selection, providing tion, design, data mining, control, telecommunications, and
thereby an efficient technique for achieving similar or even many others. To date, applications of PSO algorithms are
better classification accuracy by not using all features but reported in hundreds of publications [46, 59]. Although the
instead selecting only a small set of relevant features from main use of PSO is to solve unconstrained, single-objective
a wide range of features [58]. The key challenges in this optimization problems, a development has been made to
area include novel approaches like Monte Carlo methods PSO algorithms for solving constrained problems, multi-
[22] that simultaneously minimize the number of chosen objective optimization problems, problems with dynamically
features and maximize the classification accuracy in PSO changing landscapes, discrete problems, and for finding mul-
applications. tiple solutions as well [16]. However, several PSO methods
Memory requirement All living systems have memory and applications can be sophisticated in some areas as future
as an essential feature. According to the evolutionism per- research directions in this topic as follows:
spective, it is worthwhile examining the potential helpful Nuero Fuzzy Network (NFN) Nuero Fuzzy Network
role of historical memory in the process of EAs. That is (NFN) is an intelligent method for system identification,

13
2554 A. G. Gad

modelling, prediction, and control. In NFN, gradient-based affected by the correction rate [196] is another major trend
algorithms are generally used for training. However, these for future research.
algorithms have some disadvantages concerning, for exam- Computational biology A long DNA chain first needs
ple, getting stuck at local minima, which needs for complex to be divided into subset fragments for determining its
gradient computations to be conquered. Accordingly, Kara- sequence. Therefore, combinatorial optimization researchers
kuzu et al. [88] used improved PSO to introduce the first used the DNA Fragment Assembly method (DFA) to solve
embedded high-speed, low-cost implementation of NFN the NP-hard problem of reassembling the fragment. The
hardware through online training. It has been observed that DFA problem is solved by applying the overlap-layout-con-
that the effectiveness of the proposed NFN implementation sensus model to maximize the overlapping score measure-
is similar to other approaches in the literature, thereby gener- ment using a memetic PSO algorithm based on two initiali-
ating a novel idea for future research. In a different approach zation operators, as well as the local search operator [67].
to NFNs, Vijay and Jena [186] utilized the PSO to minimize On 19 DNA fragment datasets, the results revealed that the
the quadratic performance indices for obtaining optimal slid- PSO algorithm combining SA-based variable neighborhood
ing control parameters and PID parameters. In this study, local search and TS achieves the best overlap scores. Future
Lyapunov stability theorem was used to improve the stabil- works could address interesting issues, such as reducing
ity of the system by integrating sliding mode control into the computational time by using DNA sequence compres-
artificial neuro fuzzy inference system. Thus, it can be con- sion, improving the initialization method of the algorithm,
cluded that NFN can be applied to the robot manipulator for using an alternative search approach, and so on. In a simi-
real-time control, where good control performance can be lar biological application, some variants of the PSO algo-
obtained by adaptively changing sliding control parameters rithm have been utilized to extensively tackle the molecular
under different input disturbances. docking problem and the problem of conformational search
Color image processing In this regard, Gaussian PSO in protein-ligands [55, 136]. Ultimately, the literature has
(GPSO) was used to develop a color image quantization reported significant results. Thus, it is natural to solve bio-
algorithm [15]. In the proposed algorithm, each particle in logical problems by applying the PSO algorithm.
the swarm is randomly initialized with k centroids (i.e. color Recommender systems PSO was utilized as a tuning
triplets). The cluster with the closest centroid to a pixel in mechanism in a further area in which software tools are cre-
the image is recognized to assign this pixel to that cluster. ated to develop recommendations to entrepreneurs or even
The centroids obtained by K-means clustering are refined end users. Explicit feedback data (i.e., votes or ratings) are
by applying the GPSO. Experimentation showed that using usually used to build most of the existing recommender
GPSO method for the color image quantization gives sig- models. However, real-life scenarios do not always contain
nificantly better image quality than conventional PSO-based explicit feedback data. For example, a hybrid music recom-
approaches. In anther work, a multi-level color image thresh- mender system was suggested based on implicit feedback
olding algorithm was developed on the bases of GLLA histo- data by utilizing graph-based algorithms for making songs
gram and Tsallis-Havrda-Charvát entropy of degree 𝛼 [27]. recommendations based on the user-s preferences and behav-
Compared with the existing models, the proposed model ior [90]. Furthermore, PSO-based web-page recommenda-
achieved better performance results by applying the PSO tion system was developed on health-care multimedia data
algorithm. However, so far only a limited number of studies to track user navigation behavior by utilizing semantic web
reported how to apply together the quantization and multi- mining [121]. In the future work, these systems can be fur-
level thresholding techniques to a color image. ther extended to include other E-health care application, and/
Image registration Image registration is defined as tak- or social networking sites like twitter and Facebook.
ing several 2D images from various sources, such as Com-
puter Assisted Tomography (CAT) and Magnetic Resonance
Imaging (MRI) scans, and combining them into a 3D image. 9 Conclusions and Recommendations
Recently, a hybrid approach for registering medical images
has been developed by employing a PSO method [1] and In this literature, the exploration query on 3,632 papers pub-
an adjusted mutual information as a similarity index, or lished between 2017 and 2019 was executed to fulfill the
general meta-heuristics [184]. However, there are still trend proposed SR process. 2,140 papers related to PSO methods
applications for future research, including registering the and applications were selected. Amongst them, I briefly
images of a printed circuit board placed on a conveyor belt discussed and analyzed only 84 papers to retain the proper
using an improved scale invariant feature transform, feature length of the paper. According to AQ1, on the one hand, the
extraction technique combined with PSO [39]. Also, using PSO variants have the most percentage of the PSO methods
PSO algorithm for remote sensing images registration less in the literature by 42% of quota. Of course, hybridization
techniques have 32%, and improved PSO techniques have

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2555

26% of portions of PSO methods. On the other hand, the Conflict of interest The single/corresonding author declares no con-
general approach has the most percentage of the applica- flict of interest.
tion domains in the literature by 27% of quota, emphasiz-
Open Access This article is licensed under a Creative Commons Attri-
ing the broad application of PSO in diverse subject areas. bution 4.0 International License, which permits use, sharing, adapta-
Also, industrial applications, environmental applications, tion, distribution and reproduction in any medium or format, as long
smart city applications, health-care applications, commer- as you give appropriate credit to the original author(s) and the source,
cial applications have 20%, 17%, 16%, 11%, and 9% por- provide a link to the Creative Commons licence, and indicate if changes
were made. The images or other third party material in this article are
tions of all existing PSO applications, respectively. It has included in the article's Creative Commons licence, unless indicated
also been observed that electrical engineering approaches otherwise in a credit line to the material. If material is not included in
have the highest percentage, with 13 studies, and prediction, the article's Creative Commons licence and your intended use is not
image processing, and mechanical engineering approaches permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
are with 6 studies, based on AQ2. According to AQ3, 47% copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
of the research studies applied simulation tools to evaluate
the case studies presented in the PSO environment.
According to the adopted SR process, I must acknowledge References
that not all present studies may have been analyzed; non-
peer reviewed, non-English book chapters, survey articles, 1. Abdel-Basset M, Fakhry AE, El-Henawy I, Qiu T, Sangaiah AK
and editorial papers were omitted. In this study, the findings (2017) Feature and intensity based medical image registration
of more than 120 authors and diverse studies were consid- using particle swarm optimization. J Med Syst 41(12):197
2. Abdelkader HE, Gad AG, Abohany AA, Sorour SE (2022) An
ered to perform a comprehensive research of the methods efficient data mining technique for assessing satisfaction level of
and application approaches in PSO. However, as the research online learning for higher education students during the covid-
ended in November 2019, it is not possible to guarantee that 19. IEEE Access
all relevant seminal works are covered herein, especially 3. Aberbour J, Graba M, Kheldoun A (2015) Effect of cost function
and pso topology selection on the optimum design of pid param-
with the increasing number of studies in this field. eters for the avr system. In: 2015 4th international conference on
Finally, the SR method presented in this review focused electrical engineering (ICEE). IEEE, pp 1–5
primarily on PSO methods and applications. Through this 4. Abid S, Zafar A, Khalid R, Javaid S, Qasim U, Khan ZA, Javaid
study, the PSO methods and applications were comprehen- N (2017) Managing energy in smart homes using binary particle
swarm optimization. In: Conference on complex, intelligent, and
sively understood. In addition, open issues and future per- software intensive systems. Springer, pp 189–196
spectives were considered. However, the PSO algorithm still 5. Adhikari M, Srirama SN (2019) Multi-objective accelerated par-
needs further research in the following areas: ticle swarm optimization with a container-based scheduling for
internet-of-things in cloud environment. J Netw Comput Appl
137:35–61
• Studying the effects of heterogeneity in PSO algorithms. 6. Al-Thanoon NA, Qasim OS, Algamal ZY (2019) A new hybrid
• Hybridizing PSO with novel optimization techniques, firefly algorithm and particle swarm optimization for tun-
such as Salp Swarm Algorithm (SSA) [130], Whale ing parameter estimation in penalized support vector machine
Optimization Algorithm (WOA) [129], Lion Optimiza- with application in chemometrics. Chemom Intell Lab Syst
184:142–152
tion Algorithm (LOA) [204], Elephant Herding Optimi- 7. Alam S, Dobbie G, Koh YS, Riddle P, Rehman SU (2014)
zation (EHO) [191], and Jaya Algorithm (JA) [63, 153]. Research on particle swarm optimization based clustering: a sys-
• Implementation of innovated smart city applications, tematic review of literature and techniques. Swarm Evol Comput
such as smart metering, smart farming, smart logistics, 17:1–13
8. Ali Ghorbani M, Kazempour R, Chau KW, Shamshirband S,
and smart buildings. Taherei Ghazvinei P (2018) Forecasting pan evaporation with
• Further applications for addressing different issues in an integrated artificial neural network quantum-behaved particle
cloud computing. swarm optimization model: A case study in talesh, northern iran.
• Addressing more complex/large-scale real-world prob- Eng Appl Comput Fluid Mech 12(1):724–737
9. Ali Yahya A (2018) Centroid particle swarm optimisation for
lems, NP-hard problems, and discrete optimization prob- high-dimensional data classification. J Exp Theor Artif Intell
lems to discover new limitations of PSO. 30(6):857–886
10. Alnaqi AA, Moayedi H, Shahsavar A, Nguyen TK (2019) Pre-
diction of energetic performance of a building integrated pho-
tovoltaic/thermal system thorough artificial neural network and
hybrid particle swarm optimization models. Energy Convers
Manag 183:137–148
Funding Open access funding provided by The Science, Technology & 11. Alswaitti M, Albughdadi M, Isa NAM (2018) Density-based par-
Innovation Funding Authority (STDF) in cooperation with The Egyp- ticle swarm optimization algorithm for data clustering. Expert
tian Knowledge Bank (EKB). Syst Appl 91:170–186

Declaration

13
2556 A. G. Gad

12. Aydilek IB (2018) A hybrid firefly and particle swarm optimiza- 31. Chen CH, Liu TK, Chou JH (2014) A novel crowding genetic
tion algorithm for computationally expensive numerical prob- algorithm and its applications to manufacturing robots. IEEE
lems. Appl Soft Comput 66:232–249 Trans Ind Inf 10(3):1705–1716
13. Banks A, Vincent J, Anyakoha C (2007) A review of particle 32. Chen K, Zhou F, Yin L, Wang S, Wang Y, Wan F (2018) A
swarm optimization. Part i: background and development. Nat hybrid particle swarm optimizer with sine cosine acceleration
Comput 6(4):467–484 coefficients. Inf Sci 422:218–241
14. Banks A, Vincent J, Anyakoha C (2008) A review of particle 33. Chen S, Jq Wang, Hy Zhang (2019) A hybrid pso-svm model
swarm optimization. Part ii: hybridisation, combinatorial, multic- based on clustering algorithm for short-term atmospheric pol-
riteria and constrained optimization, and indicative applications. lutant concentration forecasting. Technol Forecast Soc Chang
Nat Comput 7(1):109–124 146:41–54
15. Barman D, Hasnat A, Sarkar S, Murshidanad MAR (2016) Color 34. Chen Y, Li L, Peng H, Xiao J, Wu Q (2018) Dynamic multi-
image quantization using gaussian particle swarm optimization swarm differential learning particle swarm optimizer. Swarm
(ciq-gpso). In: 2016 international conference on inventive com- Evol Comput 39:209–221
putation technologies (ICICT). IEEE, vol 1, pp 1–4 35. Chernbumroong S, Cang S, Yu H (2014) Genetic algorithm-
16. Beheshti Z, Shamsuddin SM, Hasan S (2015) Memetic binary based classifiers fusion for multisensor activity recognition of
particle swarm optimization for discrete optimization problems. elderly people. IEEE J Biomed Health Inform 19(1):282–289
Inf Sci 299:58–84 36. Clerc M, Kennedy J (2002) The particle swarm-explosion, stabil-
17. Beni G (1988) The concept of cellular robotic system. In: Pro- ity, and convergence in a multidimensional complex space. IEEE
ceedings IEEE international symposium on intelligent control Trans Evol Comput 6(1):58–73
1988. IEEE, pp 57–62 37. Colorni A, Dorigo M, Maniezzo V et al (1992) Distributed
18. Beni G, Hackwood S (1992) Stationary waves in cyclic swarms. optimization by ant colonies. In: Proceedings of the first Euro-
In: Proceedings of the 1992 IEEE international symposium on pean conference on artificial life, Cambridge, MA, vol 142, pp
intelligent control. IEEE, pp 234–242 134–142
19. Beni G, Wang J (1993) Swarm intelligence in cellular robotic 38. Cui H, Shu M, Song M, Wang Y (2017) Parameter selection and
systems. In: Robots and biological systems: towards a new bion- performance comparison of particle swarm optimization in sen-
ics? Springer, pp 703–712 sor networks localization. Sensors 17(3):487
20. Benioff P (1980) The computer as a physical system: a micro- 39. Dai L, Guan Q, Liu H (2018) Robust image registration of
scopic quantum mechanical Hamiltonian model of computers as printed circuit boards using improved sift-pso algorithm. J Eng
represented by turing machines. J Stat Phys 22(5):563–591 16:1793–1797
21. Bernardino HS, Barbosa HJ, Fonseca LG (2011) Surrogate- 40. Del Valle Y, Venayagamoorthy GK, Mohagheghi S, Hernandez
assisted clonal selection algorithms for expensive optimization JC, Harley RG (2008) Particle swarm optimization: basic con-
problems. Evol Intel 4(2):81–97 cepts, variants and applications in power systems. IEEE Trans
22. Beskos A, Crisan D, Jasra A, Kamatani K, Zhou Y (2017) A Evol Comput 12(2):171–195
stable particle filter for a class of high-dimensional state-space 41. Deng W, Yao R, Zhao H, Yang X, Li G (2019) A novel intel-
models. Adv Appl Probab 49(1):24–48 ligent diagnosis method using optimal ls-svm with improved pso
23. Bhattacharya A, Goswami RT, Mukherjee K (2018) A feature algorithm. Soft Comput 23(7):2445–2462
selection technique based on rough set and improvised pso algo- 42. Dorigo M, Bonabeau E, Theraulaz G (2000) Ant algorithms and
rithm (psors-fs) for permission based detection of android mal- stigmergy. Future Gen Comput Syst 16(8):851–871
wares. Int J Mach Learn Cybern, pp 1–15 43. Durán-Rosal AM, Gutiérrez PA, Carmona-Poyato Á, Hervás-
24. Bhattacharya A, Goswami RT, Mukherjee K (2019) A feature Martínez C (2019) A hybrid dynamic exploitation barebones par-
selection technique based on rough set and improvised pso algo- ticle swarm optimisation algorithm for time series segmentation.
rithm (psors-fs) for permission based detection of android mal- Neurocomputing 353:45–55
wares. Int J Mach Learn Cybern 10(7):1893–1907 44. Eberhart R, Kennedy J (1995) A new optimizer using particle
25. Bonabeau E, Marco DdRDF, Dorigo M, Théraulaz G, Theraulaz swarm theory. In: MHS’95. Proceedings of the sixth interna-
G et al (1999) Swarm intelligence: from natural to artificial sys- tional symposium on micro machine and human science. IEEE,
tems, 1st edn. Oxford University Press, Oxford pp 39–43
26. Bonyadi MR, Michalewicz Z (2015) Stability analysis of the par- 45. Ehteram M, Binti Othman F, Mundher Yaseen Z, Abdulmoh-
ticle swarm optimization without stagnation assumption. IEEE sin Afan H, Falah Allawi M, Najah Ahmed A, Shahid S, Singh
Trans Evol Comput 20(5):814–819 PV, El-Shafie A (2018) Improving the muskingum flood routing
27. Borjigin S, Sahoo PK (2019) Color image segmentation based method using a hybrid of particle swarm optimization and bat
on multi-level tsallis-havrda-charvát entropy and 2d histogram algorithm. Water 10(6):807
using pso algorithms. Pattern Recogn 92:107–118 46. Elsheikh A, Elaziz MA (2019) Review on applications of particle
28. Camci E, Kripalani DR, Ma L, Kayacan E, Khanesar MA (2018) swarm optimization in solar energy systems. Int J Environ Sci
An aerial robot for rice farm quality inspection with type-2 fuzzy Technol 16(2):1159–1170
neural networks tuned by particle swarm optimization-sliding 47. Emary E, Zawbaa HM, Sharawi M (2019) Impact of Lévy
mode control hybrid algorithm. Swarm Evol Comput 41:1–8 flight on modern meta-heuristic optimizers. Appl Soft Comput
29. Cao Y, Ye Y, Zhao H, Jiang Y, Wang H, Shang Y, Wang J (2018) 75:775–789
Remote sensing of water quality based on hj-1a hsi imagery with 48. Esmin AA, Lambert-Torres G, De Souza AZ (2005) A hybrid
modified discrete binary particle swarm optimization-partial least particle swarm optimization applied to loss power minimization.
squares (mdbpso-pls) in inland waters: a case in weishan lake. IEEE Trans Power Syst 20(2):859–866
Eco Inform 44:21–32 49. Esmin AA, Coelho RA, Matwin S (2015) A review on particle
30. Cao Y, Zhang H, Li W, Zhou M, Zhang Y, Chaovalitwongse swarm optimization algorithm and its variants to clustering high-
WA (2018) Comprehensive learning particle swarm optimiza- dimensional data. Artif Intell Rev 44(1):23–45
tion algorithm with local search for multimodal functions. IEEE 50. Ewees AA, Elaziz MA, Houssein EH (2018) Improved grass-
Trans Evol Comput hopper optimization algorithm using opposition-based learning.
Expert Syst Appl 112:156–172

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2557

51. Fan JC, Li Y, Tang LY, Wu GK (2018) Roughpso: rough set- optimization for feature selection. J Ambient Intelli Hum Com-
based particle swarm optimisation. Int J Bio Inspir Comput put, pp 1–15
12(4):245–253 71. Imran M, Hashim R, Khalid NEA (2013) An overview of particle
52. Ferdowsi A, Farzin S, Mousavi SF, Karami H (2019) Hybrid bat swarm optimization variants. Procedia Eng 53:491–496
& particle swarm algorithm for optimization of labyrinth spill- 72. Ismail FH, Houssein EH, Hassanien AE (2018) Chaotic bird
way based on half & quarter round crest shapes. Flow Measure- swarm optimization algorithm. In: International conference
ment and Instrumentation on advanced intelligent systems and informatics. Springer, pp
53. Fister I, Strnad D, Yang XS (2015) Adaptation and hybridization 294–303
in nature-inspired algorithms. In: Adaptation and hybridization 73. Issa M, Hassanien AE, Oliva D, Helmi A, Ziedan I, Alzohairy
in computational intelligence. Springer, pp 3–50 A (2018) Asca-pso: adaptive sine cosine optimization algorithm
54. Gao H, Xu W (2011) Particle swarm algorithm with hybrid muta- integrated with particle swarm for pairwise local sequence align-
tion strategy. Appl Soft Comput 11(8):5129–5142 ment. Expert Syst Appl 99:56–70
55. García-Nieto J, López-Camacho E, García-Godoy MJ, Nebro 74. Jain I, Jain VK, Jain R (2018) Correlation feature selection based
AJ, Aldana-Montes JF (2019) Multi-objective ligand-protein improved-binary particle swarm optimization for gene selection
docking with particle swarm optimizers. Swarm Evol Comput and cancer classification. Appl Soft Comput 62:203–215
44:439–452 75. Jana B, Mitra S, Acharyya S (2019) Repository and mutation
56. Ghaddar B, Naoum-Sawaya J (2018) High dimensional data clas- based particle swarm optimization (rmpso): a new pso variant
sification and feature selection using support vector machines. applied to reconstruction of gene regulatory network. Appl Soft
Eur J Oper Res 265(3):993–1004 Comput 74:330–355
57. Ghorbani N, Kasaeian A, Toopshekan A, Bahrami L, Maghami A 76. Ji B, Song X, Sciberras E, Cao W, Hu Y, Pickert V (2014) Multi-
(2018) Optimizing a hybrid wind-pv-battery system using ga-pso objective design optimization of igbt power modules considering
and mopso for reducing cost and increasing reliability. Energy power cycling and thermal cycling. IEEE Trans Power Electron
154:581–591 30(5):2493–2504
58. Gu S, Cheng R, Jin Y (2018) Feature selection for high-dimen- 77. Jiang H, Kwong C, Park W, Yu K (2018) A multi-objective pso
sional classification using a competitive swarm optimizer. Soft approach of mining association rules for affective design based
Comput 22(3):811–822 on online customer reviews. J Eng Des 29(7):381–403
59. Hajihassani M, Armaghani DJ, Kalatehjari R (2018) Applications 78. Jiao R, Huang X, Ouyang H, Li G, Zheng Q, Jiang Z (2019) Opti-
of particle swarm optimization in geotechnical engineering: a mal electric business centre location by centre-decentre quantum
comprehensive review. Geotech Geol Eng 36(2):705–722 particle swarm optimization. Syst Sci Control Eng 7(1):222–233
60. Hino T, Ito S, Liu T, Maeda M (2016) Set-based particle swarm 79. Jin Y, Sendhoff B (2009) A systems approach to evolutionary
optimization with status memory for knapsack problem. Artif multiobjective structural optimization and beyond. IEEE Comput
Life Robot 21(1):98–105 Intell Mag 4(3):62–76
61. Hoang TT, Cho MY, Alam MN, Vu QT (2018) A novel differen- 80. Jin Y, Olhofer M, Sendhoff B (2002) A framework for evolution-
tial particle swarm optimization for parameter selection of sup- ary optimization with approximate fitness functions. IEEE Trans
port vector machines for monitoring metal-oxide surge arrester Evol Comput 6(5):481–494
conditions. Swarm Evol Comput 38:120–126 81. Jmal S, Haddar B, Chabchoub H (2019) Apply the quantum par-
62. Houssein EH, Gad AG, Hussain K, Suganthan PN (2021) Major ticle swarm optimization for the k-traveling repairman problem.
advances in particle swarm optimization: theory, analysis, and Soft Computing, pp 1–14
application. Swarm Evol Comput 63:100868 82. Jordehi AR (2019) Binary particle swarm optimisation with
63. Houssein EH, Gad AG, Wazery YM (2021) Jaya algorithm and quadratic transfer function: a new binary optimisation algorithm
applications: a comprehensive review. Metaheuristics and Opti- for optimal scheduling of appliances in smart homes. Appl Soft
mization in Computer and Electrical Engineering, pp 3–24 Comput
64. Houssein EH, Gad AG, Wazery YM, Suganthan PN (2021) Task 83. Juang CF (2004) A hybrid of genetic algorithm and particle
scheduling in cloud computing based on meta-heuristics: review, swarm optimization for recurrent network design. IEEE Trans
taxonomy, open challenges, and future trends. Swarm Evol Com- Syst Man Cybernet Part B (Cybern) 34(2):997–1006
put 62:100841 84. Junior FEF, Yen GG (2019) Particle swarm optimization of deep
65. Hu W, Wang H, Qiu Z, Nie C, Yan L (2018) A quantum particle neural networks architectures for image classification. Swarm
swarm optimization driven urban traffic light scheduling model. and Evolutionary Computation
Neural Comput Appl 29(3):901–911 85. Kane BE (1998) A silicon-based nuclear spin quantum computer.
66. Huang H, Lv L, Ye S, Hao Z (2019) Particle swarm optimiza- Nature 393(6681):133
tion with convergence speed controller for large-scale numerical 86. Kang Q, Xiong C, Zhou M, Meng L (2018) Opposition-based
optimization. Soft Comput 23(12):4421–4437 hybrid strategy for particle swarm optimization in noisy environ-
67. Huang KW, Chen JL, Yang CS, Tsai CW (2015) A memetic par- ments. IEEE Access 6:21888–21900
ticle swarm optimization algorithm for solving the dna fragment 87. Karaboga D (2005) An idea based on honey bee swarm for
assembly problem. Neural Comput Appl 26(3):495–506 numerical optimization. Technical report, Technical report-tr06,
68. Huang L, Guan K, Xu T, Zhang J, Wang Q (2019) Investigation Erciyes university, engineering faculty, computer
of the mechanical properties of steel using instrumented indenta- 88. Karakuzu C, Karakaya F, Çavuşlu MA (2016) Fpga implementa-
tion test with simulated annealing particle swarm optimization. tion of neuro-fuzzy system with improved pso learning. Neural
Theor Appl Fract Mech 102:116–121 Netw 79:128–140
69. Hussien AG, Houssein EH, Hassanien AE (2017) A binary whale 89. Karbassi Yazdi A, Kaviani MA, Emrouznejad A, Sahebi H
optimization algorithm with hyperbolic tangent fitness function (2019) A binary particle swarm optimization algorithm for ship
for feature selection. In: 2017 eighth international conference on routing and scheduling of liquefied natural gas transportation.
intelligent computing and information systems (ICICIS). IEEE, Transp Lett, pp 1–10
pp 166–172 90. Katarya R, Verma OP (2018) Efficient music recommender sys-
70. Ibrahim RA, Ewees AA, Oliva D, Elaziz MA, Lu S (2018) tem using context graph and particle swarm. Multimed Tools
Improved salp swarm algorithm based on particle swarm Appl 77(2):2673–2687

13
2558 A. G. Gad

91. Kennedy J, Eberhart R (1995) Particle swarm optimization (pso). based on particle swarm optimization. Expert Syst Appl
In: Proceedings of IEEE international conference on neural net- 37(3):2629–2636
works, Perth, Australia, pp 1942–1948 110. Liu F, Huang H, Li X, Hao Z (2019) Automated test data gen-
92. Kennedy J, Eberhart R, Shi Y (2001) Swarm intelligence. Mor- eration based on particle swarm optimisation with convergence
gan Kaufmann, San Francisco speed controller. CAAI Trans Intell Technol 2(2):73–79
93. Kennedy J, Eberhart R (1995) Particle swarm optimization. In: 111. Liu G, Chen W, Chen H (2019) Quantum particle swarm with
Proceedings of IEEE international conference on neural networks teamwork evolutionary strategy for multi-objective optimization
IV, vol 1000, p 33 on electro-optical platform. IEEE Access 7:41205–41219
94. Kothari V, Anuradha J, Shah S, Mittal P (2011) A survey on 112. Liu Q, Wei W, Yuan H, Zhan ZH, Li Y (2016) Topology selec-
particle swarm optimization in feature selection. In: International tion for particle swarm optimization. Inf Sci 363:154–173
conference on computing and communication systems.. Springer, 113. Lopes RF, Costa FF, Oliveira A, Lima ACdC (2018) Algorithm
pp 192–201 based on particle swarm applied to electrical load scheduling in
95. Kour VP, Arora S (2019) Particle swarm optimization based sup- an industrial setting. Energy 147:1007–1015
port vector machine (p-svm) for the segmentation and classifica- 114. López MG, Ponce P, Soriano LA, Molina A, Rivas JJR (2019)
tion of plants. IEEE Access 7:29374–29385 A novel fuzzy-pso controller for increasing the lifetime in
96. Kulkarni RV, Venayagamoorthy GK (2011) Particle swarm power electronics stage for brushless dc drives. IEEE Access
optimization in wireless-sensor networks: a brief survey. IEEE 7:47841–47855
Trans Syst Man Cybern Part C (Applications and Reviews) 115. Lorenzo PR, Nalepa J, Ramos LS, Pastor JR (2017) Hyper-
41(2):262–267 parameter selection in deep neural networks using parallel par-
97. Kumar S, Pal SK, Singh R (2019) A novel hybrid model based ticle swarm optimization. In: Proceedings of the genetic and
on particle swarm optimisation and extreme learning machine for evolutionary computation conference companion. ACM, pp
short-term temperature prediction using ambient sensors. Sustain 1864–1871
Cities Soc 49:101601 116. Løvbjerg M, Rasmussen TK, Krink T (2001) Hybrid particle
98. Laskar NM, Guha K, Chatterjee I, Chanda S, Baishnab KL, Paul swarm optimiser with breeding and subpopulations. In: Proceed-
PK (2019) Hwpso:a new hybrid whale-particle swarm optimiza- ings of the 3rd annual conference on genetic and evolutionary
tion algorithm and its application in electronic design optimiza- computation. Morgan Kaufmann Publishers Inc., pp 469–476
tion problems. Appl Intell 49(1):265–291 117. Luo W, Sun J, Bu C, Liang H (2016) Species-based particle
99. Le LT, Nguyen H, Zhou J, Dou J, Moayedi H et al (2019) Esti- swarm optimizer enhanced by memory for dynamic optimiza-
mating the heating load of buildings for smart city planning using tion. Appl Soft Comput 47:130–140
a novel artificial intelligence technique pso-xgboost. Appl Sci 118. Ma K, Hu S, Yang J, Xu X, Guan X (2018) Appliances schedul-
9(13):2714 ing via cooperative multi-swarm pso under day-ahead prices and
100. Li H, Zhang S, Zhang C, Li P, Cropp R (2017) A novel unsuper- photovoltaic generation. Appl Soft Comput 62:504–513
vised Lévy flight particle swarm optimization (ulpso) method for 119. Mahdavi S, Rahnamayan S, Deb K (2018) Opposition based
multispectral remote-sensing image classification. Int J Remote learning: a literature review. Swarm Evol Comput 39:1–23
Sens 38(23):6970–6992 120. Maiyar LM, Thakkar JJ (2019) Environmentally conscious
101. Li J, Zhang J, Jiang C, Zhou M (2015) Composite particle swarm logistics planning for food grain industry considering wastages
optimizer with historical memory for function optimization. employing multi objective hybrid particle swarm optimization.
IEEE Trans Cybern 45(10):2350–2363 Transp Res Part E Log Transp Rev 127:220–248
102. Li L, Qin L, Qu X, Zhang J, Wang Y, Ran B (2019) Day-ahead 121. Manikandan R, Saravanan V (2019) A novel approach on par-
traffic flow forecasting based on a deep belief network optimized ticle agent swarm optimization (paso) in semantic mining for
by the multi-objective particle swarm algorithm. Knowl Based web page recommender system of multimedia data: a health care
Syst 172:1–14 perspective. Multimedia Tools and Applications, pp 1–23
103. Li W (2018) Improving particle swarm optimization based on 122. Mansouri N, Zade BMH, Javidi MM (2019) Hybrid task sched-
neighborhood and historical memory for training multi-layer uling strategy for cloud computing by modified particle swarm
perceptron. Information 9(1):16 optimization and fuzzy theory. Comput Ind Eng 130:597–633
104. Li Y, Bai X, Jiao L, Xue Y (2017) Partitioned-cooperative quan- 123. Marini F, Walczak B (2015) Particle swarm optimization (pso).
tum-behaved particle swarm optimization based on multilevel A tutorial. Chemom Intell Lab Syst 149:153–165
thresholding applied to medical image segmentation. Appl Soft 124. Mavrovouniotis M, Li C, Yang S (2017) A survey of swarm intel-
Comput 56:345–356 ligence for dynamic optimization: algorithms and applications.
105. Li Z, Shi K, Dey N, Ashour AS, Wang D, Balas VE, McCauley Swarm Evol Comput 33:1–17
P, Shi F (2017) Rule-based back propagation neural networks 125. Mehmood Y, Sadiq M, Shahzad W, Amin F (2018) Fitness-based
for various precision rough set presented kansei knowledge pre- acceleration coefficients to enhance the convergence speed of
diction: a case study on shoe product form features extraction. novel binary particle swarm optimization. In: 2018 international
Neural Comput Appl 28(3):613–630 conference on frontiers of information technology (FIT). IEEE,
106. Lin A, Sun W, Yu H, Wu G, Tang H (2019) Global genetic learn- pp 355–360
ing particle swarm optimization with diversity enhancement by 126. Melton RG (2018) Differential evolution/particle swarm
ring topology. Swarm Evol Comput 44:571–583 optimizer for constrained slew maneuvers. Acta Astronaut
107. Lin G, Guan J, Li Z, Feng H (2019) A hybrid binary particle 148:246–259
swarm optimization with tabu search for the set-union knapsack 127. Millonas MM, et al (1993) Swarms, phase transitions, and collec-
problem. Expert Syst Appl tive intelligence (paper 1); and a nonequilibrium statistical field
108. Lin Q, Ma Y, Chen J, Zhu Q, Coello CAC, Wong KC, Chen F theory of swarms and other spatially extended complex systems
(2018) An adaptive immune-inspired multi-objective algorithm (paper 2). Technical rep
with multiple differential evolution strategies. Inf Sci 430:46–64 128. Miranda V, Fonseca N (2002) Epso-evolutionary particle swarm
109. Lin TL, Horng SJ, Kao TW, Chen YH, Run RS, Chen RJ, Lai optimization, a new algorithm with applications in power sys-
JL, Kuo IH (2010) An efficient job-shop scheduling algorithm tems. In IEEE/PES transmission and distribution conference and
exhibition, vol 2. IEEE, pp 745–750

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2559

129. Mirjalili S, Lewis A (2016) The whale optimization algorithm. of support vector machine and genetic algorithms and particle
Adv Eng Softw 95:51–67 swarm. Arab J Geosci 12(9):277
130. Mirjalili S, Gandomi AH, Mirjalili SZ, Saremi S, Faris H, Mir- 150. Rahman NHA, Zobaa AF (2017) Integrated mutation strategy
jalili SM (2017) Salp swarm algorithm: a bio-inspired optimizer with modified binary pso algorithm for optimal pmus placement.
for engineering design problems. Adv Eng Softw 114:163–191 IEEE Trans Ind Inf 13(6):3124–3133
131. Moayedi H, Mehrabi M, Mosallanezhad M, Rashid ASA, Prad- 151. Raj S, Ray KC (2017) Ecg signal analysis using dct-based
han B (2018) Modification of landslide susceptibility mapping dost and pso optimized svm. IEEE Trans Instrum Meas
using optimized pso-ann technique. Engineering with Comput- 66(3):470–478
ers, pp 1–18 152. Ramya ST, Arunagiri B, Rangarajan P (2017) Novel effective
132. Mohebbi A, Achiche S, Baron L (2019) Integrated and concur- x-path particle swarm optimization based deprived video data
rent detailed design of a mechatronic quadrotor system using a retrieval for smart city. Cluster Computing, pp 1–10
fuzzy-based particle swarm optimization. Eng Appl Artif Intell 153. Rao R (2016) Jaya: a simple and new optimization algorithm for
82:192–206 solving constrained and unconstrained optimization problems.
133. Mokhtari H, Noroozi A (2018) An efficient chaotic based pso for Int J Ind Eng Comput 7(1):19–34
earliness/tardiness optimization in a batch processing flow shop 154. Rashid M, Baig AR (2010) Psogp: a genetic programming based
scheduling problem. J Intell Manuf 29(5):1063–1081 adaptable evolutionary hybrid particle swarm optimization. Int J
134. Nagra AA, Han F, Ling QH (2018) An improved hybrid self- Innov Comput Inf Control 6(1):287–296
inertia weight adaptive particle swarm optimization algorithm 155. Rojas-García ÁA, Hernández-Aguirre A, Valdez SI (2019) On
with local search. Engineering Optimization, pp 1–18 the selection of the optimal topology for particle swarm optimi-
135. Nenavath H, Jatoth RK, Das S (2018) A synergy of the sine- zation: a study of the tree as the universal topology. In: Proceed-
cosine algorithm and particle swarm optimizer for improved ings of the genetic and evolutionary computation conference.
global optimization and object tracking. Swarm Evol Comput ACM, pp 55–62
43:1–30 156. Rouff CA, Hinchey MG, Truszkowski WF, Rash JL (2006)
136. Ng MC, Fong S, Siu SW (2015) Psovina: the hybrid particle Experiences applying formal approaches in the development of
swarm optimization algorithm for protein-ligand docking. J Bio- swarm-based space exploration systems. Int J Softw Tools Tech-
inform Comput Biol 13(03):1541007 nol Transf 8(6):587–603
137. Nguyen S, Zhang M, Johnston M, Tan KC (2014) Automatic 157. Salajegheh F, Salajegheh E (2019) Psog: enhanced particle
programming via iterated local search for dynamic job shop swarm optimization by a unit vector of first and second order
scheduling. IEEE Trans Cybern 45(1):1–14 gradient directions. Swarm Evol Comput 46:28–51
138. Nobile MS, Cazzaniga P, Besozzi D, Colombo R, Mauri G, Pasi 158. Sameer F, Bakar MA, Zaidan A, Zaidan B (2019) A new algo-
G (2018) Fuzzy self-tuning pso: a settings-free algorithm for rithm of modified binary particle swarm optimization based on
global optimization. Swarm Evol Comput 39:70–85 the Gustafson-Kessel for credit risk assessment. Neural Comput
139. Nouiri M, Bekrar A, Jemai A, Niar S, Ammari AC (2018) An Appl 31(2):337–346
effective and distributed particle swarm optimization algo- 159. Sanchez IAL, Vargas JM, Santos CA, Mendoza MG, Mocte-
rithm for flexible job-shop scheduling problem. J Intell Manuf zuma CJM (2018) Solving binary cutting stock with matheuris-
29(3):603–615 tics using particle swarm optimization and simulated annealing.
140. Pashaei E, Pashaei E, Aydin N (2019) Gene selection using Soft Comput 22(18):6111–6119
hybrid binary black hole algorithm and modified binary particle 160. Saremi S, Mirjalili S, Lewis A, Liew AWC, Dong JS (2018)
swarm optimization. Genomics 111(4):669–686 Enhanced multi-objective particle swarm optimisation for esti-
141. Passino KM (2002) Biomimicry of bacterial foraging for dis- mating hand postures. Knowl Based Syst 158:175–195
tributed optimization and control. IEEE Control Syst Mag 161. Sato M, Fukuyama Y, Iizaka T, Matsui T (2018) Total optimi-
22(3):52–67 zation of energy networks in a smart city by multi-swarm dif-
142. Pawlak Z (1982) International of computer and information sci- ferential evolutionary particle swarm optimization. IEEE Trans
ence. Rough Set 11:341–356 Sustain Energy
143. Poli R, Kennedy J, Blackwell T (2007) Particle swarm optimiza- 162. Serani A, Leotardi C, Iemma U, Campana EF, Fasano G, Diez
tion. Swarm Intell 1(1):33–57 M (2016) Parameter selection in synchronous and asynchronous
144. Pradeepkumar D, Ravi V (2017) Forecasting financial time series deterministic particle swarm optimization for ship hydrodynam-
volatility using particle swarm optimization trained quantile ics problems. Appl Soft Comput 49:313–334
regression neural network. Appl Soft Comput 58:35–52 163. Shahzad F, Masood S, Khan NK (2014) Probabilistic opposi-
145. Qi C, Fourie A, Chen Q (2018) Neural network and particle tion-based particle swarm optimization with velocity clamping.
swarm optimization for predicting the unconfined compres- Knowl Inf Syst 39(3):703–737
sive strength of cemented paste backfill. Constr Build Mater 164. Sheikholeslami F, Navimipour NJ (2017) Service allocation in
159:473–478 the cloud environments using multi-objective particle swarm
146. Qi X, Ju G, Xu S (2018) Efficient solution to the stagnation optimization algorithm based on crowding distance. Swarm Evol
problem of the particle swarm optimization algorithm for phase Comput 35:53–64
diversity. Appl Opt 57(11):2747–2757 165. Shen J, Han L (2019) Design process optimization and profit
147. Qian Q, Wu J, Wang Z (2019) Optimal path planning for two- calculation module development simulation analysis of financial
wheeled self-balancing vehicle pendulum robot based on quan- accounting information system based on particle swarm optimi-
tum-behaved particle swarm optimization algorithm. Personal zation (pso). Information Systems and e-Business Management,
and Ubiquitous Computing, pp 1–11 pp 1–14
148. Qu B, Zhu Y, Jiao Y, Wu M, Suganthan PN, Liang J (2018) A 166. Sivaranjani R, Roomi SMM, Senthilarasi M (2019) Speckle noise
survey on multi-objective evolutionary algorithms for the solu- removal in sar images using multi-objective pso (mopso) algo-
tion of the environmental/economic dispatch problems. Swarm rithm. Appl Soft Comput 76:671–681
Evol Comput 38:1–11 167. Song M, Chen K, Wang J (2018) Three-dimensional wind turbine
149. Rahgoshay M, Feiznia S, Arian M, Hashemi SAA (2019) Simula- positioning using gaussian particle swarm optimization with dif-
tion of daily suspended sediment load using an improved model ferential evolution. J Wind Eng Ind Aerodyn 172:317–324

13
2560 A. G. Gad

168. Srisukkham W, Zhang L, Neoh SC, Todryk S, Lim CP (2017) 186. Vijay M, Jena D (2017) Pso based neuro fuzzy sliding mode
Intelligent leukaemia diagnosis with bare-bones pso based fea- control for a robot manipulator. J Electr Syst Inf Technol
ture optimization. Appl Soft Comput 56:405–419 4(1):243–256
169. Stork J, Friese M, Zaefferer M, Bartz-Beielstein T, Fischbach 187. Villarrubia G, De Paz JF, Chamoso P, De la Prieta F (2018)
A, Breiderhoff B, Naujoks B, Tušar T (2020) Open issues in Artificial neural networks used in optimization problems. Neu-
surrogate-assisted optimization. In: High-performance simula- rocomputing 272:10–16
tion-based optimization. Springer, pp 225–244 188. Wang C, Yu T, Curiel-Sosa JL, Xie N, Bui TQ (2019) Adaptive
170. Sun J, Feng B, Xu W (2004) Particle swarm optimization with chaotic particle swarm algorithm for isogeometric multi-objec-
particles having quantum behavior. In: Proceedings of the tive size optimization of fg plates. Structural and Multidiscipli-
2004 congress on evolutionary computation (IEEE Cat. No. nary Optimization, pp 1–22
04TH8753), vol 1. IEEE, pp 325–331 189. Wang D, Qiu H, Wu CC, Lin WC, Lai K, Cheng SR (2018)
171. Sun L, Song X, Chen T (2019) An improved convergence particle Dominance rule and opposition-based particle swarm optimi-
swarm optimization algorithm with random sampling of control zation for two-stage assembly scheduling with time cumulated
parameters. J Control Sci Eng 2019 learning effect. Soft Computing, pp 1–12
172. Sun Z, Liu Y, Tao L (2018) Attack localization task allocation in 190. Wang D, Tan D, Liu L (2018) Particle swarm optimization algo-
wireless sensor networks based on multi-objective binary particle rithm: an overview. Soft Comput 22(2):387–408
swarm optimization. J Netw Comput Appl 112:29–40 191. Wang GG, Deb S, Coelho LdS (2015) Elephant herding optimi-
173. Suresh S, Lal S (2017) Multilevel thresholding based on chaotic zation. In: 2015 3rd international symposium on computational
Darwinian particle swarm optimization for segmentation of satel- and business intelligence (ISCBI). IEEE, pp 1–5
lite images. Appl Soft Comput 55:503–522 192. Wang H, Peng Mj, Hines JW, Zheng Gy, Liu Yk, Upadhyaya BR
174. Tang B, Han J, Guo G, Chen Y, Zhang S (2019) Building mate- (2019) A hybrid fault diagnosis methodology with support vector
rial prices forecasting based on least square support vector machine and improved particle swarm optimization for nuclear
machine and improved particle swarm optimization. Archit Eng power plants. ISA Trans
Des Manag 15(3):196–212 193. Wang JJ, Liu GY (2019) Saturated control design of a quadro-
175. Tang W, Cha H, Wei M, Tian B (2019) Estimation of surface- tor with heterogeneous comprehensive learning particle swarm
based duct parameters from automatic identification system using optimization. Swarm Evol Comput 46:84–96
the Lévy flight quantum-behaved particle swarm optimization 194. Wang S, Li Y, Yang H (2019) Self-adaptive mutation differential
algorithm. J Electromagn Waves Appl, 1–11 evolution algorithm based on particle swarm optimization. Appl
176. Thabit S, Mohades A (2018) Multi-robot path planning based Soft Comput 81:105496
on multi-objective particle swarm optimization. IEEE Access 195. Wu TY, Lin CH (2014) Low-sar path discovery by particle
7:2138–2147 swarm optimization algorithm in wireless body area networks.
177. Thangaraj R, Pant M, Abraham A, Bouvry P (2011) Particle IEEE Sens J 15(2):928–936
swarm optimization: hybridization perspectives and experimental 196. Wu Y, Miao Q, Ma W, Gong M, Wang S (2017) Psosac: parti-
illustrations. Appl Math Comput 217(12):5208–5226 cle swarm optimization sample consensus algorithm for remote
178. Thangaraj R, Pant M, Abraham A, Snasel V (2012) Modified sensing image registration. IEEE Geosci Remote Sens Lett
particle swarm optimization with time varying velocity vector. 15(2):242–246
Int J Innov Comput Inf Control 8(1):201–218 197. Xu G, Wu ZH, Jiang MZ (2015) Premature convergence of stand-
179. Tharwat A, Hassanien AE (2019) Quantum-behaved particle ard particle swarm optimisation algorithm based on Markov
swarm optimization for parameter optimization of support vec- chain analysis. Int J Wirel Mobile Comput 9(4):377–382
tor machine. J Classif, pp 1–23 198. Xu G, Cui Q, Shi X, Ge H, Zhan ZH, Lee HP, Liang Y, Tai R,
180. Ting T, Yang XS, Cheng S, Huang K (2015) Hybrid metaheuris- Wu C (2019) Particle swarm optimization based on dimensional
tic algorithms: past, present, and future. In: Recent advances in learning strategy. Swarm Evol Comput 45:33–51
swarm intelligence and evolutionary computation. Springer, pp 199. Xu L, Muhammad A, Pu Y, Zhou J, Zhang Y (2019) Frac-
71–83 tional-order quantum particle swarm optimization. PLoS ONE
181. Tizhoosh HR (2005) Opposition-based learning: a new scheme 14(6):e0218285
for machine intelligence. In: International conference on compu- 200. Xu X, Rong H, Trovati M, Liptrott M, Bessis N (2018) Cs-pso:
tational intelligence for modelling, control and automation and chaotic particle swarm optimization algorithm for solving com-
international conference on intelligent agents, web technologies binatorial optimization problems. Soft Comput 22(3):783–795
and internet commerce (CIMCA-IAWTIC’06), vol 1. IEEE, pp 201. Yang XS (2009) Firefly algorithms for multimodal optimization.
695–701 In: International symposium on stochastic algorithms. Springer,
182. Tran B, Zhang M, Xue B (2016) A pso based hybrid feature pp 169–178
selection algorithm for high-dimensional classification. In: 2016 202. Yang XS, Cui Z, Xiao R, Gandomi AH, Karamanoglu M (2013)
IEEE congress on evolutionary computation (CEC). IEEE, pp Swarm intelligence and bio-inspired computation: theory and
3801–3808 applications. Newnes, London
183. Vafashoar R, Meybodi MR (2019) Cellular learning automata 203. Yang Z, Qiu H, Gao L, Cai X, Jiang C, Chen L (2019) A surro-
based bare bones pso with maximum likelihood rotated muta- gate-assisted particle swarm optimization algorithm based on
tions. Swarm Evol Comput 44:680–694 efficient global optimization for expensive black-box problems.
184. Valsecchi A, Bermejo E, Damas S, Cordón O (2018) Metaheuris- Eng Optim 51(4):549–566
tics for medical image registration. Handbook of Heuristics, pp 204. Yazdani M, Jolai F (2016) Lion optimization algorithm (loa):
1079–1101 a nature-inspired metaheuristic algorithm. J Comput Des Eng
185. Vandersypen LM, Steffen M, Breyta G, Yannoni CS, Sherwood 3(1):24–36
MH, Chuang IL (2001) Experimental realization of Shor’s 205. Yi T, Zheng H, Tian Y, Liu Jp (2018) Intelligent prediction of
quantum factoring algorithm using nuclear magnetic resonance. transmission line project cost based on least squares support vec-
Nature 414(6866):883 tor machine optimized by particle swarm optimization. Math-
ematical Problems in Engineering 2018

13
Particle Swarm Optimization Algorithm and Its Applications: A Systematic Review 2561

206. Yoon Y, Kim YH (2013) An efficient genetic algorithm for maxi- 216. Zhang Y, Wang S, Ji G (2015) A comprehensive survey on par-
mum coverage deployment in wireless sensor networks. IEEE ticle swarm optimization algorithm and its applications. Math-
Trans Cybern 43(5):1473–1483 ematical Problems in Engineering 2015
207. Yu H, Tan Y, Zeng J, Sun C, Jin Y (2018) Surrogate-assisted 217. Zhang Y, Zhang Q, Farnoosh A, Chen S, Li Y (2019) Gis-based
hierarchical particle swarm optimization. Inf Sci 454:59–72 multi-objective particle swarm optimization of charging stations
208. Yuan Q, Yin G (2014) Analyzing convergence and rates of for electric vehicles. Energy 169:844–853
convergence of particle swarm optimization algorithms using 218. Zhang YD, Wang S, Dong Z (2014) Classification of Alzheimer
stochastic approximation methods. IEEE Trans Autom Control disease based on structural magnetic resonance imaging by ker-
60(7):1760–1773 nel support vector machine decision tree. Prog Electromagn Res
209. Yue Y, Cao L, Hu J, Cai S, Hang B, Wu H (2019) A novel hybrid 144:171–184
location algorithm based on chaotic particle swarm optimization 219. Zheng J, Lu C, Gao L (2019) Multi-objective cellular particle
for mobile position estimation. IEEE Access 7:58541–58552 swarm optimization for wellbore trajectory design. Appl Soft
210. Zarei A, Mousavi SF, Gordji ME, Karami H (2019) Optimal Comput 77:106–117
reservoir operation using bat and particle swarm algorithm and 220. Zhong Y, Lin J, Wang L, Zhang H (2018) Discrete comprehen-
game theory based on optimal water allocation among consum- sive learning particle swarm optimization algorithm with metrop-
ers. Water Resources Management, pp 1–23 olis acceptance criterion for traveling salesman problem. Swarm
211. Zarrouk R, Bennour IE, Jemai A (2019) A two-level particle Evol Comput 42:77–88
swarm optimization algorithm for the flexible job shop schedul- 221. Zhou A, Qu BY, Li H, Zhao SZ, Suganthan PN, Zhang Q (2011)
ing problem. Swarm Intelligence, pp 1–24 Multiobjective evolutionary algorithms: a survey of the state of
212. Zeng N, Qiu H, Wang Z, Liu W, Zhang H, Li Y (2018) A new the art. Swarm Evol Comput 1(1):32–49
switching-delayed-pso-based optimized svm algorithm for diag- 222. Zhou Z, Ong YS, Lim MH, Lee BS (2007) Memetic algorithm
nosis of Alzheimer’s disease. Neurocomputing 320:195–202 using multi-surrogates for computationally expensive optimiza-
213. Zhang H, Xie J, Hu Q, Shao L, Chen T (2018) A hybrid dpso tion problems. Soft Comput 11(10):957–971
with Lévy flight for scheduling mimo radar tasks. Appl Soft 223. Zhu M, Li J, Chang D, Zhang G, Chen J (2018) Optimization
Comput 71:242–254 of antenna array deployment for partial discharge localization in
214. Zhang X, Zheng X, Cheng R, Qiu J, Jin Y (2018) A competitive substations by hybrid particle swarm optimization and genetic
mechanism based multi-objective particle swarm optimizer with algorithm method. Energies 1(11):1813
fast convergence. Inf Sci 427:63–76
215. Zhang Y, Balochian S, Agarwal P, Bhatnagar V, Housheya OJ Publisher's Note Springer Nature remains neutral with regard to
(2014) Artificial intelligence and its applications. Mathematical jurisdictional claims in published maps and institutional affiliations.
Problems in Engineering 2014

13