mathematics

Article
An Improved Pity Beetle Algorithm for Solving Constrained
Engineering Design Problems
Yu Peng and Xianjun Du *

College of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China;
[email protected]
* Correspondence: [email protected]

Abstract: To cope with increasingly complex models of engineering design problems and to obtain
more accurate design solutions, this paper proposed an improved population-based, bio-inspired
optimization algorithm, called the pity beetle algorithm based on pheromone dispersion model
(PBA-PDM). PBA-PDM enables a local and global search for optimization problems through the
pheromone release mechanisms in female beetles and the interaction relationship between male
beetles. The experimental results compared with other state-of-the-art metaheuristic optimization
algorithms show that PBA-PDM has an ideal performance when dealing with both classical test
functions and CEC2017 benchmark test functions. Then, the PBA-PDM is applied in dealing with
real-world constrained engineering design problems to verify the effectiveness and applicability.
The above experimental results show that the PBA-PDM proposed in this paper is an effective and
efficient algorithm for solving real-world optimization problems.

Keywords: pity beetle algorithm (PBA); pheromone dispersion model (PDM); metaheuristic
algorithm; engineering optimization problems; continuous optimization problems

MSC: 49K35
Citation: Peng, Y.; Du, X. An
Improved Pity Beetle Algorithm for
Solving Constrained Engineering
1. Introduction
Design Problems. Mathematics 2022,
10, 2211. https://doi.org/10.3390/
1.1. Optimization Problem
math10132211 The optimization problem refers to determining the value of the variable in the objec-
tive function under the condition of satisfying the given constraints, so that the dependent
Academic Editor: Frank Werner
variable (fitness value) of the objective function obtains the minimum (maximum) value [1].
Received: 6 June 2022 Wang et al. [2] noted that the optimization process is in fact the search for an optimal value
Accepted: 22 June 2022 that minimizes or maximizes the output of a given system from among all possible values
Published: 24 June 2022 of the system parameters. Therefore, optimization problems can usually be expressed in
Publisher’s Note: MDPI stays neutral
the form of mathematical programming, as follows:
with regard to jurisdictional claims in →
published maps and institutional affil- f or any x = [ x1 , · · · , x Dim ], where li ≤ xi ≤ ui , i = 1, 2, · · · , Dim
, x2
→
iations. consider : min f x
→ (1)
subject to : g j x ≤ 0, j = 1, 2, · · · , m
→
hk x = 0, k = 1, 2, · · · , n
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland. →
where x refers to the decision variable, Dim is the number of variables to be determined
→ in
This article is an open access article
distributed under the terms and
the problem, li and ui are the lower and the upper boundary of objective space, f x is
conditions of the Creative Commons called the objective function, and g j and hk refer to the inequality constraint and equality
Attribution (CC BY) license (https:// constraint, respectively.
creativecommons.org/licenses/by/
4.0/).

Mathematics 2022, 10, 2211. https://doi.org/10.3390/math10132211 https://www.mdpi.com/journal/mathematics

Mathematics 2022, 10, 2211 2 of 38

1.2. Optimization Algorithm

The essence of an optimization algorithm is a search process based on specific inspi-
rations and mechanisms. Such methods extract optimal solutions from the predefined
objective space that satisfy the constraints by means of specific paths and rules. In the face
of non-deterministic problems such as NP-Hard and NPC, researchers prefer to use heuris-
tic algorithms to solve them. The heuristic algorithm is a frame for solving optimization
problems through inductive reasoning based on past experience and experimental analysis,
that is, with the help of some intuitive judgments or trial and error methods, to find a
suboptimal solution that satisfies the accuracy or find a global optimal solution within a
certain probability. Applicability, stability, and convergence are the main criteria used to
evaluate the performance of heuristic algorithms.
Heuristic algorithms can be divided into traditional heuristics and metaheuristics.
Traditional heuristic algorithms include construction methods, local search algorithms,
relaxation methods, solution space reduction algorithms, etc. The metaheuristic algorithm
is an improvement of the heuristic algorithm, which is a combination of the stochastic
algorithm and the local search algorithm. The metaheuristic is an iterative generation
process, which achieves exploration and exploitation through the intelligent combination
of different concepts. Exploration aims to explore the promising areas of the search space
and avoid stagnating to the local optimum [3]. Meanwhile, exploitation aims to obtain a
solution from the search space with high accuracy [4].
In iterations, metaheuristics usually employ different learning strategies to acquire
and master information to efficiently discover near-optimal solutions. For example, particle
swarm optimization (PSO) [5] simulates the foraging behavior of birds, chemical reaction
optimization (CRO) [6] simulates the process of high-energy molecules gradually changing
to low-energy stable forms during the reaction, and ant colony optimization (ACO) [7]
simulates the mechanism of ants accomplishing information exchange through pheromones.
The application of the metaheuristic algorithm is relatively easy and does not require
complex mathematical operations and reasoning; only the objective function is essential as a
search indicator while it can be adapted to different optimization problems through flexible
parameter settings. Therefore, they have been widely used in theoretical optimization and
practical engineering fields [8–10]. With the emergence of complex optimization problems
in the industry, there is a growing need to address these problems through optimization
algorithms. Researchers have carried out a lot of work to improve the performance of
the algorithm.
The pity beetle algorithm [11] is a successful biology-based metaheuristic proposed
recently, which was inspired by the aggregation behavior of the six-toothed spruce bark bee-
tle in nature. Results from numerical experiments demonstrated the excellent performance
of PBA on the well-established test functions and CEC2014 benchmark functions.
The initialization phase of the PBA uses RST (random sampling technique) to ensure
the diversity of the initial population. RST divides the objective space into a number of
equal marginal hypervolumes. In the first stage of PBA, the initial population is randomly
assigned into the hypervolumes, so that the entire objective space is effectively covered
with equal probability. During the search phase, PBA has five different update modes
depending on the hypervolume selection pattern, including neighboring search hypervolume,
mid-scale search hypervolume, large-scale search hypervolume, global-scale search hypervolume,
and memory consideration search hypervolume. There are specific sequential conditions for
different search patterns. For example, mid- and large-scale search will only be performed
after a neighboring search has failed, and global search will only be performed after all
other methods have failed. In other words, PBA may execute two or three different search
strategies successively in the same iteration, which greatly increases the computational
effort of the algorithm. In addition, the memory consideration search mimics the role of
pheromones in the aggregation behavior of beetle populations. However, it is not clear from
the original paper how pheromones function to achieve spherical expansion of populations.
Mathematics 2022, 10, 2211 3 of 38

In order to overcome the shortcomings in PBA and to refine the relationship between
the beetle population model and the numerical implementations, this paper proposes an
improved PBA based on the pheromone dispersion model (PDM).
This technique (PDM) is inspired by the infestation and reproduction behavior of
beetles. The six-toothed spruce bark beetle in the Ipini tribe has the ability to infest the
whole forest [12]. The infestation always starts with a small brood; beetles attack susceptible
trees to search for a suitable host, feed on tree bark, and build nests in trees. When a suitable
host is found, some beetles will start breeding offspring and others will act as pioneers to
explore a wider area. In the process of exploration, pheromones are the medium through
which pioneers exchange information. During the reproduction period, female beetles
will release sex pheromones to attract surrounding males. They represent one of the most
well-studied insect groups in terms of chemical ecology [13,14]. Since the concentration
of sex pheromones decreases with the divergence distance, males close to female beetles
have a better chance of receiving the sex pheromone. Males that receive the sex pheromone
will compete with others for the right to mate and the winner in the competition will breed
with the female to produce offspring.
The main contributions of this paper are as follows:
1. Proposal of an improved pity beetle algorithm (PBA), which enhanced the searching
ability and particle diversity in high-dimension NP-hard optimization problems.
2. Proposal of a mathematical pheromone dispersion model (PDM) and refinement of
the PBA’s movement strategy with the help of this model.
3. Detailed and sufficient experimental results proved the superior performance of
the proposed PBA-PDM in both classical test functions and CEC2017 benchmark
functions. Furthermore, real-world engineering design problems verified the excellent
performance of PBA-PDM when dealing with constrained optimization problems.
The remainder of this paper is divided into six sections. Section 2 discusses the original
methods of PBA. Section 3 presents the mathematical model of PDM and embeds this model
in PBA. Section 4 represents the analysis of the experimental results, including classical
test function sets and CEC2017 benchmark functions. The results of PBA-PDM in solving
real-world constrained engineering design problems are presented in Section 5. The last
section summarizes our work and provides perspectives on future work.

2. Pity Beetle Algorithm (PBA)

The pity beetle algorithm mimics the infestation and reproduction of beetles in the
forest, which consists of three main stages: the search stage, the aggregation stage, and the
anti-aggregation stage. During the search stage, pioneers identify chemical traits emitted
by susceptible trees to locate suitable hosts for reproduction. When beetles successfully
infest the objective tree, they produce sex pheromones, which attract nearby males and
females to congregate. The colonies of beetles that aggregate in the host will proliferate
and when the population exceeds a specific threshold, the forest’s defense mechanisms will
be unable to contain the mass infestation (aggregation stage). However, there is a limit to
the number of beetles a host can support, and an overcrowded habitat means a smaller
share of food and a higher probability of disease. As a result, when population densities
become too high, beetles release an anti-aggregation pheromone that directs beetles away
from their current hosts to infest other trees (anti-aggregation stage).
Initialization: PBA generalizes the initial population using the random sampling
technique (RST). RST decomposes the objective space into multiple hypervolumes with
equal marginal boundaries, and then generates an initial coordinate at random on the
boundary of each hypervolume.
The implementation steps used to achieve initialization are as follows:
Mathematics 2022, 10, x FOR PEER REVIEW 4 of 35

Mathematics 2022, 10, 2211 4 of 38

 ui , k  li , k
 ,i  1
Step 1. Hypervolumes are divided t  into N popNpop parts using Equation (2):
Leni , k   (2)
 t ui , k  li , k
 u 
i,k Len
− l i,k 
i,i1, k= 1 , i  2
t N N pop
∆Leni,k =  pop t ui,k −li,k (2)
∆Len
i −1,k + Npop , i ≥ 2
where i = 1, 2, ..., Npop; t = 1, 2, ..., itermax; k = 1, 2, ..., Dim. Δ , refers to the length of the
i-th hypervolume in the k-th dimension of the t-th iteration. ui,k andt li,k are the upper bound
where i = 1, 2, ..., Npop ; t = 1, 2, ..., itermax; k = 1, 2,..., Dim. ∆Leni,k refers to the length of
and the lower bound of the k-th dimension, respectively.
the i-th hypervolume in the k-th dimension of the t-th iteration. ui,k and li,k are the upper
bound Step 2.the
and Δ bound
Thelower , obtained
of the in the
k-th previous respectively.
dimension, step forms a matrix of Npop rows and
Dim columns, i.e.,∆Len
each tdimension of the objective space is partitioned into Npop equidistant
Step 2. The i,k obtained in the previous step forms a matrix of Npop rows and
boundaries:
Dim columns, i.e., each dimension of the objective space is partitioned into N equidis- pop
tant boundaries:  Len  Len t t

1,1 1, Dim
 t   
MLen ∆Len   1,1 t · · · ∆Len1,Dim t
 (3)
 t t 
MLent =   LenNpop.,1. . Len Npop , Dim  i  k 
 
  (3)
∆LentN pop,1 ∆LentN pop,Dim
According to the study by Kallioras et al. [11], six initial broods i ×k (composed of Npop
particles each) are formed by randomly pairing the elements of each column in the matrix
According to the study by Kallioras et al. [11], six initial broods (composed of N
MLen. The RST-based population generation strategy is noted as: RST(li,k, ui,k, Dim, Npoppop ).
particles each) are formed by randomly pairing the elements of each column in the matrix
In some exceptional situations the boundaries of randomly paired hypervolume and the
MLen. The RST-based population generation strategy is noted as: RST(li,k , ui,k , Dim, Npop ).
global bounds do not match. In this case, the sampling interval of the current hypervol-
In some exceptional situations the boundaries of randomly paired hypervolume and the
ume should be adjusted using the global bounds as a criterion, as shown in Figure 1:
global bounds do not match. In this case, the sampling interval of the current hypervolume
should be adjusted using the global  Lbi , k bounds
 ui , k  las a criterion, as shown in Figure 1:
i , k  , if li , k  Lbi , k , ui , k  Ubi , k
 
+ i(, kui,k−
Leni , kLbi,kUb )l,i , k  ,i fif lui,ki , k <
ui , kli,k  Ub i , k ,, lu
Lbi,k k <
i ,i,k LbUb
i , k i,k (4)
∆Leni,k = Ubi,k  − (u − l ), i f u > Ub , l > Lb (4)
 Ubi , k i,kLbi , k i,k , if uii,k , k  Ubi , ki,k , li , k i,k Lbi , k i,k
Ubi,k − Lbi,k , i f ui,k > Ubi,k , li,k < Lbi,k

Figure 1.
Figure RST correction
1. RST correction when
when the
the bounds
bounds of the hypervolume are outside of the global bounds.

Searching strategies:
Searching PBA has
strategies: PBA hasdevised
devisedmainly
mainly two
twodifferent
different search
search modes,
modes, including
including
→
deterministic area search and memory consideration search. The decision variable ⃗x in
deterministic area search and memory consideration search. The decision variable in
Equation (1) is represented as the beetles in the population. Additionally,
Equation (1) is represented as the beetles in the population. Additionally, the optimum the optimum
beetle (the
beetle (thedecision
decisionvariable
variablewithwith thethe minimum
minimum fitness
fitness value)
value) in current
in the the current
brood brood is
is rep-
represents
resents as the
as the ‘pioneer’.
‘pioneer’.
The deterministic
The deterministic area areasearch
searchmethodmethodgenerates
generates a random
a random population
population around
around the the
pi-
pioneer using RST. Depending on the size of the random population,
oneer using RST. Depending on the size of the random population, the deterministic areathe deterministic area
search can
search can be
be divided
divided into
into four
four modes,
modes, including neighboring search
including neighboring search mode,
mode, mid-scale
mid-scale search
search
mode, large-scale search mode, and global-scale search mode. The size of the random
mode, large-scale search mode, and global-scale search mode. The size of the random population population
also defines
also defines the parameter ffpat
the parameter pat, , which
which determines
determinesthe the range
range ofof the
the new
new beetles
beetles generated
generated
under different search modes. Different search modes represent different
under different search modes. Different search modes represent different search strategies. search strategies.
However, the mathematical descriptions are the same for all four search modes, as shown
in Equation (5). For each search mode, only the parameter fpat is different. For example,
 However, the mathematical descriptions are the same for all four search modes, as shown
in Equation (5). For each search mode, only the parameter fpat is different. For example, in
neighboring search mode, fpat takes a value in the interval of 0.01 to 0.2. Among them, the
Mathematics 2022, 10, 2211 initial population is generated using the neighboring search mode. The other modes5 of 38 are
selected when the neighboring search mode fails to improve the performance of the global
optimum:
in neighboring search mode, fpat takes a value in the interval of 0.01 to 0.2. Among them,
the initialpopulation
lt , ut    pioneer 
t  1  f pat
is generated 
, pioneer
using   , pioneert   x1 , x2 ,..., xDim 
t  1  f pat search
the neighboring  mode. The other modes
are selected when the neighboring search mode fails to improve the performance of the

xnew  RST lt , ut , Dim, N pop
global optimum:

where



[lt , ut ] ∈ pioneert · 1 − f pat
, pioneert · 1 + f pat , pioneert = [ x1 , x2 , · · · , x Dim ]
Neighboring search mode: f pat   0.01,0.2 (5)
xnew = RST lt , ut , Dim, Npop
whereMid -scale search mode: f pat   0.1,1.0
Neighboring search mode : f pat ∈ [0.01, 0.2] (5)
Large-scale search mode: f pat 1.0
Mid-scale search mode : f pat ∈ [0.1, 1,100
] 
pat   0.05,0.25
Large-scale
Globalsearch
-scale mode f pat ∈
search: mode : f[1, 100]
Global-scale search mode : f pat ∈ [0.05, 0.25]
where pioneert refers to the optimum decision variable in the current iteration and xnew is
where pioneer refers to the optimum decision variable in the current iteration and xnew is
the new beetle tgenerated in the deterministic area.
the new beetle generated in the deterministic area.
AnAn illustration of the deterministic search is shown in Figure 2, where the brood size
illustration of the deterministic search is shown in Figure 2, where the brood size
(N(N
pop) is equal to 5 and the problem dimension (Dim) is equal to 2. The red circle is the
pop ) is equal to 5 and the problem dimension (Dim) is equal to 2. The red circle is the
current
currentpioneer andthe
pioneer and the green
green circle
circle represents
represents the newthe newrandomly
beetles beetles generated
randomlyaround
generated
around the pioneer.
the pioneer.

Figure
Figure 2. 2. Illustrationof
Illustration offour
four deterministic
deterministic area
areasearching methods.
searching methods.

The memory consideration searching method stores the optimal solution of the popula-
The memory consideration searching method stores the optimal solution of the pop-
tion at each iteration into a matrix, named memory (MEM), as shown in Equation (6). Unlike
ulation at each
the four iterationarea
deterministic intosearch
a matrix,
modes named memory
mentioned (MEM),memory
previously, as shown in Equation (6).
consideration
Unlike the four deterministic area search modes mentioned previously,
search selects a random location (position vector) from the matrix MEM and performs RST memory consid-
eration
using search selects
this vector. a random
In this locationa(position
case, to perform vector)
narrow local from
search, the the matrix
value of fpat MEM
shouldand
be per-
forms
withinRSTtheusing
rangethis vector.
of [0.005, In this
0.05]. If a case,
beetleto
inperform a narrow
the generated randomlocal search, the
population value
obtains a of fpat
smaller fitness value than the selected one, then the original one will be replaced:
should be within the range of [0.005, 0.05]. If a beetle in the generated random population
obtains a smaller fitness value than
 the selected one, then the original one will be replaced:
x1,1 x1,2 ··· x1,N pop

 x2,1 x x 2,2 x1,2 · · · x2,N pop
x1, Npop
 
MEM =   . 1,1 (6)
 
. .
. . . .
. 
 . x2,1 . x2,2 .  . x2, Npop

 

MEM x x · · · x (6)
Dim,1 Dim,2 Dim,N pop 
 
 xDim ,1 of PBA
Update strategy: The update strategy  xfrom
xDim ,2is different the traditional biology-
Dim , Npop
based algorithm. The beetles generated in the current iteration are not kept for performing
theUpdate strategy: The
search procedures update
in the strategy of
next iteration. PBA
Only theisoptimum
differentbeetles
from the traditional
(pioneers) biology-
in each
based algorithm.
brood and MEMThe beetles
remain, andgenerated
the rest arein
allthe current
removed iteration
from are not kept
the population for performing
(extinction):
the search procedures in the next
 iteration. Only the optimum beetles (pioneers) in each
brood and MEM remain, and pioneer
the rest are f ( pioneer
t , i f all removedt ) <from
f ( xnew ) population (extinction):
the
x t +1 = (7)
xnew , otherwise
Mathematics 2022, 10, 2211 6 of 38

where xt+1 is the optimal solution after one iteration, xnew is the new decision variable
generated in the brood, and f (pioneert ) and f (xnew ) are the fitness values corresponding to
pioneert and xnew , respectively.

3. Pity Beetle Algorithm Based on Pheromone Dispersion Model (PBA-PDM)

The PBA’s search strategies in the deterministic area rely entirely on the position of the
initial population and the random beetles generated by the RST. Moreover, the evaluation in
each iteration requires a comparison of the fitness values of all new beetles and the pioneer.
When the population size increases, the computational effort also increases significantly,
which is undesirable for large-scale optimization problems. In addition, the information of
the search direction is not used in PBA to guide the growth of the population. The purely
random-based RST strategy is effective in ensuring particle diversity but at the expense
of poor convergence. When the iteration has been completed, only the optimums in each
brood are saved in the MEM. The memory consideration search based on MEM controls
the search area (fpat ) to be within a narrow range to complete the operation, which is not
significantly different from the deterministic area search strategy in essence.
The disadvantages of the deterministic area search and memory consideration search
are summarized as follows:
1. The efficiency of optimization is dependent on the location of initial beetles in each
brood, i.e., the stability of the algorithm is uncertain and undesirable.
2. There is no clear moving direction to guide the exploration in the promising area, and
the generation of new individuals relies entirely on RST, i.e., the convergence of the
algorithm depends on the size of the nascent population in the brood.
3. Only the optimum beetle (pioneer) in each brood is retained after iteration, resulting
in some valid information from other individuals being ignored, i.e., the algorithm
does not make full use of the convergence information in the population.
In order to improve the performance of PBA and address the above shortcomings, this
study radically improves the existing search strategies and designs a pheromone dispersion
model to guide the search direction of the population.

3.1. Improvement of Deterministic Area Search

After the initial population is generated by RST, the position of each beetle is repre-
sented in a matrix as:  
beetle1
 beetle2 
population =  (8)
 
.. 
 . 
beetle N pop
beetlei = [ x1 , x2 , · · · , x Dim ], i = 1, 2, · · · , N pop (9)
Each individual in the initial population has a randomly obtained coordinate that
matches its fitness value. Substituting the coordinates into the objective function results in
Npop corresponding initial fitness values, which is represented in a matrix as:
 
f (beetle1 )
 f (beetle2 ) 
f itness value =  (10)
 
.. 
 .


f beetle N pop

The improved PBA sorts all individuals in the population in order of their fitness
value from the minimum to the maximum. The top 10 percent of the beetles in the ranking
sequence are defined as the current pioneers in each brood while the last 30 percent of
the beetles in the ranking sequence are extinct from broods; others are defined as workers
in the broods (the specific ratio of workers to the pioneer in each brood is discussed in
 The improved PBA sorts all individuals in the population in order of their fitness
value from the minimum to the maximum. The top 10 percent of the beetles in the ranking
sequence are defined as the current pioneers in each brood while the last 30 percent of the
beetles in the ranking sequence are extinct from broods; others are defined as workers in
Mathematics 2022, 10, 2211 7 of 38
the broods (the specific ratio of workers to the pioneer in each brood is discussed in Sec-
tion 4). It is worth noting that PBA-PDM names the beetles “workers” and “pioneers” only
to distinguish the roles of different individuals in the same brood, and they are both es-
Section 4). It is worth noting that PBA-PDM names the beetles “workers” and “pioneers”
sentially decision the
only to distinguish variables
roles ofof the optimization
different individualsproblem.
in the same brood, and they are both
In the grouping process, the top n workers
essentially decision variables of the optimization are assigned to the first-ranked pioneer to
problem.
formInthe thefirst brood.process,
grouping Similarly, thennth
the top to 2nth
workers areworkers
assigned are assigned
to the to the
first-ranked second-ranked
pioneer to
form the first brood. Similarly, the nth to 2nth workers are assigned to the second-ranked
pioneer and so on. This process is repeated until all the beetles have their own brood.
pioneer3and
Figure so on. This
illustrates process and
the sorting is repeated until all thewhere
group processes, beetlesnhave
equals their ownnumber
2. The brood. of the
Figurei”3 illustrates
“beetle subscript the sorting the
indicates andranking
group processes, where n equalsfitness
of the corresponding 2. The number
value ofofthe
thedecision
“beetlei ” subscript indicates the ranking of the corresponding fitness value of the decision
variable in the total population. After partition, the pioneer in each brood is considered as
variable in the total population. After partition, the pioneer in each brood is considered as
aa leader to guide the search direction, and workers are considered as searching agents.
leader to guide the search direction, and workers are considered as searching agents.

3.Illustration
Figure 3.
Figure Illustrationofof
thethe
sorting andand
sorting grouping processes.
grouping processes.
The deterministic area search is mainly based on the difference vector constructed by
The deterministic area search is mainly based on the difference vector constructed by
the coordinate of the pioneer and current individual. The beetles in each brood randomly
the coordinate of thearea
explore every possible pioneer
in theand current
objective individual.
space The beetles
to find a suitable in each
host, which brood randomly
is expressed
explore
as follows:every possible area in the objective space to find a suitable host, which is ex-
pressed as  follows: 
t +1 t t t
workeri,k = workeri,k ± pioneeri,k − workeri,k × ((Ubk − Lbk ) + Lbk ) × rand × pt × step (11)
workerit, k1  workerit, k   pioneerit, k  workerit, k    Ubk  Lbk   Lbk   rand  pt  step (11)
t t

pioneeri , i f f ( pioneeri ) < f ( RST )
pioneerit+1 = t) (12)
RST (Ub,  Lb, Dim,
pioneer t
, N pop ) , i f f if( RST
f ( ) < f (tpioneer
pioneer )  f ( RST
i )
 i i
pioneer t 1 
t refers to thei k-th  t (12)
where workeri,k dimension
RST (Ub, Lbof i-th, Nworker
, Dim pop ), if fin the)t-th
( RST pioneerit ) pioneeri,k
 f (iteration,
refers to the k-th dimension of the i-th pioneer in the t-th iteration, pt refers to the beetle’s
where refersdecreases
ability to explore,, which to the k-th dimension
with the number of ofi-th worker rand
iterations, in the t-th iteration,
generates a random ,
number in the interval [0, 1], and step adjusts the step size through
refers to the k-th dimension of the i-th pioneer in the t-th iteration, pt refers to the beetle’san adaptive mechanism:
ability to explore, which decreases with  the number of iterations, rand generates a random
t
number in the interval [0, 1], p = exp
t and step adjusts × cos(2π the·rand
step ) size through an adaptive (13) mecha-
T
nism: !
workerit
step = exp  t  (14)
  cosit  2  rand 
pt  exp  pioneer (13)
T
 
where T refers to the maximum iteration and pioneerbest refers to the global optimum
obtained in the current iteration.  workerit 
The steps for the deterministic areastep  expare
search  shown t in Figure 4. After the initial (14)
population has been sorted and grouped, workers  pioneer
in each i 
brood perform a directional
vector-based search according to Equation (11), and when all workers have completed their
position updates, pioneers perform an RST-based search according to Equation (12).
 where T refers to the maximum iteration and pioneerbest refers to the global optimum ob-
tained in the current iteration.
The steps for the deterministic area search are shown in Figure 4. After the initial
population has been sorted and grouped, workers in each brood perform a directional
Mathematics 2022, 10, 2211 8 of 38
vector-based search according to Equation (11), and when all workers have completed
their position updates, pioneers perform an RST-based search according to Equation (12).

Figure 4. Illustration
Figure 4. Illustrationofofthe
theimproved deterministicarea
improved deterministic area search.
search.

3.2. Improvement of Memory Consideration Search

3.2. Improvement of Memory Consideration Search
As mentioned in Section 2, the optimal solution of the population at each iteration is
As mentioned
stored in MEM. This in is
Section
mainly2,because
the optimal solutionaoflarge
PBA generates the number
population at each
of new iteration is
individuals
stored in MEM.
during the search Thisphase.
is mainly because
In order PBA generates
to maintain the populationa large number
size, of new
the current individuals
population
during
has tothe
be search
clearedphase. In order
at the end of eachtoiteration.
maintainTherefore,
the population size, thethe
MEM becomes current population
only efficient
has to be cleared at the end of each iteration. Therefore, MEM becomes the only efficient
way to preserve the convergence information and to guide the search direction.
way to In this study,
preserve the the memory consideration
convergence informationsearch
and toisguide
improved throughdirection.
the search a pheromone-
based search strategy, termed the pheromone consideration search.
In this study, the memory consideration search is improved through a pheromone- The mechanism for
storing historical information via MEM is retained; the information in MEM is not used
based search strategy, termed the pheromone consideration search. The mechanism for
as a dominant factor in guiding the search for populations but only as a reference for
storing historical information via MEM is retained; the information in MEM is not used
updating movement.
as a dominant
When the factor
pioneerin finds
guiding the search
a suitable for place,
breeding populations
it releasesbut only as a in
pheromones reference for up-
situ to lure
dating movement.
nearby beetles. During this phase, the concentration of pheromones gradually increases
When the pioneer
and diffuses. In order finds a suitable
to simulate breedingstate
the diffusion place, it releases
of the pheromone pheromones in situ to
in nature more
lureaccurately,
nearby beetles.
a pheromone During this phase,
dispersion the proposed
model was concentration of pheromones
by combining the Gaussiangradually
plume in-
modeland
creases [15].diffuses. In order to simulate the diffusion state of the pheromone in nature
The Gaussian
more accurately, plume model
a pheromone simplifies model
dispersion the nonlinear model of by
was proposed thecombining
airborne contami-
the Gauss-
nants affected by turbulence and replaces the complex solution in the nonlinear equation
ian plume model [15].
with an approximate analytical solution [16]. The contaminant concentration in this model
The Gaussian plume model simplifies the nonlinear model of the airborne contami-
can be expressed as:
nants affected by turbulence and replaces the complex solution in the nonlinear equation
with an approximate analytical " [16]. The contaminant
solution !
concentration
!#
in this model
Q y2 ( z − H )2 ( z + H )2
c ( x, y, z
can be expressed as: 4πux ) = exp − exp − + exp − (15)
4x 4x 4x
2 2
where (x, y, z) is the coordination  y2  
Q of measurement  points,
 z  H Q refers to the
 H    rate, H is
z emission
c  x, y, z   exp    exp    
 exp   (15)
the height of the emission point, 4 ux and u 4isx the 
  ambient 4 wind
x  speed. 4 x 
   
The eddy diffusion coefficient is a function of the downwind distance x and varies
with
where (x,they,weather
z) is theconditions
coordination and ofrelease time, so itpoints,
measurement is difficult to determine
Q refers in practice.
to the emission rate, H
Therefore, it is common practice to replace x with a
is the height of the emission point, and u is the ambient wind speed. new argument r:
The eddy diffusion coefficient is a 1function
Z x of the downwind distance x and varies
r
with the weather conditions and release= K ( ξ ) dξ
so it is difficult to determine in (16)
u time,
0 practice.
Therefore, it is common practice to replace x with a new argument r:
In order to simplify Equation (15), in the pheromone dispersion model, the emission
1 x that the diffusion in all directions (dimen-
height H is set to 0. At the same time, to rensure
  K  d  (16)
sions) is fair, the downwind direction is set as the
u 0
direction of the current beetle towards the
 Mathematics 2022, 10, x FOR PEER REVIEW 9 of 35

Mathematics 2022, 10, 2211 In order to simplify Equation (15), in the pheromone dispersion model, the emission 9 of 38
height H is set to 0. At the same time, to ensure that the diffusion in all directions (dimen-
sions) is fair, the downwind direction is set as the direction of the current beetle towards
the global optimum point → ⃗, and no longer replaces downwind direction with r. Equation
global optimum point d , and no longer replaces downwind direction with r. Equation (15)
(15) can be rewritten as:
can be rewritten as:
Q Dim x 2  
c  x1 , x2 ,, xDim   Q  Dim exp   i xi 2 (17)
c( x1 , x2 , · · · , x Dim ) = 4→ ud∏ i 1
exp − 4  (17)
i =1 4
4πu d
Figure 5 depicts a schematic of the Gaussian plume dispersion model with a single
Figure 5release
pheromone depictssource.
a schematic
In order of the Gaussianvisualization,
to facilitate plume dispersion model with
the pheromone a single
concentra-
pheromone release source. In order to facilitate visualization, the
tion dispersion at the downwind locations is simplified to two directions (vertical in red, pheromone concentra-
tion dispersion
horizontal at theFigure
in blue). downwind locations
6 illustrates the is simplifieddispersion
pheromone to two directions (vertical
model in in red,
a 1-D and 2-D
horizontal in blue). Figure 6 illustrates the pheromone dispersion
objective space. The pheromone concentration is represented by the sparseness of the par- model in a 1-D and
2-D objective
ticles, wherespace. The pheromone
the denser the particles concentration
(near the originis represented by the sparseness
of the coordinates), the higherof the
the
particles, where the denser the particles (near the origin of the coordinates),
concentration. The overall concentration distribution satisfies a standard normal distribu- the higher the
concentration.
tion. It is notedThethatoverall concentration
the pheromone distribution
dispersion satisfies a decreases
range gradually standard normal
with thedistribu-
number
tion. It is noted that the pheromone dispersion range gradually
of iterations. This means that at the beginning of the iteration, a large pheromone decreases with the number
spread
of iterations. This means that at the beginning of the iteration, a large pheromone spread
attracts individuals located in multiple areas and increases the efficiency of exploration.
attracts individuals located in multiple areas and increases the efficiency of exploration. As
As the iteration progresses to the later stages, the reduction in pheromone spread allows
the iteration progresses to the later stages, the reduction in pheromone spread allows the
the population to focus more on improving the accuracy of the global optimum solution.
population to focus more on improving the accuracy of the global optimum solution.

Figure5.5.AAcontaminant
Figure contaminantplume
plumeemitted
emittedfrom
fromaacontinuous
continuouspoint
pointsource,
source,with
withthe
thewind
winddirection
direction
aligned with the x-axis.
aligned with the x-axis.

When the pheromone concentration c received by the workers is sufficient to be

recognized by the olfactory receptor cells (the specific concentration of pheromone is
discussed in Section 4), the beetle starts the exploitation stage and further approaches the
global optimum. Workers who fail to identify pheromones continue the exploratory phase
to expand the search space further.
Exploitation: The workers move toward the pioneer after sensing the pheromone, as
shown in Equation (18):

workerit+1 = workerit + workerbest

t
+ workerit × rand + pioneerit + workerit × rand × c



(18)

Meanwhile, the female (pioneer) releasing pheromones moves towards a more suitable
environment (global optimal point) at a slow speed, as shown in Equation (19):

pioneerit+1 = pioneerit + MEMrand × pt + pioneerbest − pioneerit × rand × step

(19)

where MEMrand represents the random beetle selected in matrix MEM, c represents the
concentration of pheromone, workerbest refers to the optimum worker in the brood in which
it is located, and pioneerbest refers to the optimum beetle in the whole population.
Mathematics 2022,
Mathematics 10,10,
2022, x FOR
2211 PEER REVIEW 10 10 of 35
of 38

Figure
Figure6.6.Dispersion
Dispersionofofpheromones
pheromonesin
in1−D
1−Dand
and2−D
2−Ddimensionality
dimensionalityspace
spacewhen
whenthe
thenumber
numberofofiter-
ations is 10,is50,
iterations 10, and 90. 90.
50, and

When the pheromone

Reproduction: The nascentconcentration
population c received
in PBA is bygenerated
the workers is sufficient
by RST, i.e., the to be rec-
repro-
duction by
ognized process does not receptor
the olfactory consider the cellsvalid(theinformation generated during
specific concentration the iteration.
of pheromone is dis-
cussed in Section 4), the beetle starts the exploitation stage and further approachesofthe
Therefore, such a stochastic approach can seriously weaken the convergence performance
the algorithm
global optimum. in theWorkerslate iterative
who fail stage. In orderpheromones
to identify to more realistically
continue mimic the behavioral
the exploratory phase
patterns of the beetles
to expand the search space further. during the mating period, this paper proposes a novel reproduction
strategy that is more
Exploitation: The fine-grained
workers move and informed
toward the bypioneer
historical convergence
after sensing the information.
pheromone, as
It is common to observe such a behavior in natural populations of beetles: When male
shown in Equation (18):
beetles are exposed to the contact pheromone on the female’s body, they will try to ride
on the back of the
worker i
t 1 female longhorn
 workerit   workerbeetle and textend their aedeagus t andt try to pull her(18)
best  workeri   rand   pioneeri  workeri   rand  c
t

genitalia. There are two modes of interaction at this point: the recognition interaction
between males and
Meanwhile, thethe struggle
female between
(pioneer) femalespheromones
releasing for mating rights.moves It towards
follows that males
a more suit-
tendenvironment
able to move closer to females
(global optimal whopoint)
are located
at a slow in suitable
speed, as positions
shownduring the breeding
in Equation (19):
stage. In keeping with the above relationship between pioneers and workers, the definitions
of females andpioneer malesi are
t 1
pioneerit by
 replaced MEM   pioneer
rand  pt and
pioneers pioneerit   rand  step
best  respectively.
workers, (19)
Interaction between males (workers): Due to the insufficient differentiation of cu-
where
ticularMEMwax rand representsmales
components, the random beetle selected
have a probability in matrix identifying
of mistakenly MEM, c represents same-sexthe
concentration
individuals asof pheromone,
opposite sexes and worker best refers
trying to mate to the
withoptimum
them. If two worker
maleinbeetles
the brood in which
are paired,
itthey
is located, and pioneer
are regarded as the refersindividual
best same to the optimum to reduce beetle
theinpopulation
the whole size. population.
The location
Reproduction:
information The nascent
of the individual population
is expressed as: in PBA is generated by RST, i.e., the repro-
duction process does not consider the valid information  generated during  the iteration.
t t t t
workersuch
Therefore, new =a worker
stochastic p × p t + worker
approach can
q × ( 1 −
seriouslyp t ) ± worker
weaken −
the
p worker
convergence
q × rand
performance
(20)
of the algorithm in the late iterative stage. In order to more realistically mimic the behav-
Battle between
ioral patterns of the beetles femalesduring (pioneers):
the matingFemalesperiod,in a coupling
this paperpair slap other females
proposes a novelwhorepro-
are close to their territory with their tentacles.
duction strategy that is more fine-grained and informed by historical convergence If the female wins the battle, the intruder
infor-
will
mation. flee; if the battle fails, the intruder will take over. At this time, the second female
immediately
It is common moves into the such
to observe position of the original
a behavior in natural ones, and then the
populations male grabs
of beetles: When andmale
mounts her. The condition for femalep to win the battle is f (femalep ) > f (femaleq ).
beetles are exposed to the contact pheromone on the female’s body, they will try to ride
When a female (pioneer) successfully defends her territorial sovereignty, the update
on the back of the female longhorn beetle and extend their aedeagus and try to pull her
strategy is shown in Equation (21):
genitalia. There are two modes of interaction at this point: the recognition interaction be-
tween malespioneer and the t+1 struggle between females for mating rights. It follows that(21) males
 
p = pioneer tp ± pioneer t t
p − pioneerq × rand × pt × step
tend to move closer to females who are located in suitable positions during the breeding
stage. In keeping with the above relationship between pioneers and workers, the defini-
tions of females and males are replaced by pioneers and workers, respectively.
Mathematics 2022, 10, 2211 11 of 38

Otherwise:
 
pioneerqt+1 = pioneer tp ± pioneerqt − pioneer tp × rand × pt × step (22)

When
Mathematics 2022, 10, x FOR PEER REVIEW a male beetle mates with a female successfully, the female will cut a slit in 35
12 of the
oval position with her mandible and attempt to lay eggs while the male is still attached to the
female’s back [17]. Specifically, after mating, the male beetle (worker) and the corresponding
female beetle (pioneer) are regarded as the same individual (assimilation). Compared to the
Check if the individual is beyond the objective space
extinction operation in PBA where most randomly generated individuals are eliminated
t=t+1
after each iteration, the assimilation strategy in PBA-PDM improves the exchange of
Update xbest, fbest information between beetles and effectively enhances the convergence efficiency of the
if meet stop conditionalgorithm. Reproduction means the valid information of the workers is inherited by
stop_condition = 1the newly generated individuals. The assimilation strategy can effectively control the
end if population size and avoid the waste of computing resources. The pseudocode of PBA-PDM
end while is provided in Algorithm 1. The flowchart of PBA-PDM is presented in Figure 7.

Figure 7. Flowchart of PBA-PDM.

Figure 7. Flowchart of PBA-PDM.
3.3. Complexity Analysis
In general, the computational complexity of metaheuristics is an important factor to
be considered in practice. For simplicity, it is assumed that the problem to be optimized
has d dimension, the population size is n, and the maximum number of iterations is T.
PBA-PDM evaluates n individuals and divides the population into six departments in the
Mathematics 2022, 10, 2211 12 of 38

Algorithm 1: Pity Beetle Algorithm Based on Pheromone Dispersion Model

Input: initial parameters n, T, p, objective function f (x), lower bound lb , upper bound ub
Output: position of global optimum beetle xbest and the corresponding fitness value fbest .
Initialize populations according to the RST
while (stop_condition==0) do
Evaluate fitness values for all individuals
Sort all individuals and select top 10% beetles as pioneers
Construct sub-populations (broods) according to Equation (10)
Perform exploration operations, update the positions of individuals using Equation (11) and
Equation (12)
Store the optimal coordinate in matrix MEM
Calculate the pheromone concentration according to the pheromone dispersion model
Males (workers) who identified pheromone update their location using Equation (18)
Females (pioneers) update their location using Equation (19)
New populations of workers were generated by Equation (20)
Females (pioneers) will defend their territory during the mating season
if the defense is successful
Update the new position according to Equation (21)
else
The intruder replaces the current position, and update the location according to
Equation (22)
end if
Check if the individual is beyond the objective space
t=t+1
Update xbest , fbest
if meet stop condition
stop_condition = 1
end if
end while

3.3. Complexity Analysis

In general, the computational complexity of metaheuristics is an important factor to be
considered in practice. For simplicity, it is assumed that the problem to be optimized has d
dimension, the population size is n, and the maximum number of iterations is T. PBA-PDM
evaluates n individuals and divides the population into six departments in the initial stage
while each individual is evaluated once in each iteration. The total time complexity of
PBA-PDM can be calculated as:

O( PBA-PDM ) = O(1 + nd + Tnd) (23)

where T  n, p  n. When n or T goes to infinity, O( PBA-PDM ) ∼ = O( Tnd).

It can be seen that the time complexity of PBA-PDM is a first-order polynomial, which
is mainly influenced by three factors: problem dimension (d), number of iterations (T), and
population size (n).
PBA-PDM creates a matrix of historical information throughout the iteration to record
the best solutions generated in each iteration, which implies a memory requirement of about
(Td + Tnd). The space complexity is O((1 + n)Td), which can be expressed as Equation (24)
when T or n becomes infinite:

O( PBA-PDM ) = O( Tnd) (24)

4. Numerical Experiment
4.1. Experiment Setting
The performance of the algorithm was analyzed in two parts: qualitative and
quantitative. The qualitative part analyzed the superiority of the algorithm through the
convergence curve, and the quantitative part analyzed the performance of PBA-PDM
Mathematics 2022, 10, 2211 13 of 38

through the standard deviation (Std) and average of the results (Avg). All algorithms
were run by MATLAB R2020a on a computer with a Windows 10 64-bit professional
system and 16 GB RAM. The swarm size of every algorithm was 30 and maximum NOFE
was 10,000. All results were recorded and compared based on the average performance
of the optimizers over 30 independent runs.
Note that the pity beetle algorithm is a novel metaheuristic algorithm proposed
in 2018. There is only one improved version of PBA, called the one-way pioneer guide
pity beetle algorithm (OPGPBA) [18], proposed in 2020. In order to fully verify the
performance of the algorithm proposed in this article, the one-way pioneer guide pity
beetle algorithm (OPGPBA), the enhanced comprehensive learning particle swarm opti-
mization (ECLPSO) [19], the chaos cultural particle swarm optimization (CCPSO) [20],
the improved real-code genetic algorithm (IRGA) [21], hybrid optimization technique
based on DE and NMR algorithms (HDN) [22], the weighted mean of vectors (INFO) [23],
the fuzzy adaptive and enhanced imperialist competitive algorithm (FAEICA) [24], the
winner of CEC2014, L-SHADE [25], and the winner of CEC2017, EBOwithCMAR [26]
were used for comparison to fully embody the performance of PBA-PDM in handling
various types of optimization problems. All parameter settings were selected from the
original works.

4.2. Parameter Sensitivity Analysis

In the sorting operation in Section 3, the PBA-PDM divides the beetle population
into multiple broods according to the fitness value. Each brood has one and only one
pioneer; however, the number of workers in each brood is uncertain. Determining a
suitable number of workers is very important to improve the efficiency of the algorithm.
A large number of workers can support a broader exploration of the search space, but
too large a size of the populations may cause a heavy burden regarding calculation.
A small number of workers gives faster convergence, but too small a population size
may make the algorithm unable to find the optimal solution in the limited number of
iterations. In addition, the pheromone concentration bound is critical to the efficiency of
the exploration phase. Before applying the algorithm to solve optimization problems, it
is necessary to analyze the parameter sets systematically.
Three classical test functions, named the Perm function (100-D), Sphere function (30-
D), and Dixon-price function (10-D), were selected in this experiment and each test was
simulated over 30 independent runs. The variation range of the number of workers was
set within the range of 0.1n to 0.9n and the pheromone concentration bound was in the
interval of [0.1, 0.9], and the indicator used to measure the effectiveness of the parameter
was the average number of function evaluation (NOFE).
The experiment results are shown in Tables 1–3. For an intuitive distinction, three
results with the least number of NOFE in each rows of the table are shown in bold. The stop
condition of each test is that the error between the optimal value obtained by the algorithm
and the theoretical optimum is within 1 × 10−8 . If the calculation time exceeds 10 seconds
or NOFE are more than 20,000, it is considered that the current parameter combination is
not conducive to the operation of the algorithm. Meanwhile, the results marked in bold font
are the parameter combination that obtained the lowest four NOFE. It can be concluded
through a comprehensive analysis of the three experiments that the reasonable number of
workers is in the range of 0.5n and 0.7n while the interval of the pheromone concentration
bound is [0.5, 0.8].
Mathematics 2022, 10, 2211 14 of 38

Table 1. Average NOFE with different parameter settings in the Perm function (100 D).

Ratio of Workers to Pioneer

pheromone concentration bound

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

0.1 20,080 20,080 20,080 820 720 860 920 1040 1380
0.2 20,080 20,080 20,080 20,080 780 840 920 1220 1340
0.3 20,080 20,080 20,080 20,080 940 880 700 1000 1080
0.4 20,080 20,080 20,080 820 780 760 660 780 920
0.5 20,080 20,080 20,080 20,080 740 740 800 1080 1240
0.6 20,080 20,080 20,080 20,080 660 720 900 900 920
0.7 20,080 20,080 20,080 600 680 640 780 800 1040
0.8 20,080 20,080 20,080 780 800 820 760 800 880
0.9 20,080 20,080 20,080 720 600 880 880 800 960

Table 2. Average NOFE with different parameter settings in the sphere function (30 D).

Ratio of Workers to Pioneer

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
pheromone concentration bound

0.1 20,080 20,080 2040 3040 4400 5740 7200 8020 9560
0.2 20,080 20,080 20,080 3020 4100 4980 6480 7180 7740
0.3 20,080 20,080 2700 3500 3940 4300 5420 6360 6880
0.4 20,080 20,080 20,080 2800 3800 4180 5480 5660 6000
0.5 20,080 20,080 2300 2740 3520 4020 4800 5120 5460
0.6 20,080 20,080 2740 2960 3340 4160 4240 4660 5020
0.7 20,080 20,080 2800 2940 3200 3400 3460 4060 4580
0.8 20,080 20,080 20,080 2700 3240 3400 3580 4380 4460
0.9 20,080 20,080 20,080 2520 3200 3260 3360 3560 4220

Table 3. Average NOFE with different parameter settings in the Dixon-Price function (10 D).

Ratio of Workers to Pioneer

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
pheromone concentration bound

0.1 20,080 20,080 20,080 6320 3440 2700 3520 3900 4680
0.2 20,080 20,080 20,080 3340 2880 3520 3360 4400 7020
0.3 20,080 20,080 20,080 5600 2000 2400 2600 3140 6000
0.4 20,080 20,080 20,080 2340 1840 3720 3560 3380 3590
0.5 20,080 20,080 20,080 2080 2440 3300 3420 3800 4500
0.6 20,080 20,080 20,080 4200 3300 2600 2800 3820 3860
0.7 20,080 20,080 20,080 3080 1900 2440 2160 3560 3040
0.8 20,080 20,080 2780 1340 2640 2940 2900 3680 3860
0.9 20,080 20,080 6820 2260 3240 2840 2800 2000 2260

4.3. Experiment of Classical Benchmark Sets

In order to evaluate the effectiveness and accuracy of the proposed PBA-PDM, a con-
ventional set of classical test functions is selected in Appendices A and B. This benchmark
set includes two main groups of benchmark landscapes: unimodal (UM), which are selected
Mathematics 2022, 10, 2211 15 of 38

in Appendix A, and multimodal (MM), which are selected in Appendix B. All test functions
are minimization problems. The performance of the algorithm was analyzed in two parts:
qualitative and quantitative.
The qualitative part analyzed the superiority of the algorithm through the convergence
curve. Figures 8 and 9 show the convergence characteristics of PBA-PDM and competitors
in tackling the UM and MM classical test functions. It is worth noting that when dealing
with most optimization problems, PBA-PDM has a better performance than PBA, ECLPSO,
CCPSO, IRGA, L-shade, and EBothCMAR in terms of the convergence speed and accu-
racy. For low-dimensional optimization problems, PBA-PDM’s convergence speed and
accuracy are better than most competing algorithms. When dealing with high-dimensional
problems, PBA-PDM still maintains good performance, and similar convergence trends
can be observed in F1, F2, F4, F5, F14, F15, and F20. When dealing with F5 and F18, it can
be observed that ECLPSO and CCPSO have a faster convergence rate than PBA-PDM in
the initial stage. However, ECLPSO falls into a local optimum in the later stages of the
iteration, resulting in a stagnation of the convergence curve while PBA-PDM continues to
approach a better solution. For F15, PBA-PDM converges slightly less well than ECLPSO
and L-SHADE, but it still finds a very competitive optimal solution within a specified
number of iterations. The results of the quantitative analysis show that the PBA-PDM
proposed in this paper has an excellent performance in solving both low-dimensional and
high-dimensional optimization problems with the same population and iteration compared
Mathematics 2022, 10, x FOR PEER REVIEW 16 of 35
to other state-of-the-art competitors. The significant performance of PBA-PDM is probably
due to four reasons.

Figure 8. Convergence curves of the unimodal function.

Figure 8. Convergence curves of the unimodal function.
Mathematics 2022, 10, 2211 16 of 38

Figure 8. Convergence curves of the unimodal function.

Figure 9.
Figure 9. Convergence
Convergence curves
curves of
of the
the multimodal
multimodal function.
function.

Table 4the
Firstly, shows the obtained
sorting results the
process divides for PBA-PDM compared
whole population intowith other
several optimizerseach
subgroups, in
tackling
of which the
has UM functions.
one and only oneTable 5 shows
pioneer the obtained
as a leader to guideresults in dealing
the search withSubgroups
direction. the MM
functions.
are updatedNoteandthat the best
reordered results
after eachiniteration,
the comparison
ensuringfor
theeach test function
population are shown
is always in
searching
in the optimal direction. At the same time, the sorting process ensures that individuals of
the population always have the desired fitness value by eliminating poorer individuals.
Secondly, the exploration process has different strategies. Each strategy simulates the
realistic behavior of insect populations in nature and aims to maintain population diversity
while approximating a potentially optimal solution.
Thirdly, in the exploitation process, the female (pioneer) reproduces offspring and
adds them to the next generation. Therefore, optimum individuals in each subgroup can
exploit the search space more effectively by combining desirable solutions from different
high-quality beetles.
For the last reason, after one iteration is finished, the whole population is updated and
combined, and the new population is reordered and selected, which not only speeds up the
convergence of the algorithm but also reduces the computational load.
Table 4 shows the obtained results for PBA-PDM compared with other optimizers
in tackling the UM functions. Table 5 shows the obtained results in dealing with the
MM functions. Note that the best results in the comparison for each test function are
shown in bold. From Table 4, it can be concluded that according to the Avg and Std
obtained from the experiments, PBA-PDM can identify the best solution in dealing with
most problems compared with other optimizers in dealing with 70.0% of these 12 UM
classical benchmark functions. In F3, F4, F6, and F8-F12, the results are significantly better
than other metaheuristic algorithms. Although IRGA and FAEICA performs better than
PBA-PDM in dealing with F1 and F5, the difference in the average they achieved was only
4.9 × 10−14 and 8.7 × 10− 25 , respectively. In F2, PBA-PDM achieves the best standard
deviation and the second smallest optimal solution compared to the first-ranked competitor.
Even though IRGA, FAEICA, and L-SHADE are ranked first in several tests, PBA-PDM can
tackle 9/12 UM problems with the most satisfactory results. Table 5 shows the solutions of
eight algorithms in tackling F13–F20. As the results illustrate, the PBA-PDM can obtain the
minimum Avg and Std compared to the other optimizers on F13–F17 and F20. The Avg
of PBA-PDM and EBOwithCMAR on F19 are basically the same: EBOwithCMAR simply
obtains the better Std value. The results of PBA-PDM outperformed other state-of-the-art
algorithms in tackling 75% of these MM test functions.
Mathematics 2022, 10, 2211 17 of 38

Table 4. Statistical results of the unimodal test functions.

No. PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 2.41 × 10−13 4.08 × 10−09 1.08 × 10−11 3.08 × 10−12 1.92 × 10−13 2.85 × 10−13 2.72 × 10−13 1.12 × 10−12 3.11 × 10−11 2.67 × 10−10
F1
STD 2.15 × 10−13 4.73 × 10−09 6.16 × 10−11 2.49 × 10−12 6.16 × 10−26 2.63 × 10−13 1.04 × 10−13 5.31 × 10−12 3.41 × 10−11 1.45 × 10−10
Mean 1.57 × 10−27 1.14 × 10−25 2.76 × 10−25 2.35 × 10−25 1.56 × 10−24 4.31 × 10−24 5.41 × 10−24 7.17 × 10−25 8.56 × 10−30 1.42 × 10−24
F2
STD 5.12 × 10−27 7.63 × 10−25 4.52 × 10−25 2.76 × 10−25 1.83 × 10−24 2.28 × 10−24 8.14 × 10−24 3.96 × 10−25 9.50 × 10−27 6.79 × 10−23
Mean 2.59 × 10−10 1.78 × 10−08 4.19 × 10−09 3.94 × 10−10 3.25 × 10−09 4.32 × 10−10 5.54 × 10−10 4.63 × 10−10 1.24 × 10−07 6.89 × 10−06
F3
STD 1.05 × 10−10 1.49 × 10−08 6.77 × 10−10 4.14 × 10−10 3.40 × 10−09 4.57 × 10−10 1.61 × 10−10 3.25 × 10−10 4.08 × 10−07 3.90 × 10−05
Mean 1.15 × 10−25 2.99 × 10−22 5.67 × 10−22 2.21 × 10−20 3.28 × 10−20 2.18 × 10−22 2.08 × 10−23 3.51 × 10−22 7.18 × 10−21 1.19 × 10−23
F4
STD 1.65 × 10−24 3.17 × 10−21 1.35 × 10−22 2.71 × 10−20 2.12 × 10−20 3.53 × 10−22 2.59 × 10−23 5.08 × 10−22 2.21 × 10−20 2.39 × 10−23
Mean 8.39 × 10−24 2.08 × 10−20 8.73 × 10−24 3.51 × 10−22 1.37 × 10−22 2.14 × 10−22 9.28 × 10−24 7.52 × 10−24 9.04 × 10−24 9.55 × 10−24
F5
STD 4.65 × 10−24 2.59 × 10−20 8.92 × 10−24 5.08 × 10−21 6.69 × 10−22 4.49 × 10−22 6.91 × 10−24 3.41 × 10−24 2.81 × 10−21 1.50 × 10−23
Mean −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00 −1.99 × 10+00
F6
STD 0.00 × 10+00 2.19 × 10−10 3.29 × 10−12 5.55 × 10−16 5.42 × 10−13 2.21 × 10−17 7.36 × 10−11 5.59 × 10−12 1.26 × 10−12 4.97 × 10−11
Mean 1.37 × 10−13 3.98 × 10−10 5.15 × 10−12 4.43 × 10−11 5.41 × 10−08 2.18 × 10−09 9.91 × 10−10 4.18 × 10−13 6.19 × 10−13 4.44 × 10−13
F7
STD 1.13 × 10−13 4.09 × 10−10 8.72 × 10−12 7.36 × 10−11 3.15 × 10−07 3.11 × 10−09 2.96 × 10−03 3.45 × 10−13 2.26 × 10−13 2.81 × 10−13
Mean 1.49 × 10−25 1.61 × 10−25 7.37 × 10−24 2.28 × 10−25 6.25 × 10−20 3.75 × 10−23 4.08 × 10−22 3.63 × 10−23 1.45 × 10−23 4.22 × 10−23
F8
STD 5.46 × 10−25 1.47 × 10−25 3.74 × 10−24 2.63 × 10−25 1.49 × 10−19 2.28 × 10−23 4.74 × 10−22 4.43 × 10−23 5.35 × 10−24 2.14 × 10−22
Mean −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00 −1.91 × 10+00
F9
STD 2.79 × 10−12 1.79 × 10−08 1.66 × 10−09 2.18 × 10−07 2.56 × 10−09 2.08 × 10−08 1.67 × 10−09 2.44 × 10−11 2.50 × 10−09 1.12 × 10−07
Mean 2.29 × 10−35 4.38 × 10−35 5.52 × 10−26 3.75 × 10−24 1.96 × 10−25 2.25 × 10−33 1.20 × 10−28 2.68 × 10−34 2.68 × 10−35 9.26 × 10−27
F10
STD 1.59 × 10−35 4.90 × 10−35 4.72 × 10−26 2.28 × 10−24 1.04 × 10−25 3.52 × 10−33 1.66 × 10−28 5.42 × 10−34 2.82 × 10−35 1.69 × 10−26
Mean 1.14 × 10−41 5.52 × 10−38 8.36 × 10−30 5.03 × 10−29 1.28 × 10−30 3.41 × 10−35 2.35 × 10−39 6.31 × 10−34 2.15 × 10−35 8.59 × 10−41
F11
STD 5.44 × 10−40 4.72 × 10−38 1.39 × 10−29 1.22 × 10−29 5.74 × 10−30 1.78 × 10−35 2.10 × 10−39 3.05 × 10−34 1.41 × 10−35 1.24 × 10−41
Mean −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 −1.03 × 10+00 1.63 × 10+08
F12
STD 0.00 × 10+00 1.41 × 10−16 5.24 × 10−12 1.56 × 10−15 0.00 × 10+00 0.00 × 10+00 7.47 × 10−15 1.23 × 10−16 6.66 × 10−16 1.12 × 10+07
Mathematics 2022, 10, 2211 18 of 38

Table 5. Statistical results of the multimodal test functions.

No. PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00 1.00 × 10+00
F13
STD 0.00 × 10+00 0.00 × 10+00 1.58 × 10−13 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00
Mean −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00 −1.80 × 10+00
F14
STD 0.00 × 10+00 1.78 × 10−15 1.53 × 10−18 5.31 × 10−12 6.78 × 10−16 3.98 × 10−12 8.10 × 10−18 1.56 × 10−14 6.66 × 10−16 1.99 × 10−13
Mean 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 1.17 × 10−17 0.00 × 10+00
F15
STD 0.00 × 10+00 2.11 × 10−18 4.52 × 10−21 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 2.18 × 10−17 0.00 × 10+00
Mean 1.34 × 10−12 3.62 × 10−12 8.59 × 10−10 5.46 × 10−11 3.91 × 10−09 2.51 × 10−09 1.62 × 10−09 4.33 × 10−10 4.37 × 10−09 3.36 × 10−09
F16
STD 7.17 × 10−12 4.43 × 10−12 7.72 × 10−10 2.85 × 10−10 7.86 × 10−09 3.03 × 10−09 2.77 × 10−09 4.58 × 10−10 2.09 × 10−09 5.66 × 10−09
Mean 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00 3.00 × 10+00
F17
STD 0.00 × 10+00 3.4 × 10−15 0.00 × 10+00 1.15 × 10−14 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 1.44 × 10−14
Mean 1.00 × 10−01 2.27 × 10−03 1.02 × 10−01 3.39 × 10−03 2.55 × 10−02 5.15 × 10−02 7.39 × 10−02 5.38 × 10−02 6.19 × 10−03 3.87 × 10−02
F18
STD 0.00 × 10+00 2.44 × 10−03 5.73 × 10−01 5.88 × 10−03 5.17 × 10−01 1.96 × 10−05 3.49 × 10−02 3.47 × 10−01 2.26 × 10−01 8.08 × 10−01
Mean −9.49 × 10+02 −9.44 × 10+02 −9.47 × 10+02 −9.55 × 10+02 −9.57 × 10+02 −9.58 × 10+02 −9.49 × 10+02 −9.53 × 10+02 −9.44 × 10+02 −9.59 × 10+02
F19
STD 3.74 × 10+02 2.73 × 10+01 1.16 × 10+01 1.26 × 10+01 6.05 × 10+01 8.47 × 10+01 1.46 × 10+01 1.59 × 10+01 2.81 × 10+01 3.67 × 10−01
Mean 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 2.45 × 10−20 4.19 × 10−21 3.35 × 10−19 2.08 × 10−18 3.00 × 10−20 6.05 × 10−17 3.44 × 10−18
F20
STD 0.00 × 10+00 0.00 × 10+00 0.00 × 10+00 1.22 × 10−20 2.33 × 10−21 1.25 × 10−19 2.59 × 10−18 3.17 × 10−20 1.04 × 10−15 5.64 × 10−18
Mathematics 2022, 10, 2211 19 of 38

These quantitative results of different tests show the superior performance of PBA-
PDM against OPGPBA and other competitors. For most classical test functions, the
proposed PBA-PDM can achieve desirable solutions and does not encounter obstacles
when dealing with the high dimensionality problem. Hence, it is detected that PBA-
PDM is capable of identifying most high-accuracy solutions and excelling in many well-
established algorithms in most of the UM and MM classical test functions with strong
convergence capability.
In addition, Table 6 summarizes the improvement percentage of the best solution
obtained by PBA-PDM compared to its secondary competitors. Note that since most of the
algorithms obtained the same optimal results for F6, F9, F12-F15, F17, and F20, they are not
computed in this table. The results show that the proposed PBA-PDM not only converges
to the global optimum after a finite number of iterations but also significantly improves the
accuracy of the optimal solution compared to other algorithms.

Table 6. Percentage improvement of PBA-PDM compared with secondary competitors.

Percentage
No. PBA-PDM OPGPBA CCPSO FAEICA L-SHADE EBOwithCMAR
Improvement
F3 Mean 2.59 × 10−10 3.94 × 10−10 34.26%
F4 Mean 1.15 × 10−25 1.19 × 10−23 99.03%
F7 Mean 1.37 × 10−13 4.18 × 10−13 67.22%
F8 Mean 1.49 × 10−25 1.61 × 10−25 7.45%
F10 Mean 2.29 × 10−35 2.68 × 10−35 14.55%
F11 Mean 1.14 × 10−41 8.59 × 10−41 86.73%
F16 Mean 1.34 × 10−12 3.62 × 10−12 62.98%

4.4. Experiment of CEC2017 Benchmark Sets

It can be seen from the conclusions in the previous section that many algorithms are
able to achieve satisfactory results in the classic test function, but when the dimensionality
of the problem continues to increase and the optimal value of each dimension is not the
same, the performance of many algorithms deteriorates [27]. This phenomenon is known
as center-seeking bias (CSB) and initialization region bias (IRB) [28]. CSB refers to the fact
that no matter how the fitness of the objective function changes, the global optimal solution
obtained by the algorithm will eventually move to the origin as the iteration progresses [29].
Meanwhile, IRB refers to the fact that the optimal solution obtained by an algorithm always
moves around the optimal point of its initial population. If the algorithm has structural
bias, then the solution of the problem will always fail to cover the entire search space;
therefore, if the global optimal value of the problem to be processed is not at the origin, or
the initial population randomly generated by PRNGS is not located near the global optimal
value, the search process will stagnate at the local optimal value. The CEC2017 benchmark
functions were selected to verify whether the proposed PBA-PDM has structural bias or
not. The results of the PBA-PDM and other competitors for the CEC2017 test functions are
presented in Tables 7–9.
Mathematics 2022, 10, 2211 20 of 38

Table 7. Comparative results of the CEC2017 benchmark test functions in 10 Dim.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 6.95 ×10−04 1.99 ×10+05 1.76 × 10+05 1.15 × 10+08 9.48 × 10+08 5.70 × 10+09 8.51 ×10+00 5.43 ×10−01 1.60 ×10−02 1.49 × 10−06
F1
STD 1.57 × 10−04 4.13 × 10+05 1.11 × 10+05 3.75 × 10+07 8.28 × 10+08 4.38 × 10+09 1.46 × 10+03 3.74 × 10+01 2.92 × 10−02 1.51 × 10−06
Mean 8.75 × 10−06 1.00 × 10+10 2.84 × 10+02 1.99 × 10+27 9.92 × 10+08 8.73 × 10+09 1.54 × 10−05 2.93 × 10+01 4.06 × 10−08 3.00 × 10−09
F2
STD 4.48 × 10+01 1.43 × 10+11 4.88 × 10+02 8.80 × 10+27 3.64 × 10+09 1.91 × 10+13 1.42 × 10−03 2.80 × 10+01 5.37 × 10−08 8.85 × 10−09
Mean 1.06 × 10−10 7.62 × 10+05 8.35 × 10−01 1.68 × 10+05 1.96 × 10+04 1.02 × 10+04 7.12 × 10−07 3.61 × 10+00 3.29 × 10−03 1.18 × 10−07
F3
STD 2.25 × 10−10 1.32 × 10+06 3.30 × 10−01 1.87 × 10+04 1.47 × 10+04 9.48 × 10+03 3.62 × 10−05 3.56 × 10+01 2.50 × 10−01 3.43 × 10−07
Mean 1.74 × 10+00 7.11 × 10+04 1.61 × 10+01 2.79 × 10+02 9.47 × 10+01 1.98 × 10+02 2.31 × 10−02 5.72 × 10−01 9.10 × 10+00 1.30 × 10+00
F4
STD 8.62 × 10−01 9.46 × 10+03 2.45 × 10+01 2.66 × 10+01 5.89 × 10+01 4.56 × 10+02 1.97 × 10+00 2.74 × 10−01 7.30 × 10−01 5.28 × 10−01
Mean 2.98 × 10+00 1.03 × 10+03 3.88 × 10+01 3.66 × 10+02 7.40 × 10+01 6.80 × 10+01 5.97 × 10+00 1.99 × 10+01 1.90 × 10+01 2.56 × 10+00
F5
STD 1.50 × 10+00 4.81 × 10+01 1.81 × 10+01 5.64 × 10+01 2.57 × 10+01 1.93 × 10+01 9.32 × 10+00 1.07 × 10+01 3.47 × 10+00 1.23 × 10+00
Mean 2.30 × 10−06 1.34 × 10+02 2.58 × 10+01 3.69 × 10+00 4.41 × 10+01 4.89 × 10+01 1.03 × 10−03 1.58 × 10+00 7.48 × 10−05 1.71 × 10−04
F6
STD 1.19 × 10−03 6.68 × 10+00 1.09 × 10+01 5.52 × 10−01 1.31 × 10+01 8.44 × 10+00 2.83 × 10−01 9.25 × 10+00 3.76 × 10−05 8.09 × 10−05
Mean 1.22 × 10+01 4.51 × 10+03 7.68 × 10+01 4.79 × 10+02 1.06 × 10+02 1.33 × 10+02 1.55 × 10+01 3.82 × 10+01 3.18 × 10+01 1.18 × 10+01
F7
STD 3.65 × 10+00 2.59 × 10+02 1.78 × 10+01 1.82 × 10+01 2.76 × 10+01 1.59 × 10+01 8.43 × 10+00 1.96 × 10+01 3.13 × 10+00 5.86 × 10−01
Mean 2.98 × 10+00 1.04 × 10+03 2.64 × 10+01 3.84 × 10+02 5.30 × 10+01 6.34 × 10+01 5.97 × 10+00 1.49 × 10+01 1.86 × 10+01 3.04 × 10+00
F8
STD 1.70 × 10+00 4.70 × 10+01 1.12 × 10+01 3.83 × 10+01 1.42 × 10+01 1.06 × 10+01 7.63 × 10+00 7.20 × 10+00 3.83 × 10+00 1.61 × 10+00
Mean 1.02 × 10−13 7.49 × 10+04 4.21 × 10+02 2.51 × 10+02 1.00 × 10+03 8.32 × 10+02 1.73 × 10−11 7.80 × 10+00 4.35 × 10−11 0.00 × 10+00
F9
STD 8.09 × 10−14 6.89 × 10+03 2.16 × 10+02 7.17 × 10+01 7.17 × 10+02 3.25 × 10+02 2.16 × 10+01 1.41 × 10+01 2.81 × 10−11 0.00 × 10+00
Mean 1.27 × 10+02 1.55 × 10+04 9.82 × 10+02 1.35 × 10+04 1.43 × 10+03 1.76 × 10+03 1.33 × 10+02 1.18 × 10+01 8.98 × 10+02 5.78 × 10+02
F10
STD 1.22 × 10+02 5.91 × 10+02 2.82 × 10+02 3.12 × 10+02 3.24 × 10+02 2.32 × 10+02 3.23 × 10+02 3.45 × 10+01 2.17 × 10+02 1.71 × 10+02
Mean 1.12 × 10+00 5.57 × 10+04 5.59 × 10+01 5.12 × 10+02 4.21 × 10+02 4.97 × 10+02 1.21 × 10+00 1.29 × 10+01 3.36 × 10+00 3.45 × 10+00
F11
STD 2.49 × 10+00 1.44 × 10+04 3.75 × 10+01 1.45 × 10+02 7.09 × 10+02 6.37 × 10+03 1.45 × 10+01 4.18 × 10+01 1.45 × 10+00 5.96 × 10−01
Mean 6.43 × 10+02 9.60 × 10+02 5.70 × 10+05 1.18 × 10+07 7.84 × 10+06 1.33 × 10+07 5.87 × 10+02 2.87 × 10+03 1.34 × 10+02 1.52 × 10+02
F12
STD 1.32 × 10+04 1.28 × 10+10 9.34 × 10+05 5.63 × 10+06 7.08 × 10+06 5.04 × 10+08 1.25 × 10+04 1.61 × 10+03 1.35 × 10+02 5.74 × 10+01
Mean 4.52 × 10+00 3.84 × 10+01 1.23 × 10+04 3.62 × 10+03 1.94 × 10+04 1.05 × 10+05 7.94 × 10+00 2.71 × 10+02 6.42 × 10+00 7.11 × 10+00
F13
STD 1.02 × 10+00 2.66 × 10+03 1.01 × 10+04 4.71 × 10+03 1.86 × 10+04 7.56 × 10+07 1.47 × 10+02 7.46 × 10+03 3.81 × 10+00 2.68 × 10+00
Mean 2.82 × 10+01 1.02 × 10+08 1.10 × 10+02 5.94 × 10+04 3.05 × 10+03 2.28 × 10+02 8.31 × 10+00 4.82 × 10+01 8.24 × 10+00 1.30 × 10+01
F14
STD 4.02 × 10+01 4.33 × 10+07 3.02 × 10+01 3.54 × 10+04 2.85 × 10+03 2.54 × 10+05 2.14 × 10+01 1.26 × 10+02 7.04 × 10+00 4.85 × 10+00
Mean 9.00 × 10+00 1.50 × 10+10 3.63 × 10+02 2.86 × 10+03 1.96 × 10+04 2.44 × 10+03 3.54 × 10+00 1.30 × 10+02 2.24 × 10+00 9.44 × 10−01
F15
STD 1.14 × 10+02 4.21 × 10+09 6.40 × 10+02 2.27 × 10+03 1.90 × 10+04 7.18 × 10+05 3.34 × 10+01 2.42 × 10+03 1.24 × 10+00 2.73 × 10−01
Mean 5.50 × 10−01 8.77 × 10+03 1.78 × 10+02 3.03 × 10+03 4.88 × 10+02 5.35 × 10+02 9.29 × 10−01 4.26 × 10+00 7.94 × 10+00 8.64 × 10+00
F16
STD 4.03 × 10−01 8.63 × 10+02 1.25 × 10+02 3.74 × 10+02 2.03 × 10+02 1.51 × 10+02 8.74 × 10+01 1.38 × 10+01 4.03 × 10+00 3.14 × 10+00
Mathematics 2022, 10, 2211 21 of 38

Table 7. Cont.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 1.04 × 10+00 1.13 × 10+05 6.32 ×10+01 1.93 ×10+03 1.46 ×10+02 1.40 ×10+02 5.94 ×10+00 3.73 ×10+01 4.99 ×10+01 2.73 × 10+01
F17
STD 1.77 × 10+01 8.17 × 10+04 2.46 × 10+01 1.58 × 10+02 7.95 × 10+01 1.37 × 10+02 3.14 × 10+01 2.16 × 10+01 7.13 × 10+00 4.20 × 10+00
Mean 2.89 × 10+02 2.27 × 10+02 1.31 × 10+04 2.28 × 10+06 1.90 × 10+04 5.21 × 10+06 3.18 × 10+01 1.43 × 10+03 8.48 × 10+00 1.96 × 10+00
F18
STD 2.27 × 10+03 1.39 × 10+03 1.26 × 10+04 1.37 × 10+06 1.37 × 10+04 1.66 × 10+08 5.21 × 10+03 1.39 × 10+04 8.63 × 10+00 1.40 × 10+00
Mean 5.49 × 10+00 6.58 × 10+00 7.25 × 10+03 1.00 × 10+04 8.14 × 10+05 1.41 × 10+05 2.76 × 10+00 2.40 × 10+01 1.00 × 10+00 9.94 × 10−01
F19
STD 2.32 × 10+02 1.48 × 10+00 7.76 × 10+03 8.93 × 10+03 2.07 × 10+06 7.11 × 10+06 2.58 × 10+01 7.68 × 10+01 5.38 × 10−01 3.16 × 10−01
Mean 8.52 × 10+00 2.90 × 10+03 1.11 × 10+02 1.57 × 10+03 2.54 × 10+02 1.56 × 10+02 9.95 × 10−01 2.47 × 10+01 4.11 × 10+01 1.20 × 10+01
F20
STD 9.47 × 10+00 2.55 × 10+02 5.38 × 10+01 3.75 × 10+02 1.01 × 10+02 1.04 × 10+02 1.32 × 10+01 6.16 × 10+01 1.13 × 10+01 4.93 × 10+00
Mean 1.04 × 10+02 1.26 × 10+03 1.59 × 10+02 5.67 × 10+02 2.61 × 10+02 1.84 × 10+02 1.00 × 10+02 1.02 × 10+02 1.12 × 10+02 1.56 × 10+02
F21
STD 2.12 × 10+01 5.22 × 10+01 7.00 × 10+01 7.65 × 10+01 3.65 × 10+01 3.90 × 10+01 6.08 × 10+01 5.73 × 10+01 3.55 × 10+01 4.59 × 10+01
Mean 1.00 × 10+02 1.61 × 10+04 1.12 × 10+02 5.54 × 10+03 6.17 × 10+02 2.57 × 10+02 2.21 × 10+01 1.01 × 10+02 1.00 × 10+02 1.00 × 10+02
F22
STD 1.03 × 10+00 4.52 × 10+02 9.83 × 10+00 6.55 × 10+03 7.61 × 10+02 3.72 × 10+02 1.79 × 10+01 7.16 × 10+00 5.96 × 10−06 1.47 × 10−01
Mean 3.04 × 10+02 2.31 × 10+03 3.56 × 10+02 8.00 × 10+02 3.73 × 10+02 3.60 × 10+02 3.05 × 10+02 3.28 × 10+02 3.03 × 10+02 3.05 × 10+02
F23
STD 2.10 × 10+00 1.50 × 10+02 2.67 × 10+01 6.21 × 10+01 3.32 × 10+01 3.43 × 10+01 9.87 × 10+00 1.15 × 10+01 2.97 × 10+00 2.80 × 10+00
Mean 2.03 × 10+02 2.67 × 10+03 2.70 × 10+02 8.83 × 10+02 4.12 × 10+02 3.03 × 10+02 1.00 × 10+02 3.28 × 10+02 1.83 × 10+02 2.23 × 10+02
F24
STD 2.46 × 10+01 2.14 × 10+02 7.64 × 10+01 4.62 × 10+01 3.28 × 10+01 5.21 × 10+01 8.12 × 10+01 1.15 × 10+01 1.11 × 10+02 8.27 × 10+01
Mean 3.98 × 10+02 3.57 × 10+04 4.42 × 10+02 6.23 × 10+02 5.53 × 10+02 6.30 × 10+02 3.98 × 10+02 2.33 × 10+01 4.09 × 10+02 4.19 × 10+02
F25
STD 1.62 × 10+01 4.77 × 10+03 3.64 × 10+01 2.15 × 10+01 8.75 × 10+01 3.04 × 10+02 2.22 × 10+01 1.19 × 10+02 1.94 × 10+01 2.31 × 10+01
Mean 3.00 × 10+02 2.22 × 10+04 5.75 × 10+02 4.34 × 10+03 1.20 × 10+03 8.80 × 10+02 2.00 × 10+02 3.98 × 10+02 2.97 × 10+02 2.87 × 10+02
F26
STD 2.31 × 10+02 2.07 × 10+03 3.13 × 10+02 7.46 × 10+02 5.77 × 10+02 4.37 × 10+02 7.11 × 10+01 2.71 × 10+01 1.83 × 10+01 3.29 × 10+01
Mean 3.93 × 10+02 4.31 × 10+03 4.09 × 10+02 8.07 × 10+02 4.75 × 10+02 4.52 × 10+02 3.89 × 10+02 8.51 × 10−03 3.94 × 10+02 3.92 × 10+02
F27
STD 2.62 × 10+00 4.68 × 10+02 3.25 × 10+01 6.74 × 10+01 5.59 × 10+01 5.86 × 10+01 3.00 × 10+00 2.08 × 10+02 1.01 × 10+00 2.40 × 10+00
Mean 2.84 × 10+02 1.53 × 10+04 5.27 × 10+02 6.14 × 10+02 7.84 × 10+02 7.57 × 10+02 3.00 × 10+02 3.90 × 10+02 3.00 × 10+02 3.17 × 10+02
F28
STD 1.18 × 10+01 1.90 × 10+03 1.34 × 10+02 3.38 × 10+01 1.49 × 10+02 1.48 × 10+02 1.45 × 10+02 1.25 × 10+01 1.35 × 10−03 6.11 × 10+01
Mean 2.55 × 10+02 1.11 × 10+05 3.97 × 10+02 1.37 × 10+03 6.14 × 10+02 5.75 × 10+02 2.41 × 10+02 3.00 × 10+02 2.61 × 10+02 2.69 × 10+02
F29
STD 1.19 × 10+01 9.63 × 10+04 9.41 × 10+01 4.71 × 10+02 1.68 × 10+02 1.14 × 10+02 4.87 × 10+01 1.25 × 10+02 9.79 × 10+00 9.26 × 10+00
Mean 1.05 × 10+03 9.89 × 10+09 6.24 × 10+05 2.81 × 10+06 3.82 × 10+06 4.78 × 10+06 7.37 × 10+02 2.52 × 10+02 4.17 × 10+02 4.08 × 10+02
F30
STD 3.41 × 10+05 2.29 × 10+09 1.06 × 10+06 4.03 × 10+05 4.02 × 10+06 3.13 × 10+07 3.48 × 10+05 6.80 × 10+01 3.56 × 10+01 1.54 × 10+01
Mathematics 2022, 10, 2211 22 of 38

Table 8. Comparative results of the CEC2017 benchmark test functions in 30 Dim.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 4.96 ×10+01 2.58 × 10+02 7.54 × 10+06 5.51 × 10+03 5.38 × 10+09 4.82 × 10+10 6.08 ×10+01 6.54 ×10+05 0.00 × 10+00 3.53 × 10−07
F1
STD 2.83 × 10+03 2.53 × 10+02 1.92 × 10+06 4.42 × 10+03 2.34 × 10+09 8.94 × 10+09 3.29 × 10+03 4.18 × 10+06 0.00 × 10+00 2.34 × 10−07
Mean 4.96 × 10+01 1.03 × 10+09 2.40 × 10+16 2.47 × 10+50 2.25 × 10+35 1.27 × 10+42 3.43 × 10+07 3.00 × 10+16 3.23 × 10−02 2.36 × 10+00
F2
STD 2.83 × 10+03 4.05 × 10+09 5.91 × 10+16 1.03 × 10+51 1.03 × 10+36 6.93 × 10+47 9.48 × 10+15 1.07 × 10+21 1.80 × 10−01 1.00 × 10+01
Mean 2.44 × 10+04 4.61 × 10+03 2.36 × 10+03 6.28 × 10+04 2.76 × 10+05 8.38 × 10+04 2.87 × 10+02 1.75 × 10+04 0.00 × 10+00 2.92 × 10−01
F3
STD 6.40 × 10+03 1.03 × 10+04 9.90 × 10+02 8.33 × 10+03 8.21 × 10+04 3.12 × 10+04 8.89 × 10+02 8.97 × 10+03 0.00 × 10+00 6.83 × 10−01
Mean 7.26 × 10+01 8.75 × 10+01 1.25 × 10+02 9.70 × 10+01 9.60 × 10+02 7.30 × 10+03 6.02 × 10+01 9.63 × 10+01 8.16 × 10+01 7.54 × 10+01
F4
STD 1.52 × 10+01 5.77 × 10+00 4.03 × 10+01 1.79 × 10+01 3.98 × 10+02 5.04 × 10+03 1.60 × 10+01 2.26 × 10+01 2.67 × 10+00 9.32 × 10+00
Mean 1.17 × 10+01 7.63 × 10+01 2.01 × 10+02 8.48 × 10+01 3.65 × 10+02 3.98 × 10+02 9.45 × 10+01 1.29 × 10+02 1.22 × 10+01 1.08 × 10+01
F5
STD 1.14 × 10+01 2.31 × 10+01 3.27 × 10+01 6.31 × 10+01 5.50 × 10+01 4.05 × 10+01 2.46 × 10+01 2.78 × 10+01 2.32 × 10+00 3.94 × 10+00
Mean 2.61 × 10−02 1.77 × 10−02 5.53 × 10+01 2.62 × 10−02 8.27 × 10+01 9.56 × 10+01 7.19 × 10+00 3.38 × 10+01 1.56 × 10−08 2.18 × 10−05
F6
STD 2.92 × 10−02 6.95 × 10−03 6.41 × 10+00 9.81 × 10−03 1.12 × 10+01 4.54 × 10+00 6.67 × 10+00 7.03 × 10+00 6.02 × 10−08 4.82 × 10−06
Mean 4.75 × 10+01 9.62 × 10+01 5.05 × 10+02 2.01 × 10+02 5.99 × 10+02 7.11 × 10+02 1.48 × 10+02 3.21 × 10+02 4.36 × 10+01 3.71 × 10+01
F7
STD 2.22 × 10+01 4.15 × 10+01 6.14 × 10+01 4.50 × 10+01 7.40 × 10+01 4.75 × 10+01 4.56 × 10+01 8.33 × 10+01 2.49 × 10+00 2.26 × 10+00
Mean 1.31 × 10+01 8.34 × 10+01 1.47 × 10+02 7.58 × 10+01 2.79 × 10+02 3.29 × 10+02 8.76 × 10+01 1.20 × 10+02 1.32 × 10+01 8.92 × 10+00
F8
STD 1.48 × 10+01 2.52 × 10+01 2.65 × 10+01 6.15 × 10+01 4.85 × 10+01 3.09 × 10+01 1.70 × 10+01 2.63 × 10+01 1.81 × 10+00 2.15 × 10+00
Mean 7.10 × 10−05 9.46 × 10−07 4.72 × 10+03 1.41 × 10−01 1.00 × 10+04 1.04 × 10+04 7.02 × 10+02 2.99 × 10+03 0.00 × 10+00 0.00 × 10+00
F9
STD 1.64 × 10−01 1.38 × 10−06 6.70 × 10+02 1.91 × 10−01 3.17 × 10+03 1.26 × 10+03 5.65 × 10+02 6.83 × 10+02 0.00 × 10+00 0.00 × 10+00
Mean 2.69 × 10+03 4.76 × 10+03 4.43 × 10+03 7.12 × 10+03 6.62 × 10+03 7.96 × 10+03 3.26 × 10+03 3.17 × 10+03 1.43 × 10+03 3.76 × 10+03
F10
STD 1.28 × 10+03 4.97 × 10+02 6.29 × 10+02 2.88 × 10+02 7.98 × 10+02 3.96 × 10+02 5.61 × 10+02 3.17 × 10+02 2.27 × 10+02 4.85 × 10+02
Mean 2.39 × 10+01 6.68 × 10+01 1.59 × 10+02 6.30 × 10+01 9.21 × 10+03 1.10 × 10+04 5.59 × 10+01 1.03 × 10+02 1.14 × 10+01 3.53 × 10+01
F11
STD 3.64 × 10+01 2.15 × 10+01 4.53 × 10+01 3.46 × 10+01 4.84 × 10+03 6.36 × 10+03 4.53 × 10+01 4.48 × 10+01 1.81 × 10+01 2.30 × 10+01
Mean 2.56 × 10+04 9.72 × 10+03 7.00 × 10+06 2.51 × 10+05 4.49 × 10+08 9.58 × 10+09 2.75 × 10+04 9.20 × 10+05 3.03 × 10+02 1.33 × 10+03
F12
STD 1.12 × 10+05 2.62 × 10+03 4.66 × 10+06 1.99 × 10+05 2.94 × 10+08 3.81 × 10+09 7.93 × 10+04 9.24 × 10+06 1.82 × 10+02 6.34 × 10+02
Mean 3.69 × 10+01 1.29 × 10+02 1.69 × 10+05 1.18 × 10+04 1.01 × 10+07 7.67 × 10+09 1.44 × 10+03 1.41 × 10+04 1.87 × 10+01 6.10 × 10+01
F13
STD 4.60 × 10+01 2.69 × 10+01 8.90 × 10+04 1.14 × 10+04 9.73 × 10+06 3.48 × 10+09 1.22 × 10+04 7.33 × 10+04 2.05 × 10+01 1.51 × 10+01
Mean 7.40 × 10+02 7.02 × 10+01 3.92 × 10+04 7.17 × 10+03 3.37 × 10+06 1.75 × 10+06 2.11 × 10+02 5.60 × 10+03 2.01 × 10+01 4.92 × 10+01
F14
STD 4.43 × 10+00 5.72 × 10+00 3.61 × 10+04 6.72 × 10+03 3.75 × 10+06 8.44 × 10+06 3.07 × 10+02 1.22 × 10+05 5.82 × 10+00 4.42 × 10+00
Mean 1.49 × 10+01 4.96 × 10+01 5.92 × 10+04 7.55 × 10+03 7.69 × 10+06 1.68 × 10+09 1.11 × 10+02 1.26 × 10+03 3.15 × 10+00 2.12 × 10+01
F15
STD 2.19 × 10+03 6.99 × 10+00 3.79 × 10+04 7.12 × 10+03 1.08 × 10+07 5.15 × 10+08 7.09 × 10+03 9.73 × 10+03 1.80 × 10+00 5.12 × 10+00
Mathematics 2022, 10, 2211 23 of 38

Table 8. Cont.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 1.75 × 10+01 1.28 × 10+03 1.56 ×10+03 1.05 ×10+03 2.76 ×10+03 3.25 ×10+03 7.12 ×10+02 1.02 ×10+03 1.86 ×10+01 7.02 × 10+02
F16
STD 2.56 × 10+02 1.74 × 10+02 3.53 × 10+02 4.41 × 10+02 7.74 × 10+02 7.64 × 10+02 1.85 × 10+02 1.89 × 10+02 2.13 × 10+01 1.76 × 10+02
Mean 1.39 × 10+01 3.51 × 10+02 7.35 × 10+02 9.11 × 10+01 1.07 × 10+03 1.85 × 10+03 2.97 × 10+02 1.01 × 10+02 2.69 × 10+01 1.51 × 10+02
F17
STD 9.19 × 10+01 8.33 × 10+01 2.89 × 10+02 7.46 × 10+01 3.48 × 10+02 5.58 × 10+02 2.36 × 10+02 2.96 × 10+02 6.64 × 10+00 2.70 × 10+01
Mean 2.43 × 10+01 7.57 × 10+01 7.41 × 10+05 3.68 × 10+05 1.65 × 10+07 1.77 × 10+07 2.25 × 10+04 5.21 × 10+05 2.11 × 10+01 3.50 × 10+01
F18
STD 1.56 × 10+01 1.26 × 10+01 5.47 × 10+05 2.24 × 10+05 1.85 × 10+07 7.38 × 10+07 6.30 × 10+04 1.95 × 10+06 7.46 × 10−01 2.80 × 10+00
Mean 3.25 × 10+02 3.43 × 10+02 1.18 × 10+05 5.89 × 10+03 1.79 × 10+07 1.62 × 10+09 3.75 × 10+02 4.36 × 10+02 5.74 × 10+00 2.15 × 10+01
F19
STD 3.23 × 10+03 3.69 × 10+03 1.18 × 10+05 6.60 × 10+03 1.99 × 10+07 1.00 × 10+09 3.58 × 10+03 2.43 × 10+04 1.89 × 10+00 2.05 × 10+00
Mean 2.08 × 10+01 4.20 × 10+02 6.61 × 10+02 1.03 × 10+02 9.22 × 10+02 1.20 × 10+03 1.98 × 10+02 4.36 × 10+02 2.85 × 10+01 2.11 × 10+02
F20
STD 1.04 × 10+02 8.37 × 10+01 1.86 × 10+02 9.14 × 10+01 2.53 × 10+02 1.07 × 10+02 1.30 × 10+02 2.43 × 10+04 6.56 × 10+00 4.58 × 10+01
Mean 2.11 × 10+02 2.78 × 10+02 4.09 × 10+02 2.77 × 10+02 5.41 × 10+02 6.06 × 10+02 2.83 × 10+02 3.93 × 10+02 2.12 × 10+02 2.14 × 10+02
F21
STD 2.61 × 10+01 3.68 × 10+01 3.91 × 10+01 5.63 × 10+01 6.48 × 10+01 4.05 × 10+01 3.45 × 10+01 1.46 × 10+02 2.91 × 10+00 1.10 × 10+01
Mean 1.00 × 10+02 1.00 × 10+02 2.47 × 10+03 1.00 × 10+02 5.96 × 10+03 7.10 × 10+03 1.00 × 10+02 3.25 × 10+02 1.00 × 10+02 1.00 × 10+02
F22
STD 5.03 × 10+02 4.23 × 10−03 2.27 × 10+03 4.72 × 10−02 1.88 × 10+03 6.57 × 10+02 1.86 × 10−03 3.35 × 10+01 0.00 × 10+00 2.61 × 10−08
Mean 3.54 × 10+02 4.54 × 10+02 7.17 × 10+02 3.87 × 10+02 8.44 × 10+02 1.08 × 10+03 4.41 × 10+02 1.06 × 10+02 3.65 × 10+02 3.72 × 10+02
F23
STD 9.48 × 10+00 4.16 × 10+01 8.44 × 10+01 1.49 × 10+01 1.16 × 10+02 1.22 × 10+02 4.62 × 10+01 1.43 × 10+03 3.54 × 10+00 1.84 × 10+01
Mean 4.39 × 10+02 5.39 × 10+02 8.15 × 10+02 4.89 × 10+02 9.00 × 10+02 1.18 × 10+03 4.89 × 10+02 5.45 × 10+02 4.29 × 10+02 4.32 × 10+02
F24
STD 2.35 × 10+01 7.55 × 10+01 9.98 × 10+01 6.02 × 10+01 8.53 × 10+01 1.41 × 10+02 7.65 × 10+01 3.24 × 10+01 3.30 × 10+00 4.55 × 10+01
Mean 3.87 × 10+02 3.87 × 10+02 4.11 × 10+02 3.87 × 10+02 7.00 × 10+02 3.69 × 10+03 3.94 × 10+02 5.69 × 10+02 3.87 × 10+02 3.87 × 10+02
F25
STD 4.53 × 10−01 4.18 × 10−02 1.51 × 10+01 3.04 × 10−01 9.42 × 10+01 8.22 × 10+02 1.78 × 10+01 6.01 × 10+01 7.85 × 10−03 1.64 × 10−02
Mean 1.05 × 10+02 1.11 × 10+03 4.20 × 10+03 1.38 × 10+03 5.84 × 10+03 8.19 × 10+03 2.00 × 10+02 4.02 × 10+02 9.44 × 10+02 7.90 × 10+02
F26
STD 1.77 × 10+02 6.28 × 10+02 1.30 × 10+03 1.58 × 10+02 1.16 × 10+03 1.11 × 10+03 9.98 × 10+02 2.41 × 10+01 5.14 × 10+01 2.99 × 10+02
Mean 5.10 × 10+02 5.08 × 10+02 6.01 × 10+02 5.15 × 10+02 7.73 × 10+02 1.49 × 10+03 5.18 × 10+02 2.45 × 10+02 5.04 × 10+02 5.03 × 10+02
F27
STD 6.62 × 10+00 7.32 × 10+00 4.17 × 10+01 9.89 × 10+00 1.32 × 10+02 2.20 × 10+02 1.45 × 10+01 1.74 × 10+03 5.18 × 10+00 5.28 × 10+00
Mean 4.04 × 10+02 3.97 × 10+02 4.67 × 10+02 4.16 × 10+02 1.14 × 10+03 3.72 × 10+03 4.03 × 10+02 5.20 × 10+02 3.25 × 10+02 3.64 × 10+02
F28
STD 1.29 × 10+01 1.06 × 10+01 2.55 × 10+01 1.24 × 10+01 2.81 × 10+02 8.74 × 10+02 2.53 × 10+01 2.62 × 10+01 4.78 × 10+01 4.18 × 10+01
Mean 6.95 × 10+01 8.92 × 10+01 2.55 × 10+02 9.74 × 10+01 6.18 × 10+02 4.22 × 10+03 5.81 × 10+02 4.58 × 10+02 6.42 × 10+00 4.86 × 10+01
F29
STD 2.33 × 10+03 2.61 × 10+03 8.80 × 10+05 8.03 × 10+03 6.04 × 10+07 1.95 × 10+03 2.25 × 10+02 3.57 × 10+01 1.98 × 10+03 2.01 × 10+03
Mean 1.63 × 10+03 2.31 × 10+02 5.40 × 10+05 3.64 × 10+03 5.41 × 10+07 1.41 × 10+09 4.17 × 10+03 6.98 × 10+02 5.13 × 10+01 4.09 × 10+01
F30
STD 4.96 × 10+01 2.58 × 10+02 7.54 × 10+06 5.51 × 10+03 5.38 × 10+09 6.83 × 10+08 1.18 × 10+04 2.54 × 10+02 0.00 × 10+00 3.53 × 10−07
Mathematics 2022, 10, 2211 24 of 38

Table 9. Comparative results of the CEC2017 benchmark test functions in 100 Dim.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 1.63 ×10+04 1.81 ×10+09 1.81 × 10+09 2.36 × 10+10 3.42 × 10+10 2.76 × 10+11 4.39 ×10+04 5.95 ×10+07 0.00 × 10+00 8.98 × 10−01
F1
STD 6.17 × 10+04 1.21 × 10+09 1.21 × 10+09 4.61 × 10+09 8.38 × 10+09 6.07 × 10+09 2.14 × 10+05 1.13 × 10+08 0.00 × 10+00 8.94 × 10+01
Mean 1.21 × 10+05 3.79 × 10+68 3.79 × 10+68 8.54 × 10+174 5.05 × 10+139 1.51 × 10+168 1.09 × 10+79 2.11 × 10+38 2.00 × 10+00 9.15 × 10+03
F2
STD 1.83 × 10+05 2.53 × 10+85 2.53 × 10+85 Inf Inf Inf 6.30 × 10+92 8.23 × 10+42 2.00 × 10+12 3.05 × 10+13
Mean 7.58 × 10+04 2.15 × 10+04 2.15 × 10+04 4.09 × 10+05 7.15 × 10+05 4.88 × 10+05 4.47 × 10+04 6.99 × 10+04 0.00 × 10+00 5.74 × 10+01
F3
STD 1.10 × 10+04 3.98 × 10+03 3.98 × 10+03 2.72 × 10+04 1.07 × 10+05 2.47 × 10+05 1.84 × 10+04 1.10 × 10+04 0.00 × 10+00 4.93 × 10+03
Mean 5.30 × 10+01 6.15 × 10+02 6.15 × 10+02 2.97 × 10+03 6.10 × 10+03 1.10 × 10+05 2.92 × 10+02 2.95 × 10+02 1.82 × 10+02 2.85 × 10+01
F4
STD 2.72 × 10+01 8.41 × 10+01 8.41 × 10+01 8.27 × 10+02 9.31 × 10+02 1.73 × 10+04 3.34 × 10+01 7.73 × 10+01 8.50 × 10+00 4.53 × 10+01
Mean 3.30 × 10+01 2.91 × 10+02 2.91 × 10+02 9.70 × 10+02 1.11 × 10+03 1.66 × 10+03 5.98 × 10+02 2.93 × 10+02 3.43 × 10+01 1.20 × 10+01
F5
STD 2.46 × 10+01 1.44 × 10+02 1.44 × 10+02 3.18 × 10+01 5.83 × 10+01 4.10 × 10+01 8.28 × 10+01 2.48 × 10+01 6.93 × 10+00 3.07 × 10+01
Mean 8.29 × 10−02 1.45 × 10+01 1.45 × 10+01 3.08 × 10+01 8.26 × 10+01 1.16 × 10+02 5.06 × 10+01 3.63 × 10+01 3.02 × 10−05 2.82 × 10−05
F6
STD 4.59 × 10−02 2.10 × 10+00 2.10 × 10+00 2.52 × 10+00 9.03 × 10+00 2.47 × 10+00 4.63 × 10+00 8.06 × 10+00 2.15 × 10−05 8.06 × 10−05
Mean 1.00 × 10+02 8.94 × 10+02 8.94 × 10+02 1.51 × 10+03 2.55 × 10+03 3.30 × 10+03 1.34 × 10+03 6.35 × 10+02 1.47 × 10+02 5.99 × 10+01
F7
STD 3.65 × 10+01 3.67 × 10+01 3.67 × 10+01 1.08 × 10+02 1.22 × 10+02 8.96 × 10+01 2.30 × 10+02 6.39 × 10+01 6.83 × 10+00 5.49 × 10+01
Mean 1.35 × 10+01 2.54 × 10+02 2.54 × 10+02 9.71 × 10+02 1.18 × 10+03 1.78 × 10+03 7.02 × 10+02 2.60 × 10+02 4.51 × 10+01 2.98 × 10+01
F8
STD 2.56 × 10+01 1.53 × 10+02 1.53 × 10+02 3.17 × 10+01 1.17 × 10+02 7.82 × 10+01 6.77 × 10+01 4.59 × 10+01 5.20 × 10+00 1.07 × 10+01
Mean 3.12 × 10−01 1.47 × 10+03 1.47 × 10+03 1.06 × 10+04 5.31 × 10+04 8.68 × 10+04 1.69 × 10+04 9.41 × 10+03 0.00 × 10+00 0.00 × 10+00
F9
STD 7.52 × 10−01 1.17 × 10+03 1.17 × 10+03 2.38 × 10+03 1.34 × 10+04 3.05 × 10+03 2.13 × 10+03 1.87 × 10+03 0.00 × 10+00 0.00 × 10+00
Mean 6.30 × 10+03 2.88 × 10+04 2.88 × 10+04 2.89 × 10+04 2.24 × 10+04 3.11 × 10+04 1.17 × 10+04 5.59 × 10+03 9.74 × 10+03 5.52 × 10+03
F10
STD 1.85 × 10+03 2.72 × 10+02 2.72 × 10+02 6.11 × 10+02 1.66 × 10+03 1.04 × 10+03 1.92 × 10+03 6.58 × 10+02 3.55 × 10+02 7.92 × 10+02
Mean 1.44 × 10+02 1.13 × 10+03 1.13 × 10+03 7.51 × 10+04 9.87 × 10+04 2.72 × 10+05 1.54 × 10+03 2.94 × 10+02 1.80 × 10+01 9.06 × 10+01
F11
STD 7.85 × 10+01 2.17 × 10+02 2.17 × 10+02 1.22 × 10+04 6.82 × 10+04 8.78 × 10+04 2.20 × 10+02 7.15 × 10+02 4.18 × 10+01 9.88 × 10+00
Mean 4.67 × 10+05 8.89 × 10+07 8.89 × 10+07 1.54 × 10+09 5.21 × 10+09 1.75 × 10+11 6.26 × 10+06 7.78 × 10+06 2.95 × 10+03 3.55 × 10+03
F12
STD 6.94 × 10+05 5.89 × 10+07 5.89 × 10+07 4.73 × 10+08 2.54 × 10+09 2.15 × 10+10 1.65 × 10+07 1.29 × 10+07 7.33 × 10+02 4.70 × 10+03
Mean 1.21 × 10+02 1.59 × 10+04 1.59 × 10+04 1.45 × 10+04 6.89 × 10+07 3.62 × 10+10 9.66 × 10+03 2.33 × 10+04 4.23 × 10+02 1.82 × 10+02
F13
STD 2.02 × 10+02 3.52 × 10+03 3.52 × 10+03 3.31 × 10+03 6.00 × 10+07 9.05 × 10+09 8.44 × 10+03 2.30 × 10+04 2.94 × 10+02 3.26 × 10+01
Mean 7.75 × 10+03 3.84 × 10+02 3.84 × 10+02 1.17 × 10+06 3.67 × 10+06 1.52 × 10+08 5.93 × 10+04 1.52 × 10+05 3.72 × 10+01 9.66 × 10+01
F14
STD 6.72 × 10+04 6.21 × 10+01 6.21 × 10+01 2.05 × 10+06 3.24 × 10+06 1.05 × 10+08 8.85 × 10+04 1.21 × 10+06 7.30 × 10+00 8.08 × 10+00
Mean 2.37 × 10+03 6.10 × 10+03 6.10 × 10+03 1.18 × 10+03 2.65 × 10+06 2.19 × 10+10 2.37 × 10+03 1.52 × 10+05 4.15 × 10+01 8.14 × 10+01
F15
STD 2.28 × 10+03 1.31 × 10+03 1.31 × 10+03 1.29 × 10+03 6.90 × 10+06 3.88 × 10+09 2.40 × 10+03 1.21 × 10+06 3.87 × 10+01 8.93 × 10+00
Mathematics 2022, 10, 2211 25 of 38

Table 9. Cont.

PBA-PDM OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
Mean 4.12 × 10+02 2.03 × 10+03 2.03 ×10+03 7.30 ×10+03 1.04 ×10+04 1.71 ×10+04 2.96 ×10+03 6.22 ×10+03 4.39 ×10+02 1.14 × 10+03
F16
STD 4.72 × 10+02 7.94 × 10+02 7.94 × 10+02 3.32 × 10+02 2.56 × 10+03 3.12 × 10+03 7.40 × 10+02 6.53 × 10+03 2.68 × 10+02 2.65 × 10+02
Mean 2.43 × 10+02 1.22 × 10+03 1.22 × 10+03 4.28 × 10+03 6.16 × 10+03 1.90 × 10+06 3.17 × 10+03 1.31 × 10+03 5.74 × 10+02 8.37 × 10+02
F17
STD 3.71 × 10+02 1.25 × 10+03 1.25 × 10+03 2.42 × 10+02 1.26 × 10+03 8.08 × 10+06 5.29 × 10+02 5.53 × 10+02 1.76 × 10+02 1.34 × 10+02
Mean 7.88 × 10+04 8.30 × 10+03 8.30 × 10+03 8.49 × 10+06 2.62 × 10+06 3.90 × 10+08 1.72 × 10+05 1.17 × 10+03 4.85 × 10+01 3.55 × 10+01
F18
STD 6.13 × 10+05 3.39 × 10+03 3.39 × 10+03 7.38 × 10+06 5.16 × 10+06 1.12 × 10+08 1.53 × 10+05 3.73 × 10+02 1.75 × 10+01 1.39 × 10+01
Mean 8.42 × 10+03 5.64 × 10+03 5.64 × 10+03 4.80 × 10+02 2.64 × 10+07 2.14 × 10+10 5.37 × 10+02 1.06 × 10+06 4.17 × 10+01 4.77 × 10+01
F19
STD 3.28 × 10+03 5.83 × 10+03 5.83 × 10+03 1.22 × 10+03 2.53 × 10+07 2.57 × 10+09 4.39 × 10+03 4.50 × 10+06 6.04 × 10+00 5.66 × 10+00
Mean 3.42 × 10+02 4.16 × 10+03 4.16 × 10+03 4.88 × 10+03 3.30 × 10+03 6.19 × 10+03 2.71 × 10+03 1.82 × 10+03 8.67 × 10+02 6.41 × 10+02
F20
STD 2.45 × 10+02 1.58 × 10+02 1.58 × 10+02 1.39 × 10+02 8.20 × 10+02 1.85 × 10+02 4.82 × 10+02 1.35 × 10+04 1.45 × 10+02 1.73 × 10+02
Mean 2.40 × 10+02 5.68 × 10+02 5.68 × 10+02 1.23 × 10+03 1.88 × 10+03 2.70 × 10+03 7.93 × 10+02 9.70 × 10+02 2.66 × 10+02 2.21 × 10+02
F21
STD 3.72 × 10+01 1.75 × 10+02 1.75 × 10+02 2.78 × 10+01 1.80 × 10+02 9.47 × 10+01 1.34 × 10+02 2.95 × 10+02 6.43 × 10+00 1.54 × 10+01
F22 Mean 5.36 × 10+03 3.95 × 10+02 3.95 × 10+02 2.94 × 10+04 2.42 × 10+04 3.35 × 10+04 1.45 × 10+04 4.49 × 10+02 9.55 × 10+03 1.00 × 10+02
r STD 2.00 × 10+03 1.16 × 10+02 1.16 × 10+02 5.91 × 10+02 1.21 × 10+03 5.74 × 10+02 1.30 × 10+03 5.02 × 10+01 5.89 × 10+02 2.84 × 10+03
Mean 4.33 × 10+02 1.46 × 10+03 1.46 × 10+03 1.58 × 10+03 2.48 × 10+03 3.97 × 10+03 1.30 × 10+03 6.41 × 10+03 5.61 × 10+02 4.46 × 10+02
F23
STD 2.53 × 10+01 6.59 × 10+01 6.59 × 10+01 3.78 × 10+01 1.91 × 10+02 3.55 × 10+02 1.29 × 10+02 1.17 × 10+03 1.35 × 10+01 1.07 × 10+02
Mean 5.39 × 10+02 1.39 × 10+03 1.39 × 10+03 2.03 × 10+03 3.32 × 10+03 6.39 × 10+03 1.94 × 10+03 8.51 × 10+02 9.01 × 10+02 5.08 × 10+02
F24
STD 5.56 × 10+01 1.05 × 10+02 1.05 × 10+02 6.36 × 10+01 4.99 × 10+02 9.32 × 10+02 1.41 × 10+02 9.96 × 10+01 1.15 × 10+01 1.30 × 10+02
Mean 5.27 × 10+02 1.15 × 10+03 1.15 × 10+03 3.29 × 10+03 3.18 × 10+03 2.75 × 10+04 7.74 × 10+02 8.58 × 10+02 6.36 × 10+02 4.80 × 10+02
F25
STD 1.69 × 10+01 1.03 × 10+02 1.03 × 10+02 5.28 × 10+02 5.54 × 10+02 2.02 × 10+03 7.04 × 10+01 6.95 × 10+01 4.43 × 10+01 1.64 × 10+01
Mean 1.44 × 10+02 2.09 × 10+03 2.09 × 10+03 1.41 × 10+04 2.77 × 10+04 5.07 × 10+04 1.34 × 10+04 6.58 × 10+02 2.89 × 10+03 4.39 × 10+02
F26
STD 2.74 × 10+02 2.33 × 10+03 2.33 × 10+03 5.79 × 10+02 2.88 × 10+03 4.01 × 10+03 3.87 × 10+03 7.43 × 10+01 1.09 × 10+02 4.22 × 10+02
Mean 5.84 × 10+02 7.73 × 10+02 7.73 × 10+02 1.61 × 10+03 1.92 × 10+03 7.40 × 10+03 8.24 × 10+02 2.21 × 10+03 5.65 × 10+02 5.10 × 10+02
F27
STD 2.68 × 10+01 5.89 × 10+01 5.89 × 10+01 1.53 × 10+02 7.93 × 10+02 1.09 × 10+03 2.05 × 10+02 1.95 × 10+03 1.68 × 10+01 7.14 × 10+00
Mean 4.18 × 10+02 1.04 × 10+03 1.04 × 10+03 4.07 × 10+03 5.39 × 10+03 3.32 × 10+04 6.35 × 10+02 8.38 × 10+02 4.77 × 10+02 4.59 × 10+02
F28
STD 2.32 × 10+01 1.21 × 10+02 1.21 × 10+02 6.23 × 10+02 1.05 × 10+03 3.25 × 10+03 3.20 × 10+01 1.22 × 10+02 1.50 × 10+01 1.99 × 10+01
Mean 3.97 × 10+02 3.20 × 10+03 3.20 × 10+03 2.06 × 10+03 1.05 × 10+04 2.21 × 10+05 4.43 × 10+03 7.41 × 10+02 8.81 × 10+02 6.15 × 10+02
F29
STD 1.84 × 10+02 9.83 × 10+02 9.83 × 10+02 5.28 × 10+02 1.76 × 10+03 8.07 × 10+05 5.84 × 10+02 1.59 × 10+02 1.45 × 10+02 1.27 × 10+02
Mean 7.94 × 10+05 5.99 × 10+05 5.99 × 10+05 2.86 × 10+03 5.84 × 10+08 2.62 × 10+10 3.35 × 10+04 1.59 × 10+03 2.12 × 10+03 5.80 × 10+05
F30
STD 8.47 × 10+04 5.73 × 10+05 5.73 × 10+05 1.89 × 10+02 4.64 × 10+08 6.60 × 10+09 2.32 × 10+05 3.65 × 10+02 2.25 × 10+02 4.52 × 10+04
Mathematics 2022, 10, 2211 26 of 38

It can be seen from Table 7 that when the dimension of the test function problem
is 10, PBA-PDM can optimize 14 test functions to the optimal value, including F3, F6,
F8, F10, F11, F13, F16, F17, F20, F21, F22, F25, and F28. Note that PBA-PDM, L-SHADE,
and EBOwithCMAR reached the same optimal value in solving F22. As mentioned in
Section 4.1, L-SHADE is the winner of “CEC2014 Special Session and Competition”, and
EBOwithCMAR is the winner of “CEC2017 Special Session and Competition”. For F1 and
F7, PBA-PDM obtained the second best result compared to EBOwithCMAR. For F2, F4, F5,
F9, F12, F14, F15, F19, F23, F26, and F30, both L-SHADE and EBOwithCMAR showed better
performances than PBA-PDM, but the differences between the best solutions are all within
two orders of magnitude, which indicates that PBA-PDM is still very competitive with
other competitors. For F26, F27, and F29, INFO obtained a better result than PBA-PDM;
however, the error between these two results is negligible when compared to the order of
magnitude of the results obtained by other algorithms.
The performance of PBA-PDM, and the other comparison algorithms, decreased when
the dimensionality of the test problem increased. Table 8 shows the average and standard
deviation of the optimization results for each algorithm at 30-D. It can be seen that at this
point, PBA-PDM only achieved the best results in 9 out of 30 tests, including F4, F5, F16,
F17, F20, F21, F22, F25, and F26. The overall performance of PBA-PDM at this point is
slightly lower than that of L-SHADE and EBOwithCMAR. It can be seen from Table 9 that
the drop in performance is also pronounced when the problem dimension rises to 100,
with PBA-PDM achieving the best results in 9 out of 30 tests, including F8, F13, F16, F17,
F20, F23, F26, F28, and F29. This is mainly because L-SHADE and EBOwithCMAR were
developed specifically for the CEC benchmark test problems, and the superior performance
of such algorithms on the test problems comes at the cost of applicability and the ease
of use. L-SHADE and EBOwithCMAR require a large number of complex parameters to
be set, and when they are applied to other optimization problems (especially real-world
optimization problems), the proper setting of parameters is a huge challenge.
In conclusion, PBA-PDM proposed in this paper shows a superior performance in
solving low-dimensional optimization problems. When the dimensionality of the test
function increases significantly, PBA-PDM is still able to achieve a satisfactory solution
within acceptable limits. Compared to other well-established optimization algorithms, PBA-
PDM is more robust and more widely applicable. Compared to optimization algorithms
developed specifically for numerical calculations, PBA-PDM requires fewer parameters to
be set and is easier to understand and apply to real-world optimization problems.

4.5. Wilcoxon Signed-Rank Tests

In the previous section, we performed a complete analysis and discussion of the
algorithm’s performance using the statistical results of the classical test functions and the
CEC benchmark function. The null hypothesis H0 for this test is: The median number of
optimal solutions obtained after 30 independent tests of the same optimization problem
in the same dimension by Appendices A and B is the same. The results of the PBA-
PDM algorithm compared to competitors are reported in Tables 10–13, with a statistical
significance level α = 0.05. In these tables, ‘+’ indicates that the null hypothesis is rejected
and PBA-PDM has statistically superior performances and ‘−’ indicates that PBA-PDM has
inferior performances. Meanwhile, ‘=’ indicates there is no statistical difference between
PBA-PDM and competitors. From the results, it can be concluded that the proposed
PBA-PDM performs significantly better than most comparison algorithms in classical test
functions. When solving the CEC benchmark functions, the performance of PBA-PDM
decreases progressively with the increases in dimensionality. Compared to the optimization
algorithms developed specifically for CEC benchmark problems, PBA-PDM’s optimization
results are not as good as L-SHADE and EBOwithCMAR, but it is still competitive with
other well-established algorithms.
Mathematics 2022, 10, 2211 27 of 38

Table 10. Wilcoxon signed rank test results (p-values) of classical test problem analysis.

Function OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
F1 0.000(+) 0.000(+) 0.013(+) 0.035(−) 0.308(=) 0.215(=) 0.000(+) 0.001(+) 0.000(+)
F2 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+)
F3 0.000(+) 0.002(+) 0.027(+) 0.003(+) 0.073(=) 0.029(+) 0.045(+) 0.000(+) 0.000(+)
F4 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
F5 0.000(+) 0.529(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.039(+) 0.041(+)
F6 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F7 0.000(+) 0.005(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.085(+) 0.016(+)
F8 0.303(+) 0.000(+) 0.119(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
F9 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F10 0.280(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.415(+) 0.000(+)
F11 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.610(=)
F12 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F13 1.000(=) 0.902(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F14 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F15 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F16 0.049(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
F17 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=) 1.000(=)
F18 0.000(−) 0.371(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
F19 0.136(=) 0.001(+) 0.154(=) 0.410(=) 0.311(=) 0.716(=) 0.732(=) 0.239(=) 0.442(=)
F20 1.000(=) 1.000(=) 1.000(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
+/=/− 10/9/1 10/10/0 11/9/0 11/8/1 9/10/1 11/9/0 12/8/0 11/8/1 11/9/0

Table 11. Wilcoxon signed rank test results (p-values) of 10-D CEC2017 problem analysis.

Function OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
CEC01 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC02 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC03 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.057(+)
CEC04 0.000(+) 0.000(+) 0.000(+) 0.000(=) 0.000(=) 0.000(=) 0.000(=) 0.002(+) 0.602(=)
CEC05 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.037(+) 0.000(+) 0.000(+) 0.651(=)
CEC06 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.011(+) 0.000(+)
CEC07 0.000(+) 0.029(+) 0.000(+) 0.000(+) 0.000(+) 0.010(+) 0.000(+) 0.159(+) 0.036(−)
CEC08 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.209(−)
CEC09 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC10 0.000(+) 0.061(+) 0.000(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.281(+) 0.116(+)
CEC11 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.715(=) 0.544(=) 0.317(+)
CEC12 0.001(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.005(−) 0.003(−)
CEC13 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.002(−) 0.001(−)
CEC14 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.022(−)
CEC15 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(−) 0.000(−)
CEC16 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.002(+) 0.000(+) 0.000(+) 0.000(+)
CEC17 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC18 0.0411(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC19 0.055(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.036(−) 0.000(−)
CEC20 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC21 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(+) 0.000(+)
CEC22 0.000(+) 0.003(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 1.000(=) 1.000(=)
CEC23 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.733(=) 0.806(=)
CEC24 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.012(+)
CEC25 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC26 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.003(−) 0.001(−)
CEC27 0.000(+) 0.102(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.004(−) 0.437(=)
CEC28 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC29 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(−) 0.000(+) 0.030(+) 0.005(+)
CEC30 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−)
+/=/− 28/1/1 30/0/0 29/1/0 29/1/0 29/1/0 26/1/3 27/2/1 19/3/8 12/5/13
Mathematics 2022, 10, 2211 28 of 38

Table 12. Wilcoxon signed rank test results (p-values) of 30-D CEC2017 problem analysis.

Function OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
CEC01 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC02 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC03 0.000(+) 0.000(−) 0.000(−) 0.001(+) 0.001(+) 0.001(−) 0.001(+) 0.000(−) 0.000(−)
CEC04 0.034(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(−) 0.000(+) 0.005(−) 0.301(=)
CEC05 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.779(=) 0.515(=)
CEC06 0.001(−) 0.000(+) 0.663(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC07 0.000(+) 0.000(+) 0.000(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC08 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.628(=) 0.000(−)
CEC09 0.000(+) 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC10 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+)
CEC11 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+)
CEC12 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(−) 0.000(−)
CEC13 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+)
CEC14 0.000(+) 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(−) 0.000(−)
CEC15 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+)
CEC16 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.015(+) 0.000(+)
CEC17 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(+) 0.000(+)
CEC18 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.033(−) 0.001(−)
CEC19 0.751(=) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC20 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC21 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.277(=) 0.000(−)
CEC22 1.000(=) 0.000(+) 1.000(=) 0.000(+) 0.000(+) 1.000(=) 0.000(+) 1.000(=) 1.000(=)
CEC23 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.007(+) 0.000(+)
CEC24 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(−) 0.003(−)
CEC25 1.000(=) 0.000(+) 1.000(=) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 1.000(=) 1.000(=)
CEC26 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC27 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−)
CEC28 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC29 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(=)
CEC30 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
+/=/− 22/3/5 27/0/3 25/4/1 30/0/0 30/0/0 24/1/5 28/0/2 6/5/19 8/5/17

Table 13. Wilcoxon signed rank test results (p-values) of 100-D CEC2017 problem analysis.

Function OPGPBA ECLPSO CCPSO IRGA HDN INFO FAEICA L-SHADE EBOwithCMAR
CEC01 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC02 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC03 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−) 0.000(−)
CEC04 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC05 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.004(+) 0.000(−)
CEC06 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC07 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC08 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC09 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC10 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(−)
CEC11 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC12 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC13 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC14 0.000(+) 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC15 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(+) 0.015(+) 0.000(+) 0.000(−) 0.000(−)
CEC16 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC17 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC18 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−)
CEC19 0.000(−) 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(−) 0.000(−)
CEC20 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC21 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−)
CEC22 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(+) 0.000(−)
CEC23 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC24 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.001(−)
CEC25 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC26 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC27 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(−) 0.000(−)
CEC28 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC29 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC30 0.000(−) 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−) 0.000(−)
+/=/− 26/0/4 25/0/5 26/0/4 30/0/0 30/0/0 27/0/3 25/0/5 16/0/14 9/0/21
 CEC28 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC29 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+) 0.000(+)
CEC30 0.000(−) 0.000(−) 0.000(−) 0.000(+) 0.000(+) 0.000(−) 0.000(−) 0.000(−) 0.000(−)
+/=/− 26/0/4 25/0/5 26/0/4 30/0/0 30/0/0 27/0/3 25/0/5 16/0/14 9/0/21
Mathematics 2022, 10, 2211 29 of 38

5. Application of PBA-PDM to a Real-World Constrained Engineering Design Prob-

lem
5. Application of PBA-PDM
The utilization to a Real-World
of metaheuristics Constrained
algorithms to solve Engineering Designdesign
real-world optimal Problem
prob-
lemsThe
is autilization
classical, of metaheuristics
well-applied algorithms
research to solve
direction. real-world
In this section, optimal design prob-
six constrained engi-
lems is a classical, well-applied research direction. In this section, six constrained
neering design problems according to CEC2020 were selected to evaluate the effectiveness engineer-
ing design problems
of PBA-PDM. according
The results to CEC2020
of PBA-PDM arewere selected
compared to to evaluate
other the effectiveness
well-established of
state-of-
PBA-PDM. The results
the-art algorithms of PBA-PDM
proposed are compared
in previous work. to other well-established state-of-the-art
algorithms proposed in previous work.
5.1. Pressure Vessel Design
5.1. Pressure Vessel Design
The objective intention in this case, as Figure 10 indicates, is to minimize the fabrica-
The objective intention in this case, as Figure 10 indicates, is to minimize the fabrication
tion cost and it has four parameters and constraints. The variables of this case are: the
cost and it has four parameters and constraints. The variables of this case are: the thickness
thickness of the shell ( ), thickness of the head ( ), inner radius ( ), and length of the
of the shell (Th ), thickness of the head (Th ), inner radius (r), and length of the section
section without the head ( ).
without the head (L).

Figure 10. Schematic view of the pressure vessel design.

The pressure vessel design problem has a nonlinear fitness function with three linear
and one nonlinear inequality constraints as follows:
→
Consider z = [z1 , z2 , z3 , z4 ] = [ Ts , Th , r, L]
→
Minimize f z = 0.6224z1 z3 z4 + 1.7781z2 z23 + 3.1661z21 z4 + 19.84z21 z3
Subject
  to:
→
g1 z = −z1 + 0.0193z3 ≤ 0,
→
g2 z = −z3 + 0.00954z3 ≤ 0,
→
g3 z = − ∏ z23 z4 − 0.75 ∏ z33 + 1, 296, 000 ≤ 0,
→
g4 z = z4 − 240 ≤ 0,
Variable range:
1 ≤ z1 , z2 ≤ 99, 10 ≤ z3 , z4 ≤ 200,
The results of PBA-PDM are compared to OPGPBA, L-SHADE, EBOwithCMAR,
ECLPSO, CCPSO, GA [30], BA [31], MFO [32], WOA [33], CSS [34], ESs [35], GWO [36],
BIANCA [37], MDDE [38], and DELC [39]. Table 14 shows the optimum designs obtained
by the listed optimizers over 30 independent runs. It can be concluded that the search
ability of L-SHADE and EBOwithCMAR is the best and basically the same, followed by
PBA-PDM. As the third best optimizer in solving pressure vessel design problems, PBA-
PDM can obtain a competitive solution compared to most other listed algorithms such as
ECLPSO, GWO, etc.

Table 14. The comparison of PBA-PDM with other optimizers in Pressure Vessel Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR ECLPSO CCPSO GA BA

x1 0.778168 0.781329 0.875000 0.812500 0.817583 0.812500 0.865281 0.874962
x2 0.384648 0.393356 0.430000 0.437500 0.407292 0.437500 0.444743 0.432673
x3 40.3196 40.4197 43.9844 41.6623 42.0917 42.0912 43.6098 45.3349
x4 200.000 200.000 154.654 182.123 176.719 176.746 158.784 140.284
f (x) 5785.33 5950.04 5751.48 5726.18 6000.46 6061.08 6257.38 6073.19
STD 3.15 × 10+01 7.63 × 10+02 2.16 × 10+01 1.59 × 10+01 6.68 × 10+02 3.91 × 10+01 3.85 × 10+02 2.05 × 10+02
 Table 14. The comparison of PBA-PDM with other optimizers in Pressure Vessel Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR ECLPSO CCPSO GA BA

0.778168 0.781329 0.875000 0.812500 0.817583 0.812500 0.865281 0.874962
0.384648 0.393356 0.430000 0.437500 0.407292 0.437500 0.444743 0.432673
40.3196
Mathematics 2022, 10, 2211 40.4197 43.9844 41.6623 42.0917 42.0912 43.6098 45.3349
30 of 38
200.000 200.000 154.654 182.123 176.719 176.746 158.784 140.284
f(x) 5785.33 5950.04 5751.48 5726.18 6000.46 6061.08 6257.38 6073.19
STD 3.15 × 10+01 7.63Table × 10+02 2.16 × 10+01
14. Cont. 1.59 × 10+01 6.68 × 10+02 3.91 × 10+01 3.85 × 10+02 2.05 × 10+02
MFO WOA CSS ESs GWO BIANCA MDDE DELC
PBA-PDM OPGPBA L-SHADE EBOwithCMAR ECLPSO CCPSO GA BA
0.802486 0.846125 0.820145 0.810922 0.827568 0.914706 0.812500 0.812523
MFO WOA CSS ESs GWO BIANCA MDDE DELC
0.387435 0.424562 0.405557 0.400839 0.408104 0.452139 0.437500 0.437534
x1 0.802486
x2
40.3410 0.846125
0.387435
43.5473 0.820145
0.424562
42.4944
0.405557
0.810922
42.0167
0.400839
0.827568
42.6308 0.914706
0.408104
47.3941 0.812500
0.452139
42.0984
0.437500
0.812523
42.0984
0.437534
x3 199.832 43.5473
40.3410 159.478 42.4944 171.812 177.651
42.0167 170.714 47.3941
42.6308 120.464 176.636
42.0984 176.636
42.0984
x4f(x) 199.832
6070.41 159.478
6069.01 171.812
5962.07 177.651
5943.70 170.714
6016.81 120.464
6162.07 176.636
6059.70 176.636
6059.71
f (x) 6070.41 6069.01 5962.07 5943.70 6016.81 6162.07 6059.70 6059.71
STDSTD 7.757.75× 10 ×
+0110 +01 1.79
1.79 × 10 ×
+0110 +01 1.92
1.92 × 10 × 10
+02 +02 8.62
8.62 × 10 ×
+0110 +01 4.47
4.47 × 10 ×
+0110 +01 1.59 ×+02
1.59 × 10 10 +02 7.68 × +02
7.68 × 10 10 +02 4.13 × 10
4.13 × 10+01
+01

5.2. Welded Beam Design

5.2. Welded Beam Design
In this case, as Figure 11 indicates, according to the design constraints, the objective
In this case, as Figure 11 indicates, according to the design constraints, the objective of
of this
this problem
problem is toisdiscover
to discover the minimum
the minimum manufacturing
manufacturing cost.cost. The design
The design variables
variables for
for this
problem are the thickness of weld (h), length (l), height (t), thickness of the bar (b), and two),
this problem are the thickness of weld (ℎ), length ( ), height ( ), thickness of the bar (
and two
linear andlinear and five nonlinear
five nonlinear inequalityinequality constraints.
constraints.

Figure 11. Schematic view of the welded beam design.

The constraints of this design problem can be formulated as follows:

→
Consider z = [z1 , z2 , z3, z4 ] = [h, l, t, b]
→
Minimize f z = 1.10471z21 z2 + 0.04811z3 z4 (14.0 + z2 )
Subject
→ to: →
g1 z = τ z − τmax ≤ 0,
→ →
g2 z = σ z − σmax ≤ 0,
→ →
g3 z = δ z − δmax ≤ 0,
→
g4 z = z1 − z4 ≤ 0,
→ →
g5 z = P − Pc z ≤ 0,
→
g6 z = 0.125 − z1 ≤ 0,
→
g7 z = 1.10471z21 + 0.04811z3 z4 (14.0 + z2 ) − 5.0 ≤ 0,
Variable range:
0.125 ≤ z1 ≤ 5, 0.1 ≤ z2 ≤ 10, 0.1 ≤ z3 ≤ 10, 0.125 ≤ z3 ≤ 5,
where:
→ r z2 P MR  z2 
τ z = τ0 + 2τ0 τ00 + τ00 2 , τ0 = √ , τ00 = , M = P L+ ,
2R 2z1 z2 J 2
s ( "  #)  
z22 z1 + z3 2 √ z22 z1 + z3 2
  
→ 6PL
R= + , J=2 2z1 z2 + , σ z = ,
4 2 12 2 z4 z23
q
4.013E z23 z64 /36
r !
→ 4PL3 → z3 E
δ z = , Pc z = 1− ,
Ez33 z4 L2 2L 4G

P = 600 lb, L = 14 in, E = 30 × 106 psi, G = 12 × 106 psi.

Mathematics 2022, 10, 2211 31 of 38

The optimal results of PBA-PDM compared to OPGPBA, L-SHADE, EBOwithCMAR,

RANDOM [40], DAVID [40], SIMPLEX [40], APPROX [40], GSA [41], ESs, CDE [42],
CPSO [43], BA, and HS [44] are represented in Table 15. It can be demonstrated that
the PBA-PDM and EBOwithCMAR can both obtain the minimum solution with the most
ideal design settings.

Table 15. The comparison of PBA-PDM with other optimizers in Welded Beam Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR RANDOM DAVID SIMPLEX

x1 0.205734 0.201249 0.1872 0.205729 0.457500 0.243400 0.279200
x2 3.47042 3.46235 4.048 3.47048 4.73130 6.25520 5.62560
x3 9.03649 9.03650 8.9957 9.03662 5.08530 8.29150 7.75120
f (x) 0.205735 0.206319 0.2099 0.205739 0.660000 0.244400 0.279600
STD 6.18 × 10−08 2.14 × 10−03 5.17 × 10−12 6.81 × 10−09 1.03 × 10−10 6.64 × 10−02 9.31 × 10−02
APPROX GSA ESs CDE CPSO BA HS
x1 0.240000 0.182129 0.199742 0.203137 0.202369 0.215002 0.244200
x2 6.21890 3.85697 3.61206 3.54299 3.54421 7.42904 6.22310
x3 8.29150 10.0000 9.03750 9.03349 9.04821 7.16371 8.29150
f (x) 0.244400 0.208506 0.206082 0.206179 0.205768 0.331240 0.244300
STD 1.33 × 10−02 6.43 × 10−06 8.11 × 10−02 7.57 × 10−03 5.13 × 10−07 2.81 × 10−05 2.66 × 10−03

5.3. Speed Reducer Design

Mathematics 2022, 10, x FOR PEER REVIEW 28 of 35
The objective in this case is to minimize the total mass of the speed reducer while
satisfying all the constraints. The speed reducer design problem is indicated in Figure 12.

Figure12.
Figure 12. Schematic
Schematic view
view of
of the
the Speed
SpeedReducer
Reducerdesign.
design.

The design
The design variables
variables for
for this
this problem
problem are are as as follows:
follows: thethe face
face width
width (b),
( ), module
module of of
teeth (m),
teeth ( ),number
numberof ofteeth
teethon
onthe
the pinion
pinion (z),( ),length
length of of shaft
shaft 11 between
between bearings
bearings (l( 1 ),), length
length
of shaft
of shaft 22 between
between bearings
bearings(l( 2 ),),diameter
diameterof ofshaft
shaft11(d ( 1 ),),and
anddiameter
diameterofofshaft
shaft22(d( 2 ).).
The constraints of this design problem can be
The constraints of this design problem can be formulated as follows: formulated as follows:
→
Consider z⃗ =[z1, z2 , z3,, z4,, z5,, z6,, z7]] =
Consider = [[b,, m,, z,, l1,, l2,, d1,, d2]]
→
Minimize f z ( ⃗)== 075854 075854bm (3.3333
2 3.3333z2 + + 14.9334 − 43.0934
43.0934)− 1.508b (d21 +
− 1.508 + d22 ) +
+



Minimize 14.9334z −
7.477( 3 + 3 )
+ 0.7854( +
2 )2

7.477 d1 + d2 + 0.7854 l1 d1 + l2 d2
Subject to:
Subject to:
g ( →
⃗)= 27 − 1 ≤ 0,
g1 z = bm. 2 z − 1 ≤ 0,
g (→⃗)= − 1 ≤ 0,
g2 z = . 397.5 2 − 1 ≤ 0,
g (→⃗)= bm z− 1 ≤ 0,
1.93l13
g3 z = . mzd4 − 1 ≤ 0,
g ( ⃗)= 1− 1 ≤ 0,
→ 1.93l 3
g4 z = mzd42 − 1 ≤ 0,
g ( ⃗) = 1 q (745 ⁄ ) + 16.9 × 10 − 1 ≤ 0,
→
1 2
z == 110d
gg5 ( ⃗) (745l
3 (745 ⁄1 /mz + 16.9××10
) )+ 157.5 106−−11 ≤≤0,0,
1q
→
gg6 ( ⃗)
z == 85d −
1 1 ≤ 0,
3 (745l1 /mz)2 + 157.5 × 106 − 1 ≤ 0,
2
g ( ⃗) = − 1 ≤ 0,
g ( ⃗) = − 1 ≤ 0,
. .
g ( ⃗) = − 1 ≤ 0,
. .
Mathematics 2022, 10, 2211 32 of 38

→
g7 z = mz 40 − 1 ≤ 0,
→
5m
g8 z = b − 1 ≤ 0,
→
b
g9 z = 12m − 1 ≤ 0,
→
g10 z = 1.5d1l +1.9 − 1 ≤ 0,
→ 1

g11 z = 1.1d2l +1.9 − 1 ≤ 0,

2
Variable range:
2.6 ≤ z1 ≤ 3.6, 0.7 ≤ z2 ≤ 0.8, z3 ∈ {17, 28}(integer ),
7.3 ≤ z4 , z5 ≤ 8.3, 2.9 ≤ z6 ≤ 3.9, 5 ≤ z7 ≤ 5.5,
As shown in Table 16, the optimal results obtained by the proposed PBA-PDM outper-
form the competitors, including OPGPBA, L-SHADE, EBOwithCMAR, ECLPSO, SCA [45]
GWO, and MGWO variants [46], with the best fitness value of 2973.15.

Table 16. The comparison of PBA-PDM with other optimizers in Speed Reducer Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR ECLPSO SCA GWO MGWO-III

x1 3.51633 3.50850 3.50000 3.50159 3.49810 3.50875 3.50056 3.50010
x2 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7
x3 17 17 17 17 17 17 17 17
x4 7.30474 7.39284 7.30000 7.30000 7.63980 7.30000 7.31911 7.30056
x5 8.05193 7.81603 7.71532 7.81545 7.80000 7.80000 7.80414 7.80668
x6 3.19129 3.35807 3.35021 3.35127 3.35820 3.46102 3.35047 3.35068
x7
Mathematics 5.29221
2022, 5.28677
10, x FOR PEER REVIEW 5.28665 5.28874 5.28530 5.28923 5.28669 5.28690
29 of 35
f (x) 2973.15 3002.92 2994.47 2998.55 3000.00 3030.56 2996.90 2996.79
STD 4.36 × 10−12 3.55 × 10−05 0.00 × 10+00 0.00 × 10+00 8.53 × 10−04 2.11 × 10+00 1.43 × 10−15 7.21 × 10−01

5.4.
5.4. Multiple
Multiple Disk
Disk Clutch
Clutch Brake
Brake Design
Design
The
The objective
objective in
in this
this case,
case, which
which is
is shown
shown in
in Figure
Figure 13, is to
13, is to minimize
minimize the
the multiple
multiple
dick
dick clutch
clutch brake’s
brake’soverall
overallmass.
mass.The
Thedesign
designvariables
variablesforfor
this
thisproblem areare
problem thethe
number of
number
friction surfaces , actuating force , thickness of the disk , outer radius
of friction surfaces z, actuating force F, thickness of the disk t, outer radius Ro , and inner
radius Ri ..

Figure 13. Schematic view of the Multiple Disk Clutch Brake design.

The constraints of this design problem can be formulated as follows:

follows:
→
Consider z⃗ ==→[[z1,, z2,, z3,, z4, , z5]]== [ Ri , Ro , t, F, z]
Consider
[ , , , , ]
Minimize f ( z⃗) == π ( R2o−− R)2i (t( Z −− 1)1×)× 0.0000078
0.0000078
Subject
Subject to:
→ to:
( ⃗) = − − ≥ 0,
g1 z = Ro − Ri − r ≥ 0,
( ⃗) = − ( + 1)( + 0.5) ≥ 0,
→

(
g2 z = lmax−− ( Z≥+0,1)(t + 0.5) ≥ 0,
⃗) =
( ⃗) = ( ) − ( ) ≥ 0,
( ⃗) = − ≥ 0,
( ⃗) = − ≥ 0,
( ⃗) = − ( ) ≥ 0,
( ⃗) = ≥ 0,
Variable range:
Mathematics 2022, 10, 2211 33 of 38

→
g3 z = Pmax − Prz ≥ 0,
→
g4 z = Pmax (vsr max ) − Prz (vrz ) ≥ 0,
→
g5 z = vsr max − vrz ≥ 0,
→
g6 z = Tmax − T ≥ 0,
→
g7 z = Mh − s( Ms ) ≥ 0,
→
g8 z = T ≥ 0,
Variable range:
60 ≤ z1 ≤ 80, 90 ≤ z2 ≤ 110, 1 ≤ z3 ≤ 3, 0 ≤ z4 ≤ 1000, 2 ≤ z5 ≤ 9,
The optimal results obtained by PBA-PDM and PBA, OPGPBA, L-SHADE, EBOwith-
CMAR, WCA [47], APSO [48], and IAPSO [49] are shown in Table 17. It can be concluded
that the solution obtained by PBA-PDM and EBOwithCMAR is 0.124972556, which is the
lowest result among all competitors.

Table 17. The comparison of PBA-PDM with other optimizers in Multiple Disk Clutch Brake Design.

PBA-PDM PBA OPGPBA L-SHADE EBOwithCMAR WCA APSO IAPSO L-SHADE

x1 80 70 73 70 80 70 76 70 70
x2 90 90 93 90 90 90 96 90 90
x3 1 1 1 1 1 1 1 1 1
x4 600 810 990 810 700 910 840 900 810
x5 2 3 2 3 2 3 3 3 3
f (x) 0.124972 0.313657 0.282541 0.313657 0.124972 0.313657 0.337181 0.313657 0.313657
STD 3.07 × 10−02 7.95 × 10−02 8.08 × 10−02 7.35 × 10−02 1.24 × 10−02 6.57 × 10−02 3.79 × 10−02 5.04 × 10−02 3.27 × 10−03

5.5. Tension/Compression Spring Design

The objective intention in this case, as Figure 14 indicates, is to minimize the weight of
Mathematics 2022, 10, x FOR PEER REVIEW 30 of 35
the tension/compression spring. The design variables for this engineering design problem
are the wire diameter (d), mean coil diameter (D), and the number of active coils (N).

Figure 14.Schematic
Figure14. Schematicview
viewof
ofthe
theTension/Compression
Tension/Compression Spring
Springdesign.
design.

The T/C spring design problem has a nonlinear fitness function with four nonlinear
The T/C spring design problem has a nonlinear fitness function with four nonlinear
inequality constraints as follows:
inequality→ constraints as follows:
Consider z = [z1 , z2 , z3 ] = [d, D, N ]
Consider ⃗ = →[ , , ] = [ , , ]
Minimize f (z = ((z3 + 2)z2 z21
Minimize ⃗) = + 2)
Subject
Subject to:
→ to:
z32 z3
g1 (z⃗) ==11−− 71785Z 4≤≤0,0,
→ 1
2
4Z2 −z1 z2
g2 (z⃗) == + 1 ≤ 0,
12566(z2 z31 −z41 )+ 5108z21≤ 0,
→
.
g3 (z⃗) ==11−− 140.5zz22 z3
1
≤≤0,0,
→
z + z
g4 (z⃗) == 1.1.5 − 2 1 ≤ 0,
− 1 ≤ 0,
Variablerange:
Variable range:
0.05≤≤z1 ≤≤2.00,
0.05 2.00, 0.25 0.25 ≤ ≤ z2 ≤
≤ 1.30,
1.30, 2.00
2.00 ≤ z3 ≤
≤ 15.00,
15.00,
Many optimizers have been utilized to tackle this problem, including OPGPBA, L-
SHADE, EBOwithCMAR, TEO [50], MFO, SFS [51], GWO, WOA, GSA, CPSO, DEDS [52],
DELC, HEAA [53], WEO [54], and BA. Table 18 shows the best solution and the values of
the constraints obtained by PBA-PDM and other competitors over 30 independent runs.

Table 18. The comparison of PBA-PDM with other optimizers in Tension/Compression Spring De-
 Mathematics 2022, 10, 2211 34 of 38

Many optimizers have been utilized to tackle this problem, including OPGPBA, L-
SHADE, EBOwithCMAR, TEO [50], MFO, SFS [51], GWO, WOA, GSA, CPSO, DEDS [52],
DELC, HEAA [53], WEO [54], and BA. Table 18 shows the best solution and the values of
the constraints obtained by PBA-PDM and other competitors over 30 independent runs.

Table 18. The comparison of PBA-PDM with other optimizers in Tension/Compression Spring Design.

PBA-PDM OPGPBA L-SHADE TEO MFO SFS GWO EBOwithCMAR

x1 0.051318 0.051207 0.051480 0.0517750 0.0519944 0.0516890 0.0516900 0.051686
x2 0.347868 0.345215 0.351661 0.358791 0.364109 0.356717 0.356737 0.356637
x3 11.8272 12.0040 12.0000 11.1683 10.8684 11.28896 11.2888 11.0000
f (x) 0.0126637 0.0126763 0.010248 0.0126650 0.0126669 0.0126652 0.0126660 0.012665
STD 2.70 × 10−07 4.06 × 10−07 3.79 × 10−08 1.41 × 10−06 5.26 × 10−07 8.98 × 10−07 6.73 × 10−08 2.35 × 10−07
WOA GSA CPSO DELC HEAA WEO BA DEDS
x1 0.0512070 0.0502760 0.0517280 0.0516890 0.0516890 0.0516850 0.0516900 0.0516890
x2 0.345215 0.323680 0.357644 0.356717 0.356729 0.356630 0.356730 0.356717
x3 12.0000 13.5254 11.2445 11.2889 11.2882 11.2941 11.2885 11.2889
f (x) 0.0126763 0.0127022 0.0126740 0.0126650 0.0126650 0.0126650 0.0126650 0.0126650
STD 4.59 × 10−06 5.61 × 10−07 5.44 × 10−07 3.34 × 10−08 2.04 × 10−05 1.20 × 10−06 7.58 × 10−06 7.32 × 10−06

From Table 18, we can draw the conclusion that L-SHADE performs the best among
the 16 algorithms, and the result of PBA-PDM is the second best, which is very competitive
and effective with high-quality solutions in tackling this benchmark problem and it exposes
an ideal design solution.

atics 2022, 10, x FOR PEER REVIEW

5.6. Cantilever Beam Design 31 of 35
The objective in this case is to calculate the minimal weight of a cantilever beam. The
cantilever beam consists of five hollow blocks as shown in Figure 15.

Figure 15. Schematic view

Figure 15. of the Cantilever
Schematic Beam
view of the Design.
Cantilever Beam Design.

The constraints
Theofconstraints
this design of problem
this design can be formulated
problem as follows:as follows:
can be formulated
→
[ , , z, =
Consider ⃗ = Consider , z2[, z,3, z4], z5 ] = [ X, N ]
, [z]1=
→
Minimize ( ⃗) = 0.0624(
Minimize f z = + 0.0624+ (z+1 + z+ 2 + z)3 + z4 + z5 )
Subject to: Subject to:
→  
( ⃗) = +g + z = + 61+ 37 − 19
+ + + + ≤7 0, 1
− τmax ≤ 0,
1 z31 z32 z33 z34 z35
Variable range:Variable range:
0.01 ≤ ≤ 100, 0.01 ≤ = z1,≤. . .100,
,5 i = 1, · · · , 5
i
The optimal The results obtained
optimal results by byPBA-PDM
obtained PBA-PDM and and OPGPBA,
OPGPBA, L-SHADE,L-SHADE,
EBOwithCMAR,
EBOwithCMAR, SOS SOS
[55], [55],
MMA [56], GCA_I
MMA [56], GCA_II
[56], GCA_I [56], and
[56], GCA_II [56],CS [57]CS
and are shown
[57] in Table
are shown in19. It can
be be
Table 19. It can concluded
concluded thatthat
the solution obtained
the solution usingusing
obtained PBA-PDM
PBA-PDMis veryiscompetitive and is better
very competi-
thanthan
tive and is better all theallother well-established
the other comparison
well-established algorithms.
comparison algorithms.

Table 19. The comparison of PBA-PDM with other optimizers in Cantilever Beam Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR SOS MMA GCA_II CS

6.01468 6.01513 6.01452 6.01509 6.01878 6.01350 6.01510 6.01490
5.30464 5.30930 5.30928 5.30854 5.30544 5.30570 5.30630 5.30791
4.49659 4.49500 4.50126 4.49660 4.49587 4.49390 4.49750 4.50230
3.50356 3.50142 3.49862 3.50159 3.49896 3.49000 3.50400 3.50770
Mathematics 2022, 10, 2211 35 of 38

Table 19. The comparison of PBA-PDM with other optimizers in Cantilever Beam Design.

PBA-PDM OPGPBA L-SHADE EBOwithCMAR SOS MMA GCA_II CS

x1 6.01468 6.01513 6.01452 6.01509 6.01878 6.01350 6.01510 6.01490
x2 5.30464 5.30930 5.30928 5.30854 5.30544 5.30570 5.30630 5.30791
x3 4.49659 4.49500 4.50126 4.49660 4.49587 4.49390 4.49750 4.50230
x4 3.50356 3.50142 3.49862 3.50159 3.49896 3.49000 3.50400 3.50770
x5 2.15017 2.15270 2.15196 2.15067 2.15564 2.15000 2.15000 2.15043
f (x) 1.33993 1.33997 1.34010 1.33997 1.33994 1.34039 1.33996 1.34065
STD 3.05 × 10−04 2.14 × 10−04 7.04 × 10−04 6.57 × 10−06 8.90 × 10−03 4.27 × 10−03 7.36 × 10−02 5.71 × 10−03

6. Summary
Low efficiency and premature convergence are the main shortcomings of metaheuristic
optimization algorithms in tackling constrained engineering design problems. To address
this tricky problem, an improved pity beetle algorithm is proposed in this paper, named the
pity beetle algorithm based on pheromone dispersion model (PBA-PDM). The PBA-PDM
is inspired by the collective behavior of beetles, which models the interactions between
different broods.
To verify the superiority of PBA-PDM in dealing with different optimization problems,
several experiments were conducted. Firstly, the results of solving the classical test functions,
including the UM and MM types, prove the effectiveness of PBA-PDM in terms of convergence
and avoiding stagnation. Secondly, the results of solving the CEC2017 benchmark functions
demonstrate that PBA-PDM can achieve satisfying results close to the winners of the CEC
competitions, and there is no structural bias in the mathematical models and search strategies.
The above experimental results show that PBA-PDM exhibits a good convergence performance
compared with its counterparts and other state-of-the-art algorithms for both unimodal
and multimodal test problems, with high accuracy in low-dimensional problems and a fast
convergence speed in high-dimensional problems. Furthermore, PBA-PDM shows strong
applicability and robustness in the optimization of real-world constrained engineering design
problems. The experiment results show that the algorithm has more accurate parameter
designs than other widely used optimization algorithms.
In this paper, the convergence of PBA-PDM is primarily investigated through a large
number of experiments. In future work, the theoretical convergence analysis of PBA-PDM
will be an interesting topic to pursue. According to the results of the high-dimensional test
experiments, there is still much room for improvement in PBA-PDM, and attention can
be paid to improving the search efficiency and convergence speed of the algorithm in the
high-dimensional objective space. Further, given that large-scale optimization problems
often involve multiple objectives, the single-objective PBA-PDM can be extended to a multi-
objective optimization algorithm. Then, the multi-objective PBA-PDM could be applied
to solve more complex optimization problems, such as multi-objective optimization of
wastewater treatment processes, the flying sidekick traveling salesman problem, and global
smooth path planning for mobile robots.

Author Contributions: X.D.: conceptualization, software, writing—review and editing, supervision.

Y.P.: writing—original draft, software, data curation, visualization. All authors have read and agreed
to the published version of the manuscript.
Funding: The National Natural Science Foundation of China (No. 61563032), the Natural Science
Foundation of Gansu Province (No. 1506RJZA104), The University Innovation Fund Project of Gansu
Province (No. 2021A-027).
Conflicts of Interest: The authors declare no conflict of interest.
Mathematics 2022, 10, 2211 36 of 38

Appendix A. Unimodal Test Functions

No. Function D Range Optimum

  2
F1 f ( x ) = ∑id=1 ∑dj=1 ( j + β) xij − 1 100 [−3, 3] 0
ji
d 2
F2 f ( x ) = ∑ i =1 x i 50 [−5, 5] 0
f 1 ( x ) = x12 + 2x22 − 0.3 cos(3πx1 ) − 0.4 cos(4πx2 ) + 0.7
F3 2 [−5, 5] 0
f 2 ( x ) = x12 + 2x22 − 0.3 cos(3πx1 ) cos(4πx2 ) + 0.3
f 3 ( x ) = x12 + 2x22 − 0.3 cos(3πx1 + 4πx2 ) + 0.3
F4 f ( x ) = ∑id=1 ∑ij=1 x2j 100 [−65, 65] 0
F5 f ( x ) = ∑id=1 ixi2 100 [−10, 10] 0
−d(d + 4)
F6 f ( x ) = ∑i=1 ( xi − 1)2 − ∑id=2 xi xi−1
d 100 [−4, 4]
(d − 1)/6
2 2
F7 f ( x ) = ( x1 + 2x2 − 7) +
(2x1 + x2 − 5) 2 [−10, 10] 0
F8 f ( x ) = 0.26 x12 + x22 − 0.48x1 x2 2 [−10, 10] 0
x1 ∈[−1.5, 4]
F9 f ( x ) = sin( x1 + x2 ) + ( x1 − x2 )2 − 1.5x1 + 2.5x2 + 1 2 −1.9133
x2 ∈[−3, 4]

2
F10 f ( x ) = ( x1 − 1)2 + ∑id=2 i 2xi2 − xi−1 100 [−10, 10] 0
F11 x6 2 [−5, 5] 0
( x ) = 2x12 − 1.05x
f 4 1 2
 1 + 6 + x1 x2 + x2
F12 x14 2 [−2, 2] −1.0316
f ( x ) = 4 − 2.1x1 + 3 x1 + x1 x2 + −4 + 4x22 x22
2 2

Appendix B. Multimodal Test Functions

No. Function D Range Optimum

F13 sin2 ( x2 − x22 )−0.5 2 [−10, 10] 0

f ( x ) = 0.5 − 1
[1+0.001( x12 +x 2 2
2 )]
ix2

F14 f ( x ) = −∑id=1 sin( xi ) sin2m πi 100 [0, π] −1.8013
d [−5, 5]
∑ i =1 xi2
 
F15 f ( x ) = 10d + − 10 cos(2πxi ) 100 0
f (x) =
F16
2 2 [−5, 5] 0
(1.5 − xh1 + x1 x2 ) + (2.25 − x1 + x1 x2 )2 + 2.625 − x1 + x1 x23
2

2
i
F17 f ( x ) = 1 + ( x1 + x2 + 1) 19 − 14x1 + 3x1 − 14x2 + 6x1 x2 + 3x22
2
2 [−2, 2] 3
h
i
× 30 + (2x1 − 3x2 )2 18 − 32x1 + 12x12 + 48x2 − 366x1 x2 + 27x22
F18
q
2 x∈[−5, 5] 0
f ( x ) = 100 x2 + 0.01x12 + 0.01| x1 + 10| y∈[−3, 3]
f ( x )=
F19 q
x
p  2 [−600, 600] −959.6407
−( x2 + 47) sin x2 + 21 + 47 − x1 sin | x1 − ( x2 + 47)|
x2
 
F20 f ( x ) = ∑id=1 4000
i − d x
∏i=1 cos √ii + 1 100 [−10, 10] 0

References
1. Jiang, J.H.; Liu, Y.T.; Zhao, Z.Y. TriTSA: Triple Tree-Seed Algorithm for dimensional continuous optimization and constrained
engineering problems. Eng. Appl. Artif. Intell. 2021, 104, 104303. [CrossRef]
2. Wang, C.; Koh, J.M.; Yu, T.; Xie, N.G.; Cheong, K.H. Material and shape optimization of bi-directional functionally graded plates
by GIGA and an improved multi-objective particle swarm optimization algorithm. Comput. Methods Appl. Mech. Eng. 2020, 366,
113017. [CrossRef]
3. Shadravan, S.; Naji, H.R.; Bardsiri, V.K. The sailfish optimizer: A novel nature-inspired metaheuristic algorithm for solving
constrained engineering optimization problems. Eng. Appl. Artif. Intell. 2019, 80, 20–34. [CrossRef]
4. Shahrzad, S.; Mirjalili, S.; Lewis, A. Grasshopper optimisation algorithm: Theory and application. Adv. Eng. Softw. 2017, 105,
30–47.
5. Eberhart, R.; Kennedy, J. A new optimizer using particle swarm theory. In Proceedings of the Mhs95 Sixth International
Symposium on Micro Machine & Human Science, Nagoya, Japan, 4–6 October 1995; IEEE: New York, NY, USA, 2002.
6. Lam, A.Y.; Li, V.O. Chemical reaction optimization: A tutorial. Memetic Comput. 2012, 4, 3–17. [CrossRef]
7. Dorigo, M.; Birattari, M.; Stutzle, T. Ant colony optimization. IEEE Comput. Intell. Mag. 2006, 1, 28–39. [CrossRef]
8. Hussain, K.; Salleh, M.N.M.; Cheng, S.; Shi, Y. Metaheuristic research: A comprehensive survey. Artif. Intell. Rev. 2019, 52,
2191–2233. [CrossRef]
9. Mahdavi, S.; Shiri, M.E.; Rahnamayan, S. Metaheuristics in large-scale global continues optimization: A survey. Inf. Sci. 2015, 295,
407–428. [CrossRef]
10. Boussaïd, I.; Lepagnot, J.; Siarry, P. A survey on optimization metaheuristics. Inf. Sci. 2013, 237, 82–117. [CrossRef]
11. Kallioras, N.A.; Lagaros, N.D.; Avtzis, D.N. Pity Beetle Algorithm—A new metaheuristic inspired by the behaviour of bark
beetles. Adv. Eng. Softw. 2018, 121, 147–166. [CrossRef]
Mathematics 2022, 10, 2211 37 of 38

12. Schwerdtfeger, F. Ein Beitrag zur Fortpflanzungsbiologie des Borkenkäfers Pityogenes chalcographus L. J. Appl. Entomol. 2010, 15,
335–427. [CrossRef]
13. Su-Ran, W.U.; Xiong, G.R.; Cai, W.W.; Wang, J.G.; Zeng, J. Advances in the Studies on Semiochemicals in Cerambycid Beetles.
Sugarcane Canesugar 2018, 2, 47–51.
14. Rassati, D.; Marchioro, M.; Flaherty, L.; Poloni, R.; Edwards, S.; Faccoli, M.; Sweeney, J. Response of native and exotic longhorn
beetles to common pheromone components provides partial support for the pheromone-free space hypothesis. Insect Sci. 2021,
28, 793–810. [CrossRef]
15. Turner, D. Bruce. A diffusion model for an urban area. J. Appl. Meteorol. Climatol. 1964, 3, 83–91. [CrossRef]
16. Stockie, J.M. The mathematics of atmospheric dispersion modeling. Siam Rev. 2011, 53, 349–372. [CrossRef]
17. Fauziah, B.A.; Hidaka, T.; Tabata, K. The reproductive behavior of Monochamus alternatus HOPE:(Coleoptera: Cerambycidae).
Appl. Entomol. Zool. 1987, 22, 272–285. [CrossRef]
18. Wang, Y.J.; Peng, S. One-Way Pioneer Guide Pity Beetle Algorithm: A New Evolutionary Algorithm for Solving Global Optimiza-
tion Problems. IEEE Access 2020, 8, 203270–203293. [CrossRef]
19. Yu, X.; Zhang, X. Enhanced comprehensive learning particle swarm optimization. Appl. Math. Comput. 2014, 242, 265–276.
[CrossRef]
20. Ying, W.; Zhou, J.; Lu, Y.; Qin, H.; Zhang, Y. Chaos Cultural Particle Swarm Optimization and Its Application. In International
Symposium on Neural Networks; Springer: Berlin/Heidelberg, Germany, 2009; pp. 30–40.
21. Wang, Z.L.; Yang, P.; Ling, D.; Miao, Q. An Improved Real-Coded Genetic Algorithm and Its Application. J. Electron. Sci. Technol.
2008, 6, 43–46.
22. Singh, G.; Singh, U. Triple band-notched UWB antenna design using a novel hybrid optimization technique based on DE and
NMR algorithms. Expert Syst. Appl. 2021, 184, 115299. [CrossRef]
23. Almalaq, A.; Alqunun, K.; Refaat, M.M.; Farah, A.; Benabdallah, F.; Ali, Z.M.; Aleem, S.H.A. Towards Increasing Hosting Capacity
of Modern Power Systems through Generation and Transmission Expansion Planning. Sustainability 2022, 14, 2998. [CrossRef]
24. Keivanian, F.; Chiong, R. A novel hybrid fuzzy–metaheuristic approach for multimodal single and multi-objective optimization
problems. Expert Syst. Appl. 2022, 195, 116199. [CrossRef]
25. Tanabe, R.; Fukunaga, A. Success-history based parameter adaptation for differential evolution. In Proceedings of the 2013 IEEE
Congress on Evolutionary Computation, Cancun, Mexico, 20–23 June 2013; IEEE: New York, NY, USA, 2013; pp. 71–78.
26. Kumar, A. Improving the local search capability of effective butterfly optimizer using covariance matrix adapted retreat phase. In
Proceedings of the 2017 IEEE Congress on Evolutionary Computation (CEC), San Sebastián, Spain, 5–8 June 2017.
27. Liang, J.-J.; Suganthan, P.N.; Deb, K. Novel composition test functions for numerical global optimization. In Proceedings of the
2005 IEEE Swarm Intelligence Symposium, Pasadena, CA, USA, 8–10 June 2005; IEEE: New York, NY, USA, 2005; pp. 68–75.
28. Caraffini, F.; Kononova, A.V.; Corne, D. Infeasibility and structural bias in differential evolution. Inf. Sci. 2019, 496, 161–179.
[CrossRef]
29. Kononova, A.V.; Corne, D.W.; de Wilde, P.; Shneer, V.; Caraffini, F. Structural bias in population-based algorithms. Inf. Sci. 2015,
298, 468–490. [CrossRef]
30. Deb, K. Optimal design of a welded beam via genetic algorithms. AIAA J. 1991, 29, 2013–2015. [CrossRef]
31. Gandomi, A.H.; Yang, X.-S.; Alavi, A.H.; Talatahari, S. Bat algorithm for constrained optimization tasks. Neural Comput. Appl.
2013, 22, 1239–1255. [CrossRef]
32. Mirjalili, S. Moth-flame optimization algorithm: A novel nature-inspired heuristic paradigm. Knowl. Based Syst. 2015, 89, 228–249.
[CrossRef]
33. Mirjalili, S.; Lewis, A. The whale optimization algorithm. Adv. Eng. Softw. 2016, 95, 51–67. [CrossRef]
34. Kaveh, A.; Talatahari, S. A novel heuristic optimization method: Charged system search. Acta Mech. 2010, 213, 267–289. [CrossRef]
35. Mezura-Montes, E.; Coello, C.A.C. A simple multimembered evolution strategy to solve constrained optimization problems.
IEEE Trans. Evol. Comput. 2005, 9, 1–17. [CrossRef]
36. Mirjalili, S.; Mirjalili, S.M.; Lewis, A. Grey wolf optimizer. Adv. Eng. Softw. 2014, 69, 46–61. [CrossRef]
37. Montemurro, M.; Vincenti, A.; Vannucci, P. The automatic dynamic penalisation method (adp) for handling constraints with
genetic algorithms. Comput. Methods Appl. Mech. Eng. 2013, 256, 70–87. [CrossRef]
38. Mezura-Montes, E.; Coello, C.C.; Velázquez-Reyes, J.; Muñoz-Dávila, L. Multiple trial vectors in differential evolution for
engineering design. Eng. Optim. 2007, 39, 567–589. [CrossRef]
39. Wang, L.; Li, L.-P. An effective differential evolution with level comparison for constrained engineering design. Struct. Multidiscip.
Optim. 2010, 41, 947–963. [CrossRef]
40. Rao, R. Jaya: A simple and new optimization algorithm for solving constrained and unconstrained optimization problems. Int. J.
Ind. Eng. Comput. 2016, 7, 19–34.
41. Rashedi, E.; Nezamabadi-Pour, H.; Saryazdi, S. Gsa: A gravitational search algorithm. Inf. Sci. 2009, 179, 2232–2248. [CrossRef]
42. Huang, F.-Z.; Wang, L.; He, Q. An effective co-evolutionary differential evolution for constrained optimization. Appl. Math.
Comput. 2007, 186, 340–356. [CrossRef]
43. He, Q.; Wang, L. An effective co-evolutionary particle swarm optimization for constrained engineering design problems. Eng.
Appl. Artif. Intell. 2007, 20, 89–99. [CrossRef]
Mathematics 2022, 10, 2211 38 of 38

44. Lee, K.S.; Geem, Z.W. A new structural optimization method based on the harmony search algorithm. Comput. Struct. 2004, 82,
781–798. [CrossRef]
45. Mirjalili, S. Sca: A sine cosine algorithm for solving optimization problems. Knowl. Based Syst. 2016, 96, 120–133. [CrossRef]
46. Kumar, V.; Kumar, D. An astrophysics-inspired grey wolf algorithm for numerical optimization and its application to engineering
design problems. Adv. Eng. Softw. 2017, 112, 231–254. [CrossRef]
47. Eskandar, H.; Sadollah, A.; Bahreininejad, A.; Hamdi, M. Water cycle algorithm–A novel metaheuristic optimization method for
solving constrained engineering optimization problems. Comput. Struct. 2012, 110, 151–166. [CrossRef]
48. Yang, X.S.; Deb, S.; Fong, S. Accelerated particle swarm optimization and support vector machine for business optimization and
applications. In International Conference on Networked Digital Technologies; Springer: Berlin/Heidelberg, Germany, 2011; pp. 53–66.
49. Guedria, N.B. Improved accelerated pso algorithm for mechanical engineering optimization problems. Appl. Soft Comput. 2016,
40, 455–467. [CrossRef]
50. Kaveh, A.; Dadras, A. A novel meta-heuristic optimization algorithm: Thermal exchange optimization. Adv. Eng. Softw. 2017, 110,
69–84. [CrossRef]
51. Salimi, H. Stochastic fractal search: A powerful metaheuristic algorithm. Knowl. Based Syst. 2015, 75, 1–18. [CrossRef]
52. Zhang, M.; Luo, W.; Wang, X. Differential evolution with dynamic stochastic selection for constrained optimization. Inf. Sci. 2008,
178, 3043–3074. [CrossRef]
53. Wang, Y.; Cai, Z.; Zhou, Y.; Fan, Z. Constrained optimization based on hybrid evolutionary algorithm and adaptive constraint-
handling technique. Struct. Multidiscip. Optim. 2009, 37, 395–413. [CrossRef]
54. Kaveh, A.; Bakhshpoori, T. Water evaporation optimization: A novel physically inspired optimization algorithm. Comput. Struct.
2016, 167, 69–85. [CrossRef]
55. Cheng, M.-Y.; Prayogo, D. Symbiotic organisms search: A new metaheuristic optimization algorithm. Comput. Struct. 2014, 139,
98–112. [CrossRef]
56. Chickermane, H.; Gea, H.C. Structural optimization using a new local approximation method. Int. J. Numer. Methods Eng. 1996,
39, 829–846. [CrossRef]
57. Gandomi, A.H.; Yang, X.-S.; Alavi, A.H. Cuckoo search algorithm: A metaheuristic approach to solve structural optimization
problems. Eng. Comput. 2013, 29, 17–35. [CrossRef]