Journal of Bionic Engineering (2022) 19:1177–1202

https://doi.org/10.1007/s42235-022-00185-1

RESEARCH ARTICLE

An Improved Tunicate Swarm Algorithm with Best‑random Mutation

Strategy for Global Optimization Problems
Farhad Soleimanian Gharehchopogh1 

Received: 5 November 2021 / Revised: 25 February 2022 / Accepted: 28 February 2022 / Published online: 28 March 2022
© Jilin University 2022

Abstract
The Tunicate Swarm Algorithm (TSA) inspires by simulating the lives of Tunicates at sea and how food is obtained. This
algorithm is easily entrapped to local optimization despite the simplicity and optimal, leading to early convergence compared
to some metaheuristic algorithms. This paper sought to improve this algorithm's performance using mutating operators such
as the lévy mutation operator, the Cauchy mutation operator, and the Gaussian mutation operator for global optimization
problems. Thus, we introduced a version of this algorithm called the QLGCTSA algorithm. Each of these operators has a
different performance, increasing the QLGCTSA algorithm performance at a specific optimization operation stage. This algo-
rithm has been run on benchmark functions, including three different compositions, unimodal (UM), and multimodal (MM)
groups and its performance evaluate six large-scale engineering problems. Experimental results show that the QLGCTSA
algorithm had outperformed other competing optimization algorithms.

Keywords  Tunicate swarm algorithm (TSA) · Mutation strategy · Global optimization

1 Introduction of exploitation and exploration. Each successful metaheuris-

Metaheuristic algorithms are increasingly used to solve
several search problems and perform continuous and dis-
crete optimizations. Metaheuristic algorithms have become
a suitable solution for continuous optimization problems,
as constant optimization problems have become increas-
ingly complicated, and mathematical procedures have been
unable to provide optimal solutions [1–6]. Because of the
inability of precise optimization procedures to solve com-
plex high-dimensional and multi-state problems, approxima-
tion algorithms propose new solving problem approaches.
Approximate algorithms divide into two key categories,
heuristic and metaheuristic algorithms. Heuristic algorithms
were less focused since their local entrapment and used to
solve specific optimization problems. Metaheuristic algo-
rithms provide acceptable solutions at a reasonable time
while not guaranteeing optimal solutions. Searching and
producing optimal solutions for an optimization problem in
metaheuristic algorithms is based on two essential concepts
* Farhad Soleimanian Gharehchopogh
Bonab.farhad@gmail.com
1
Department of Computer Engineering, Urmia Branch,
Islamic Azad University, Urmia, Iran
tic algorithm seeks to create a balance between exploita-
tion and exploration [7–15]. A generation of metaheuristic
algorithms uses exploration to produce new solutions while
gradually reducing the exploration as the generation comes
to an end; as the exploration process decreases, exploitation
gradually increases.
That is why metaheuristic algorithms propose more than
other algorithms to solve optimization problems for the fol-
lowing four main reasons [16–19]. First, metaheuristic algo-
rithms have been inspired by simple issues in nature, result-
ing in simplicity. Simplicity supports computer scientists
in introducing hybrid metaheuristic algorithms and using
them efficiently to solve several problems; in the meantime,
simplicity makes it possible to learn these algorithms fast.
Second, most metaheuristic algorithms do not need deriva-
tives. These algorithms create random solutions for an opti-
mization problem, thereby removing the need to calculate
a search space subsidiary to find more optimal solutions.
Third, these algorithms' flexibility allows them to solve sev-
eral optimization problems without changing an algorithm
structure. The algorithms' inputs and their objective function
regulates by focusing on the optimization problem. Finally,
metaheuristic algorithms digression local optimization com-
pared to standard models.

13
Vol.:(0123456789)
 F. S. Gharehchopogh
1178
Inspired by simulating the life of Tunicates at sea and
how food provided by Satnam, Kaur et al. proposed the TSA
algorithm [20]. Moreover, it is seen as one of the most recent
metaheuristic algorithms to solve engineering optimization
problems. Tunicate can search for a food source, but they
have no idea about it. This algorithm is easily entrapped to
local optimization despite the simplicity and optimal, lead-
ing to early convergence compared to some metaheuristic
algorithms. This paper sought to improve this algorithm's
performance using mutation operators such as Lévy, Cauchy,
and Gaussian. These mutation operators use Lévy, Cauchy,
and Gaussian mutation distributions. Each of these distribu-
tions has different performances, increasing the LGCTSA
algorithm's performance at a particular stage of the opti-
mization operation. Conducted experiments suggest that
simultaneously using three mutation operators increas-
ingly improves the performance of the LGCTSA algorithm.
Gaussian mutation creates small jumps in solutions. On the
other hand, the Cauchy mutation creates more extensive
mutations than the Gaussian mutation, thus leading to the
increased overall ability of search agents to search globally
in each generation.
There are different classes of metaheuristic algorithms
is proposed in the literature. However, apart from different
classes, one can observe that several of these algorithms are
based on collective behavior and how animals hunt in nature
[21]. Our goal is to examine some nature-based metaheuris-
tics and basic algorithms to solve this subpart's optimization
problem. A Spider Monkey Optimization (SMO) algorithm
[22] proposes numerical optimization using feeding spider
monkey modeling. They have division and combination-
based social structures. Such animals move from more lead-
ing groups to smaller ones based on a deficiency of food
availability, and vice versa—the metaheuristic algorithm of
coexistence search planners on animals' interaction in nature
[23]. The algorithm has regarded three stages of co-assis-
tance, food-sharing, and parasitic life for the animals such
that two animals may benefit or harm each other. Prey-preda-
tor metaheuristic algorithm[24] randomly allocates solutions
produced as a huntsman and the prey relevant on their func-
tion in the target function. A new metaheuristic optimization
algorithm inspired by the movement of galaxies, stars, and
massive galaxies under the influence of gravity, called the
Galactic Swarm Optimization(GSO) algorithm, was pro-
posed [25] in 2015. It applies several exploration and exploi-
tation stages for establishing the best balance rate between
the exploration of new solutions and the exploitation of new
solutions. Also, a metaheuristic algorithm provided by Mir-
jalili et al. (2014) is based on the hierarchical behavior of
wolves [26] named Grey Wolf Optimizer (GWO). The three
best solutions are alpha, beta, and delta wolves, consecutive,
with the rest of the solutions considered regular wolves [27].
13
A population-based metaheuristic algorithm famous for
the Sine Cosine Algorithm (SCA) solves optimization prob-
lems [28]. In this algorithm, as a random primary candidate,
several solutions have been produced by cosine and sine
functions, as they move in a sinusoidal/co-sinusoidal way
(up and down) towards the best solution. Several random
and comparative variables have been incorporated in this
algorithm to emphasize the exploration and exploitation of
the search space at different optimization milestones. Simu-
lation results confirm that the SCA algorithm effectively
reveals different search space areas, preventing local opti-
mization, convergence to all-out global optimization, and
using promising areas of a search space during optimization.
The Spotted Hyena Optimization (SHO) algorithm [29] is
a metaheuristic algorithm with a collective approach based
on the collective hunting behavior of spotted hyenas. In this
algorithm, each solution to the problem codes as a hyena.
The hyenas approach the optimal answer or the prey by fol-
lowing the optimal answer or the group-leading hyena. The
three primary SHO stages are searching for the prey, laying
the siege, and invading the prey, all three of which are math-
ematically modeled and executed.
A population-based metaheuristic optimization algorithm
is inferred from Harris hawk's collaborative behavior and
hunting style in nature using the surprise touring model
known as the Harris Hawks Optimization (HHO) algorithm
[18]. In this algorithm, several Harris hawks collaborate and
invade from different angles to surprise prey. Harris Hawk
can produce different chase patterns based on dynamic sce-
narios' nature and prey's escape patterns. This technique
computationally imitates emotional patterns and behaviors
to create an optimization algorithm.
We only investigated several the metaheuristic algo-
rithms such as: Starling Murmuration Optimizer (SMO)
[30], Simulated Annealing (SA) algorithm [31], Biogeog-
raphy-Based Optimization (BBO) [32], Clonal Selection
Algorithm (CSA) [33], Gorilla Troops Optimizer (GTO)
[34], Conscious Neighborhood-based Crow Search Algo-
rithm (CCSA) [35], Farmland Fertility Algorithm (FFA)
[36], Quantum-based Avian Navigation Optimizer Algo-
rithm[37], Gravitational Search Algorithm (GSA) [38], Afri-
can Vultures Optimization Algorithm(AVOA)[39], League
Championship Algorithm (LCA)[40], Henry Gas Solubil-
ity Optimization(HGSO)[41], Group Counseling Optimizer
(GCO)[42], Intelligent Water Drops (IWD) algorithm[43],
and etc.
In this paper, we attained a new solution to solve engi-
neering problems. We improved the TAS algorithm; we
begin to describe the LGCTSA algorithm in detail after
the TSA algorithm was introduced in Sect. 3. Moreover,
the Lévy mutation is used to strengthen more profound
search capacities. Metaheuristic algorithms use mutation
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1179

procedures to produce new solutions. The following most ��⃗

��⃗ = G (1)
contributed to the paper, which is as follows: A
���⃗
M
• Using Quantum Rotation Gate (QRG) to improve perfor- In Eq. (1), the vector G,
��⃗ which is the force of gravity, is
mance. obtained based on Eq. (2).
• Using the Gaussian, Cauchy, and Lévy mutation as a
mutation operator and investigating how this operator ��⃗ = c2 + c3 − F
G ��⃗ (2)
performs in the TSA algorithm.
In Eq. (2), the vector F ��⃗ The water flow vector in the
• Using the Gaussian, Cauchy, and Lévy mutation in two
ocean's depth is obtained based on Eq. (3).
forms and investigating how these operators perform
when used two-by-two. ��⃗ = 2.c1
F (3)
• Using the Gaussian mutation, Cauchy mutation, and
Lévy mutation to modify the TSA algorithm's perfor- In Eqs. (2–3), the variables c1, c2, and c3 are random val-
mance and evaluate the effects of these operators when ues betwixt 0 and 1. Finally, the vector M ���⃗ represents the
incorporated in a TSA algorithm. social force between the search agents, which calculates
• Investigating the LGCTSA algorithm using 52 standard using Eq. (4).
benchmarks functions and also examining the algorithm's
performance using six engineering problems of large ���⃗ = ⌊Pmin + c1 .Pmax − Pmin ⌋
M (4)
dimensions.
• An experiment for the scalability of the LGCTSA algo- In Eq. (4), ­Pmax and P­ min represent the primary and sub-
rithm using standard benchmarks distribution of large ordinate velocities to create social interaction. In the TSA
dimensions. algorithm, ­Pmax and P ­ min values consider 1 and 4, respec-
tively. After the search agent's new position, calculate the
We organized this paper as follows: Sect. 2 introduces search agent's move towards the best neighbor shown in
the TSA algorithm; Sect. 3 investigates the LGCTSA algo- Eq. (5).
rithm and provides mutation operators; Sect. 4 evaluates �����⃗ = ||FS ���⃗p (x)||
����⃗ − rand().P
PD
| | (5)
the LGCTSA algorithm, or the standard benchmarks func-
tion along with six engineering optimization problems, and �����⃗ demonstrates the distance betwixt the food
In Eq. (5), PD
Sect. 5 offers the conclusion and future works of this paper.
source and the search agent. x is the current repeat value and
����⃗ is the optimal location of the food source while P
FS ���⃗p (x)
indicates the tunicate position. In the next stage of the TSA
2 Tunicate Swarm Algorithm algorithm, the search agents converge towards the best
search agent, where it can maintain its position relative to
Tunicate Swarm Algorithm (TSA) was constructed to simu-
the best search agent, being the food source. This stage of
late the Tunicate life at sea and how Tunicate gets food[20].
the algorithm performs in Eq. (6).
Tunicates can search for food sources at sea while having no
{
idea for a food source in a given search space. In the TSA ����⃗ + A.��⃗ PD,
�����⃗ if rand ≥ 0.5
���
⃗ FS
algorithm, two Tunicate behaviors known as jet propulsion Pp (x) =
����⃗ ��⃗ �����⃗ if rand < 0.5 (6)
FS − A.PD,
and swarm intelligence uses for optimization operations.
To algorithm jet pulsing behavior mathematically, Tuni-
In Eq. (6), P���⃗p (x) is the updated position of the Tunicate
cate must observe three main conditions: moving towards
based on the feed source position FS ����⃗ . Finally, in the math-
the best search agent's position, avoiding conflicts between
ematical modeling stage of Swarm behavior, two of the best
search agents, and residual close to the best search agent.
optimal solutions are maintained, and the position of other
While in congestion behavior, the position of other search
search agents is updated based on these two solutions. Equa-
agents is selected based on the best solution. These behav-
tion (7) is applied to algorithm this behavior.
iors are explained with their mathematical algorithm below.
( )
To prevent a conflict between search agents (in other words, ( ) P ���⃗p (x) + Pp x�⃗ + 1
other tunicates), vector A ��⃗ is used to calculate the new posi- Pp x�⃗ + 1 = (7)
2 + c1
tion of the search agent (Eq. (1)).
The TSA pseudo-algorithm shows in Algorithm (1).

13
 F. S. Gharehchopogh
1180

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1181

3 QLGCTSA Algorithm
The TSA algorithm is entrapped in the local optimization,
leading to early convergence. For this, we have improved
the TSA algorithm's performance using mutation opera-
tors. These mutation operators have been designed based
on three distributions: Lévy, Cauchy, and Gaussian. Each
of these distributions has different functions, increasing the
QLGCTSA algorithm's performance at a given stage in the
optimization operation. On the other hand, experiments con-
ducted illustrated that three mutation operators' simultane-
ous application increases the performance of the QLGCTSA
algorithm. This algorithm and how the optimization opera-
tion process conductors are present in Fig. 1.
3.1 Lévy Mutation ‑TSA (QLTSA)
mutation operators. The Lévy flight operator uses to grow
the capability of search agents. It helps all search agents find
better positions and increase the ability to search deeper dur-
ing optimization operations. Therefore, the LTSA algorithm
has more extraordinary ability and efficiency in an all-out
search. The Lévy mutation describes in Eq. (9).
Levy(𝛽) ∼ u = t−1−𝛽 , 0 < 𝛽 ≤ 2 (9)
In Eq. (9), β is a stability index, and Lévy flight calculates
as Eq. (10).
u×𝜎
Levy(𝛽) ∼ 1 (10)
|v| 𝛽
In Eq. (10), u and v are standard normal distributions. σ
was obtained using Eq. (11).
1

Like other metaheuristic algorithms, the TSA algorithm ⎡ (Γ(1 + 𝛽) × sin( 𝜋𝛽 ) ⎤ 𝛽

moves to optimal local locations. This paper uses differ- 𝜎=⎢ 2 ⎥
(11)
⎢ Γ( 1+𝛽 ) × 𝛽 × 2( 𝛽−1 )⎥
ent mutation strategies to move to new and different search ⎣ 2
2
⎦
agents' positions. Different random mutation operators have
In Eq. (11), Γ is a standard gamma function, and β is a
incorporated the LTSA algorithm. That then search agent is
constant value of 1.5. An integrated version with a post-
applied to them during the iterations of the mutation opera-
swarm behavior phase Lévy flight is based on Algorithm (2)
tion, as expressed in Eq. (8).
using Eq. (12), representing the Lévy flight.
�
xi = xi × (1 + 𝛿); i ∈ {1, … , n} (8) �
xi = xi × (1 + L(𝛽)) (12)
In Eq. (8), xi represent the current and xi represent the
′

mutated position of a search agent, and δ indicates one of the

No
Yes
Generate the initial Yes
Start t≤ T j ≤ Pop
tunicate population
No

Calculate the fitness of updated Detetmine the jet propulsion and swarm
search agents behaviors of tunicate using eq.(7)

Best-random Mutation Phase

Obtain the new positions Xnewg,
Obtain fitness fg, fl and fc
Xnewl and Xnewc using Eqs(20-22)

Save Xnew as the new Yes Dose Xnew has lower No

Save X(i,j)
posotin of X(i,j) fitness than X(i,j)

Return the best Operate QRG

Finish
optimal solution

Fig. 1  Flowchart of QLGCTSA algorithm

13
 F. S. Gharehchopogh
1182

Table 1  The selection strategy of QRG

( ) In Eq. (12), L (β) is a random value in the Lévy flight. It
𝛿𝛉𝐢 𝐬 𝛂𝐣 , 𝛃𝐣 αj βj > 0 αj βj < 0 αj = 0 βj = 0 can produce different solutions because of its heavy-tailed
( ) distribution. Thus, the Lévy flight increases the ability of
f xi == best_f itness δ 0 0 0 0
( ) all search agents to escape optimal local locations. The way
f xi > best_f itness  + 1 –1  ± 1 0
( )
–1  + 1 0  ± 1
the Lévy flight is incorporated illustrates in Algorithm (2).
f xi < best_f itness

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization…
3.2 Cauchy Mutation ‑TSA (QCTSA)
In this subsection, the Cauchy mutation is applied to the
CTSA algorithm. The Cauchy mutation shows using
Eq. (13).
( )
1 1 𝛾
y = + arctan (13)
2 𝜋 g 3.4 Quantum Rotation Gate (QRG)
The corresponding density function shows using Eq. (14).
1 g
fCauchy(0,g) (𝛾) =
𝜋 g2 + 𝛾 2
In Eq. (14), g is a ratio parameter with a value of 1. y is
a random value betwixt 0 and 1, and γ = tan(𝜋(y − 1∕2)) .
The Cauchy mutation congestion function is related to the
Gaussian mutation. Still, there are also differences in the
Cauchy mutation relative to the Gaussian mutation in that
the Cauchy mutation is it is small and is in a vertical direc-
tion. On the other hand, the Cauchy mutation is horizon-
tally wider than the Gaussian mutation. It is conceivable
to modify the capability to search in search agents or add
neighbors of each generation using the Cauchy mutation.
On the other hand, this leads to increased reliability of
search agents to improve the solutions found on a large scale
while quickly escaping from the optimal locations. To this
end, the Cauchy mutation is used as a mutation operator.
Equation (15), represents the Cauchy mutation.
�
xi = xi × (1 + C(𝛾))
In Eq. (15), C(𝛾) is a random number obtained from the
Cauchy mutation. Using the Cauchy mutation in the swarm
behavior section in the TSA algorithm leads to increased
exploitation and promising spaces by employing a much bet-
ter algorithm of the QLGCTSA algorithm. The quality of
solutions can be enhanced by employing the Cauchy muta-
tion in the entire optimization operation.
3.3 Gaussian Mutation ‑TSA (QGTSA)
In this subsection, Gaussian mutation uses to prevent losing
diversity during the search process. The Gaussian density
function shows in Eq. (16).
1 𝛼2
fgaussian(0,σ2 ) (α) = √ e− 2
2𝜋𝜎 2 2𝜎
In Eq. (16), σ2 Each search agent's variance creates a
random numerical. The Gaussian mutation with a Standard
Deviation (STD) of 1, and a mean value of 0. It is used for
producing equations such as Eq. (17).
1183
�
xi = xi × (1 + G(𝛼)) (17)
Equation (17), (α) is a uniform distribution taken from
the Gaussian mutation as a random value. It incorporates the
Gaussian mutation, such as the Lévy flight, in the rotational
motion phase. However, Eq. (17) is used instead of Eq. (12).
In QLGCTSA, the QRG mechanism[44] is proposed to
enhance local search capabilities and increase population
(14) diversity. The population data, consisting of several dimen-
sions, are generated as floating data by the GTSA algorithm.
Therefore, the quantum bit's discrete data must be converted
to continuous algorithm data, so these values ​​are placed
directly as quantum bits to work in the QRG. Setting up and
updating a QRG is as follows[44].
[ ( ) ( )]
( ) cos( 𝜃i) −sin( 𝜃)i
U 𝜃i = (18)
sin 𝜃i cos 𝜃i
[ �] [ ] [ ( ) ( ) ][ ]
𝛼i ( ) 𝛼i cos( 𝜃i) −sin( 𝜃)i 𝛼i
� = U 𝜃i = s (19).
𝛽i 𝛽i sin 𝜃i cos 𝜃i 𝛽i
Where, θi represents the angle of rotation of ith. The(method )T
of setting it in QLGCTSA is shown in Table 1. (αi , βi )
, T
shows the quantum bit state vectors before ( and) αi , βi
�
shows it in after updating the QRG.Where s αj , βj demon-
strates the direction of rotation, best_fitness demonstrates the
best
( )fitness value to date, indicates the angle of rotation,
(15) f xi demonstrates the fitness value of the ith population.
The main impress of the QRG is to show the particular popu-
lation by contrasting the fitness values of particulars in two
population groups a​ nd then changing the value of the poor
to the particular population.
3.5 QLGCTSA Algorithm
To create new solutions and to increase the performance
of the TSA algorithm, we hybridized three mutation opera-
tors of QLTSA, QCTSA, and QGTSA in a two-by-two form
and provided the QLCTSA (hybrid CTSA with LTSA),
QLGTSA (hybrid GTSA with LTSA), and QGCTSA (hybrid
CTSA with GTSA) algorithms and finally incorporated all
three and presented the hybrid algorithm of QLGCTSA.
The Gaussian mutation generates slight mutations in the
(16) solutions, and for this, the solutions produced have slight
changes or variations. Therefore, the Cauchy mutation cre-
ates more significant mutations than the Gaussian mutation,
which increases all search agents' general ability to conduct
an all-out search in each generation. Moreover, Lévy flight
has been employed to strengthen deeper search capabili-
ties. Metaheuristic algorithms make use of mutation pro-
cedures to produce new solutions. Algorithms using only
13
 F. S. Gharehchopogh
1184
one mutation operator rarely balance the exploration and
exploitation phases. For this, four significant changes were
used to increase performance. In the QLGCTSA algorithm,
Lévy flight, Gaussian mutation, and Cauchy mutation opera-
tors yielded new solutions. It was obtained by determining
the most appropriate distribution among the seven proposed
distributions. Three main Eqs. (20–22) are illustrated for
mutation operators.
xnewl = xi × (1 + L(𝛽)) (20)
xnewg = xi × (1 + G(𝛼)) (21)
xnewc = xi × (1 + C(𝛾)) (22)
Two equations (i.e., Eqs. 20 and 21) were used for the
QLGTSA algorithm and two Eqs. (20 and 22) for the
QLCTSA algorithm. As well, two Eqs. (21 and 22) were
used for the QGCTSA algorithm. Two equations were
13
simultaneously applied in these three algorithms, and the
best solution was selected. The QLGTSA algorithm con-
struct is illustrated in Fig. 1. It should be borne that the
construct used for QLCTSA and QGCTSA algorithms is the
same as that of the QLGTSA algorithm. However, two Eqs.
(20 and 22) where used in the QLCTSA algorithm, while
also two Eqs. (21 and 22) in the QGCTSA algorithm replac-
ing Eqs. (20–21). The Lévy flight makes more extensive
mutations in the solutions produced relative to the Gaussian
mutation and Cauchy mutations. The Lévy flight can take
significant strides in new solutions and, therefore, prevent
early convergence. In the QLGCTSA algorithm, three Eqs.
(20–22) were simultaneously used, and the best solution
was selected from the three created solutions. The construct
used for the QLGCTSA algorithm was the same as that pre-
sented in the LGTSA algorithms, except that Eqs. (20–22)
were employed instead of using Eqs. (20–22). QRG is used
in all algorithms. In Algorithm (3), the pseudo-code of the
QLGCTSA algorithm is shown.
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization…
3.6 Time Complexity
In this subsection, the time complexity of the QLGCTSA
algorithm is investigated. The computational complex-
ity of initial population generation in the TSA algorithm
is equal to O(n × d) . Where n is equal to the population
size and d represents the dimensions of the problem. The
(cost calculation operation
) of each solution is equal to
Maxiteration × n × d which Maxiteration is equal to the total
amount of iterations for the optimization operation. The TSA
algorithm takes O(n) time to improve solutions. The total
computational complexity)of the TSA algorithm is equal to
(
O Maxiteration × n × d × N .
4 Experiments and Results
The LGCTSA algorithm and other algorithms presented in
this paper were run and tested using Matlab 9.9 and a Win-
dows 7 Enterprise 32-bit Intel® Core™ i7-4510U Processor,
16.00 GB RAM. All the experiments and a review of algo-
rithms and algorithms' performance are conducted using 30
populations and 500 iterations. Experimental results were
according to a mean of 30 executions of the stored perfor-
mance compared. However, in solving the CEC 2017 com-
petition problem, all experiments were performed using 30
populations and 1000 iterations. Moreover, Wilcoxon's non-
parametric statistical test rank test [45] with a 0.05 emphasis
level was conducted to specify if the QLGCTSA algorithm
managed to produce statistical advancement. It should be
noted, however, that the symbols " + ," " = ," "–" suggest
whether or not the QLGCTSA algorithm had better or equal
performance or poor performance compared to other algo-
rithms compared. Hence, the Friedman test[46] was con-
ducted to assess all competitors' mean performance, while
the rating's Average Value (ARV) was provided to indicate
these test results.
4.1 Experiments on Test Functions
To investigate the QLGCTSA algorithm, a different set of
criterion functions [47, 48]. This set of criterion functions
included three UM, MM, and Cauchy mutation groups. The
criterion functions of UM (F1-F7) have one best global opti-
mization solutions. The capacity of the exploitation phase of
each optimization algorithm can be shown. While using MM
(F8-F23) criterion functions, the capabilities of the explora-
tion phase of optimization algorithms can be better shown.
Math formula and features of criterion functions of MM and
UM are illustrated in Tables 2, 3, 4.
Simulation results and findings from improving the
QTSA algorithm's performance by Gaussian, Cauchy, and
lévy flights for Fixed-dimension multimodal benchmark
1185
functions through Min, Max, Mean, and STD criteria are:
illustrated. Table 5 evaluates the TSA algorithm and its vari-
ous improvements. The benchmark functions of numbers
one to thirteen with dimensions 30 have been investigated.
It should be noted that QRG was used in all experiments
except for the TSA algorithm tested without any improve-
ment. On the other hand, in this section, the effect of muta-
tion operators on the QLGCTSA algorithm is investigated.
According to Table  5, it is clear that the LGCTSA
algorithm has the highest performance compared to other
algorithms. The best performance pertains to QLGTSA,
QGCTSA, QLCTSA, QLTSA, QGTSA, and QCTSA algo-
rithms. The results in Table 5 demonstrate that the TSA
algorithm substantially increases its performance using
mutation operators.
According to Fig. 2, the QLGCTSA algorithm has been
able to have the best performance in convergence. The
QLGTSA, QGCTSA, and QLCTSA algorithms have also
shown acceptable performance. Eventually, Examining
Fig. 2 shows that all algorithms integrated with mutation
operators have performed well.
Table  6 shows the results obtained from Wilcoxon's
nonparametric statistical test rank. According to the results
shown, it can be easily understood that the QLGCTSA algo-
rithm has a statistically good ability to discover quality solu-
tions. Because in most cases, it has been shown that it is
statistically significantly different from the solutions of other
algorithms. But functions 18 and 20 have statistically poor
results compared to other algorithms.
4.1.1 The Scalability Test for QLGCTSA
An experiment was conducted using large-dimension bench-
mark functions to examine the QLGCTSA algorithm and its
scalability (Table 7). The dimensions used to perform the
scalability test were 100, 500, and 1000, respectively.
Table  7 pertains to the scalability test findings using
the TSA and QLGCTSA algorithms. It is clear that the
QLGCTSA algorithm still finds better solutions even as
dimensions of the problems increase; using the proposed
mutation operators, the TSA algorithm's performance
increases significantly, thus yielding the desired perfor-
mance in solving solutions of high dimensions. The CEC
2017 competition benchmark functions (shown in Table 8)
were applied in the third group, covering hybrid rotated,
composite, and shifted MM cases.
The QLGCTSA algorithm results and performance are
compared with TSA [20], SCA [28], SHO [29], HHO [49],
MVO [50], and AVOA [39] optimization algorithms. In this
connection, optimizing algorithms were regulated in the
base algorithm, as seen in Table 9.
This comparison is based on STD and AVG, as shown in
Table 10. This table compares the QLGCTSA algorithm's
13
 F. S. Gharehchopogh
1186

Table 2  Describes the unimodal No Function Type Fmin Dimensions Range

benchmark functions
F1 ∑d US 0 30, 100, 500, 1000
f (x) = 2
i=1 xi
[−100, 100]d
F2 ∑d � � ∏d � � UN 0 30, 100, 500, 1000
f (x) = i=1 �xi � + i=1 �xi � [−10, 10]d
F3 ∑d �∑i �2 UN 0 30, 100, 500, 1000 [−100, 100]d
f (x) = i=1 j=1 xj
{ }
F4 f (x) = maxi |xi ||, 1 ≤ i ≤ d
| US 0 30, 100, 500, 1000 [−100, 100]d
∑ � � �2 � �2 �
F5 f (x) = d−1 100 xi+1 − xi2 + xi − 1 UN 0 30, 100, 500, 1000 [−30, 30]d
i=1

F6 ∑d �� �
� 2 US 0 30, 100, 500, 1000
f (x) = [−100, 100]d
i=1 �xi + 0.5�
F7 ∑d US 0 30, 100, 500, 1000
f (x) = i=1 ix4i + random[0, 1) [−128, 128]d

Table 3  Describes multimodal benchmark functions

No Function Type Fmin Dimensions Range
� �� ��
F8 ∑ �xi � MS  − 418.9829 × n 30, 100, 500, 1000 [−500, 500]d
f (x) = − di=1 xi sin � �

F9 ∑d � d � ��
MS 0 30, 100, 500, 1000 [−5.12, 5.12]d
f (x) = 10d + x − 10cos 2𝜋xi
�i=1 i � � � ∑ �
F10 ∑d d MN 0 30, 100, 500, 1000 [−32, 32]d
f (x) = −20exp −0.2 d1 i=1 xi2 − exp d1 i=1 cos2𝜋xi + 20 + e
� �
F11 f (x) = 1 ∑d ∏d x MN 0 30, 100, 500, 1000 [−600, 600]d
4000
x2
i=1 i
− i=1
cos √i +1
i
� � � ∑ � �2 � � �� � �2 � ∑d � �
F12 f (x) = 𝜋
10sin 𝜋y1 + d−1
i=1 yi − 1 1 + 10sin2 𝜋yi+1 + yd − 1 + i=1 U xi , 10, 100, 4
d
xi +1
yi = 1 + 4
⎧ � �m MN 0 30, 100, 500, 1000 [−50, 50]d
� � ⎪ k xi − a xi > a
U xi , a, k, m = ⎨ � 0 − a <� xi < a
⎪ k −xi − a m xi < −a
⎩
� � � ∑d � �2 � � �� � �2 � � ���
F13 f (x) = 0.1 sin2 3𝜋x1 + i=1 xi − 1 1 + sin2 3𝜋xi + 1 + xd − 1 1 + sin2 2𝜋xd MN 0 30, 100, 500, 1000 [−50, 50]d
∑d � �
+ i=1 U xi , 5, 100, 4

simulation results and other algorithms with SDT and AVG
parameters applied to the CEC2017 functions. The results
indicate that the QLGCTSA algorithm outperforms different
algorithms to be compared.
Table 10 data pertain to evaluating the LGCTSA algo-
rithm and other optimizing algorithms. That suggests a
better QLGCSTA algorithm's performance than differ-
ent optimization algorithms, including TSA, SCA, SHO,
HHO, and MVO. On the other hand, it becomes clear that
the QLGCTSA algorithm has outperformed other optimiza-
tion algorithms and has an excellent ability to solve several
benchmark functions. Figure 3 illustrates the convergence
diagram of different iterations on CEC2017 benchmark
functions on different optimization algorithms that demon-
strate that the QLGCTSA algorithm is better than the others.
It becomes clear that the QLGCTSA algorithm had out-
performed in convergence than other algorithms and had
better desirability. MVO and SHO algorithms have a minor
intersection as they found the worst solutions compared to
different optimizing algorithms. The HHO algorithm was
outperformed other optimization algorithms but under-per-
formed the QLGCTSA algorithm in most benchmark func-
tions. Finally, one can see that the QLGCTSA algorithm
outperformed the TSA algorithm while having a remark-
able ability to fight the optimal local trap. As shown in the
diagrams in Fig. 3, it becomes clear that the QLGCTSA
algorithm uses the proposed mutation operators to improve
solving optimization problems substantially.
Table 11 shows the results of the Wilcoxon rank-sum test
with 5% significance. According to the results, it can be
concluded that the performance of the QLGCTSA algorithm
in comparison with other comparable algorithms in testing
CEC2017 benchmark functions has a good capability and
has been able in most cases compared to different compari-
son algorithms. It has had a significant performance. This
shows that the solutions obtained in testing the CEC2017

13
 Table 4  Describes fixed-dimension multimodal benchmark functions
No Function Type Fmin Dimen- Range
sions
� �−1
F14 1 ∑25 1
FM 21 2 [−65, 65]d
f (x) = 500
+ i=1 i+∑2 6
j=1 (xj −aj,i )

∑ � 2 � 0.00030
F15 x (b +bi x2 ) 2 FM 4 [−5, 5]d
f (x) = di=1 ai − b12 +bi x +x
i i 3 4

F16 f (x) = 4x12 − 2.1x14 + 13 x16 + x1 x2 − 4x22 + 4x24 FM – 1.0316 2 [−5, 5]d
( )2 ( )
F17 5.1 2 5 1 FM 0.398 2 [−5, 5]d
f (x) = x2 − 4𝜋 2 x1 +
𝜋
x 1 − 6 + 10 1 − 8𝜋
cosx1 + 10
[ ( )2 ( )] [ ( )2 (
F18 f (x) = 1 + x1 + x2 + 1 19 − 14x1 + 3x21 − 14x2 + 6x1 x2 + 3x22 × 30 + 2x1 − 3x2 × 18 − 32x1 FM 3 2 [−2, 2]d
)]
+12x21 + 48x2 − 36x1 x2 + 27x22
∑ � ∑ � �2 �
F19 f (x) = − 4i=1 ai exp − 3j=1 bij xj − pij FM – 3.86 3 [1, 3]d

∑ � ∑ � �2 �
F20 f (x) = − 4i=1 ai exp − 6j=1 bij xj − pij FM – 3.32 6 [0, 1]d

∑5 �� �� �T �−1
F21
f (x) = − i=1 X − ai X − ai + ci
FM – 10.1532 4 [0, 10]d
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization…

∑7 �� �� �T �−1
F22
f (x) = − i=1 X − ai X − ai + ci
FM – 10.4028 4 [0, 10]d

∑10 �� �� �T �−1
F23
f (x) = − i=1 X − ai X − ai + ci
FM – 10.5363 4 [0, 10]d

13
1187
 F. S. Gharehchopogh
1188

Table 5  Examining the effect of Gaussian mutation, Cauchy mutation, and lévy flights on the QLGCTSA algorithm's performance
Function Metric TSA QLGCTSA QLCTSA QLGTSA QLTSA QGCTSA QGTSA QCTSA

F1 Min 2.49E-24 0.00E + 00 2.48E-241 0.00E + 00 9.77E-182 3.45E-257 1.52E-212 3.20E-140

Max 2.81E-21 0.00E + 00 4.72E-192 5.14E-264 9.81E-112 8.22E-215 2.04E-151 3.64E-96
Mean 6.13E-22 0.00E + 00 3.18E-193 3.43E-265 6.54E-113 5.48E-216 1.36E-152 2.43E-97
STD 7.33E-22 0.00E + 00 0.00E + 00 0.00E + 00 2.53E-112 0.00E + 00 5.27E-152 9.39E-97
F2 Min 2.06E-15 7.16E-240 4.54E-123 2.07E-153 5.36E-93 2.98E-131 6.10E-108 5.19E-73
Max 3.33E-13 1.73E-212 1.21E-101 1.55E-126 1.73E-67 5.81E-112 9.75E-86 3.71E-54
Mean 9.33E-14 1.15E-213 1.44E-102 1.34E-127 1.22E-68 3.88E-113 6.85E-87 2.85E-55
STD 1.09E-13 0.00E + 00 3.70E-102 4.10E-127 4.46E-68 1.50E-112 2.51E-86 9.54E-55
F3 Min 1.26E-07 0.00E + 00 1.57E-244 6.57E-292 2.16E-171 5.21E-271 1.42E-230 5.08E-132
Max 2.34E-03 0.00E + 00 1.54E-191 1.90E-256 7.30E-126 4.88E-228 1.76E-172 1.85E-94
Mean 2.08E-04 0.00E + 00 1.03E-192 1.27E-257 4.87E-127 3.43E-229 1.34E-173 1.23E-95
STD 5.95E-04 0.00E + 00 0.00E + 00 0.00E + 00 1.89E-126 0.00E + 00 0.00E + 00 4.77E-95
F4 Min 5.48E-02 3.67E-251 1.31E-120 5.68E-153 2.29E-89 5.68E-133 1.38E-99 8.44E-75
Max 8.54E-01 9.51E-208 5.78E-95 2.20E-132 2.29E-64 3.48E-110 4.84E-87 7.36E-51
Mean 2.70E-01 6.34E-209 3.86E-96 1.47E-133 2.18E-65 2.34E-111 3.38E-88 4.96E-52
STD 1.93E-01 0.00E + 00 1.49E-95 5.68E-133 6.24E-65 8.98E-111 1.25E-87 1.90E-51
F5 Min 2.72E + 01 9.23E-06 1.33E-04 9.52E-05 4.32E-04 2.09E-04 6.65E-04 8.48E-05
Max 3.07E + 01 7.96E-05 1.35E-02 5.70E-03 1.01E-01 1.49E-02 5.38E-02 3.69E-01
Mean 2.86E + 01 3.54E-05 3.80E-03 1.82E-03 2.20E-02 2.13E-03 1.05E-02 7.85E-02
STD 7.69E-01 1.81E-05 3.77E-03 2.15E-03 2.63E-02 3.82E-03 1.29E-02 9.76E-02
F6 Min 2.59E + 00 1.50E-08 5.83E-07 3.35E-07 1.06E-04 3.38E-07 9.43E-06 1.47E-04
Max 4.58E + 00 2.35E-07 9.00E-05 5.02E-06 4.34E-03 1.18E-05 2.47E-04 2.80E-02
Mean 3.68E + 00 9.15E-08 2.37E-05 1.84E-06 1.47E-03 3.97E-06 8.88E-05 5.65E-03
STD 6.05E-01 5.55E-08 2.42E-05 1.44E-06 1.35E-03 3.27E-06 7.61E-05 7.15E-03
F7 Min 2.56E-03 7.77E-06 4.97E-05 1.51E-05 7.00E-05 3.27E-05 6.28E-05 3.83E-05
Max 2.27E-02 3.30E-04 9.79E-04 5.98E-04 2.02E-03 7.60E-04 1.10E-03 2.70E-03
Mean 9.87E-03 9.06E-05 2.53E-04 2.34E-04 5.72E-04 2.55E-04 4.00E-04 8.31E-04
STD 4.70E-03 1.09E-04 2.40E-04 1.83E-04 5.14E-04 2.02E-04 3.23E-04 8.59E-04
F8 Min – 6.81E + 03 – 1.26E + 04 – 1.26E + 04 – 1.26E + 04 – 1.26E + 04 – 1.26E + 04 – 1.26E + 04 – 1.26E + 04
Max – 5.00E + 03 – 8.50E + 03 – 1.04E + 04 – 9.32E + 03 – 9.26E + 03 – 9.69E + 03 – 8.62E + 03 – 8.46E + 03
Mean – 6.04E + 03 – 1.16E + 04 – 1.22E + 04 – 1.16E + 04 – 1.16E + 04 – 1.13E + 04 – 1.17E + 04 – 1.09E + 04
STD 4.73E + 02 1.46E + 03 6.35E + 02 1.13E + 03 1.16E + 03 1.05E + 03 1.25E + 03 1.38E + 03
F9 Min 1.06E + 02 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
Max 2.29E + 02 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
Mean 1.76E + 02 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
STD 3.72E + 01 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
F10 Min 1.38E-12 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16
Max 3.40E + 00 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16
Mean 1.58E + 00 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16 8.88E-16
STD 1.55E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
F11 Min 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
Max 2.31E-02 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
Mean 7.92E-03 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
STD 9.17E-03 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00 0.00E + 00
F12 Min 3.39E + 00 7.05E-10 4.25E-08 2.80E-08 4.89E-07 2.67E-08 1.86E-07 6.15E-07
Max 1.34E + 01 6.68E-09 6.03E-06 1.29E-06 4.76E-04 1.67E-06 1.53E-05 2.67E-03
Mean 8.20E + 00 3.11E-09 2.01E-06 2.49E-07 1.08E-04 5.10E-07 4.69E-06 3.27E-04
STD 2.93E + 00 1.53E-09 1.80E-06 3.37E-07 1.30E-04 4.45E-07 4.19E-06 6.76E-04

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1189

Table 5  (continued)
Function Metric TSA QLGCTSA QLCTSA QLGTSA QLTSA QGCTSA QGTSA QCTSA

F13 Min 7.92E-01 3.93E-09 3.07E-07 3.31E-07 8.28E-06 3.61E-07 5.31E-07 3.51E-05
Max 7.09E + 00 1.40E-07 1.85E-04 1.33E-05 2.83E-02 1.22E-05 1.14E-02 2.60E-02
Mean 3.17E + 00 6.22E-08 2.57E-05 2.79E-06 2.43E-03 3.49E-06 8.36E-04 4.31E-03
STD 1.49E + 00 3.79E-08 4.63E-05 3.26E-06 7.23E-03 3.26E-06 2.91E-03 7.88E-03
F14 Min 9.98E-01 9.98E-01 9.98E-01 9.98E-01 9.98E-01 9.98E-01 9.98E-01 9.98E-01
Max 1.64E + 01 9.98E-01 2.98E + 00 2.98E + 00 1.99E + 00 3.97E + 00 2.98E + 00 5.93E + 00
Mean 8.39E + 00 9.98E-01 1.33E + 00 1.20E + 00 1.06E + 00 1.59E + 00 1.26E + 00 1.92E + 00
STD 5.79E + 00 2.59E-16 7.18E-01 5.56E-01 2.57E-01 1.11E + 00 6.98E-01 1.47E + 00
F15 Min 3.08E-04 3.07E-04 3.08E-04 3.08E-04 3.08E-04 3.07E-04 3.08E-04 3.09E-04
Max 5.66E-02 1.22E-03 2.04E-02 4.74E-04 2.04E-02 1.22E-03 2.04E-02 2.05E-02
Mean 8.24E-03 3.72E-04 3.23E-03 3.45E-04 1.94E-03 4.32E-04 1.86E-03 3.19E-03
STD 1.57E-02 2.36E-04 6.97E-03 5.25E-05 5.11E-03 2.45E-04 5.13E-03 7.02E-03
F16 Min – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00
Max – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00
Mean – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00 – 1.03E + 00
STD 3.29E-07 2.85E-16 5.57E-16 3.20E-16 5.73E-14 3.66E-16 1.78E-15 1.97E-10
F17 Min 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01
Max 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01
Mean 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01 3.98E-01
STD 3.34E-05 0.00E + 00 4.13E-14 1.30E-13 1.42E-11 6.25E-16 5.74E-12 8.00E-11
F18 Min 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00
Max 8.40E + 01 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00
Mean 2.28E + 01 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00 3.00E + 00
STD 3.30E + 01 1.22E-06 2.16E-07 2.60E-08 2.09E-06 5.47E-08 2.11E-07 4.45E-05
F19 Min – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00
Max – 3.85E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.85E + 00
Mean – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00 – 3.86E + 00
STD 2.01E-03 6.83E-14 1.39E-06 1.36E-08 7.46E-05 3.20E-08 4.64E-06 2.36E-03
F20 Min – 3.32E + 00 – 3.32E + 00 – 3.32E + 00 – 3.32E + 00 – 3.32E + 00 – 3.32E + 00 – 3.32E + 00 – 3.32E + 00
Max – 3.14E + 00 – 3.20E + 00 – 3.20E + 00 – 3.20E + 00 – 3.19E + 00 – 3.20E + 00 – 3.20E + 00 – 3.20E + 00
Mean – 3.26E + 00 – 3.26E + 00 – 3.28E + 00 – 3.27E + 00 – 3.30E + 00 – 3.28E + 00 – 3.29E + 00 – 3.30E + 00
STD 7.72E– 02 6.14E– 02 5.81E– 02 6.14E– 02 5.07E– 02 5.80E– 02 5.47E– 02 4.94E– 02
F21 Min – 1.01E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01
Max – 2.60E + 00 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01
Mean – 7.14E + 00 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01 – 1.02E + 01
STD 3.18E + 00 1.36E-12 3.73E-10 2.37E-11 5.32E-08 3.60E-11 2.22E-09 8.06E-07
F22 Min – 1.03E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01
Max – 1.84E + 00 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01
Mean – 6.46E + 00 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01 – 1.04E + 01
STD 3.72E + 00 1.23E– 12 2.65E– 10 3.30E– 11 1.06E– 07 3.04E– 11 3.63E– 09 4.76E– 07
F23 Min – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01
Max – 1.85E + 00 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01
Mean – 6.91E + 00 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01 – 1.05E + 01
STD 3.88E + 00 6.51E– 13 6.66E– 10 8.22E– 11 1.86E– 07 9.04E– 11 2.63E– 09 4.81E– 07
Overall Rank 8 1 4 2 5 3 6 7
Rank  + / = /– 22/1/0  ~  7/14/1 3/19/1 8/14/1 6/17/0 10/12/1 15/8/0
ARV 6.7941 3.6724 3.9825 3.9587 4.3715 3.8607 4.6948 4.6231

13
 F. S. Gharehchopogh
1190
Fig. 2  Convergence diagram about Gaussian, Cauchy, and lévy flight effects on the QLGCTSA algorithm
benchmark functions are of high quality. On the other hand,
trying the CEC2017 benchmark functions due to its high
complexity, the optimization algorithm to solve these prob-
lems should have a good performance in the exploration and
exploitation phases and create a good balance between them.
So, it can easily escape from the optimal local trap and has
a good ability in the diversity component in all optimization
stages. Based on the explanations given and the performance
of the QLGCTSA algorithm, it can be concluded that the
QLGCTSA algorithm has not only performed well in the
face of various standard functions. And, even if the dimen-
sions of the problem increase, it has been able to achieve a
quality solution. And maintain its optimization capabilities
if the problem size increases.
On the other hand, according to the convergence dia-
grams, it can be concluded that it has had a good ability
in terms of convergence. And, in all stages of optimization
can escape from the optimal local trap and repeatedly find
13
better solutions. Finally, according to all the experiments
performed, it can be concluded that the QLGCTSA algo-
rithm can be considered a good and effective option for solv-
ing various problems.
4.2 Engineering Optimization Problem
The QLGCTSA algorithm's performance to solve engineer-
ing problems is evaluated in this subsection. These prob-
lems were solved using P-metaheuristics as one of the hot
research fields[51]. A total of six engineering problems are
applied for this evaluation.
4.2.1 Speed Reducer Design Engineering Problem
This problem aims to minimize the total weight of the speed
reducer, which constraints are surface stress, transverse
deflection, and bending stress. It has various components
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1191

Table 6  P-values of the Wilcoxon rank-sum test with 5% significance for F1–F23 problems
No QLGCTSA algorithm QLGCTSA QLGCTSA QLGCTSA QLGCTSA QLGCTSA QLGCTSA
versus TSA algorithm versus algorithm versus algorithm versus algorithm versus algorithm versus algorithm ver-
QLGTSA QLCTSA QGCTSA QLTSA QGTSA sus QCTSA
P-values R P-values R P-values R P-values R P-values R P-values R P-values R

F1 1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  + 
F2 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  1.2118E-12  +  3.0199E-11  +  3.0199E-11  + 
F3 1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  + 
F4 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  + 
F5 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  + 
F6 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  + 
F7 3.0199E-11  +  1.3017E-02  +  1.0467E-03  +  2.1590E-03  +  4.6054E-03  +  1.1564E-02  +  4.4411E-03  + 
F8 3.0199E-11  +  3.9512E-01  =  6.4789E-01  =  4.2039E-01  =  4.4425E-01  =  5.5412E-02  =  2.6145E-01  = 
F9 1.2118E-12  +  NaN  =  NaN  =  NaN  =  NaN  =  NaN  =  NaN  = 
F10 1.2118E-12  +  NaN  =  NaN  =  NaN  =  NaN  =  NaN  =  NaN  = 
F11 1.2118E-12  +  NaN  =  NaN  =  NaN  =  NaN  =  NaN  =  NaN  = 
F12 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  + 
F13 3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  +  3.0199E-11  + 
F14 1.2108E-12  +  1.2108E-12  +  1.2108E-12  +  1.2108E-12  +  1.2108E-12  +  1.2108E-12  +  1.2108E-12  + 
F15 8.7584E-09  +  7.9022E-01  =  1.4025E-08  +  2.4038E-01  =  4.9074E-09  +  4.7984E-09  +  4.8947E-08  + 
F16 1.7203E-12  +  1.7203E-12  +  1.7203E-12  +  1.7203E-12  +  1.7203E-12  +  1.7203E-12  +  1.7203E-12  + 
F17 1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  + 
F18 1.2118E-12  +  8.1340E-05 – 2.2531E-02 – 1.8655E-03 – 7.0892E-01  =  2.5103E-02 – 1.5846E-01  = 
F19 1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  +  1.2118E-12  + 
F20 0.097091067596477  =  6.4820E-01  =  2.4549E-01  =  7.0892E-01  =  5.0691E-01  =  3.1951E-01  =  5.0691E-01  = 
F21 1.3369E-11  +  1.3369E-11  +  1.3369E-11  +  1.3369E-11  +  1.3369E-11  +  1.3369E-11  +  1.3369E-11  + 
F22 5.1436E-12  +  5.1436E-12  +  5.1436E-12  +  5.1436E-12  +  5.1436E-12  +  5.1436E-12  +  5.1436E-12  + 
F23 1.4488E-11  +  1.4488E-11  +  1.4488E-11  +  1.4488E-11  +  1.4488E-11  +  1.4488E-11  +  1.4488E-11  + 

such as the number of teeth on pinion ­(u1), the module of ( ) 1.93u3

5
teeth ­(u2), face width ­(u3), length of shaft one between bear-
⇀
g4 u = −1≤0
u2 u47 u3
ings ­(u4), length of shaft two between bearings (­ u5), the
diameter of shaft 1 (­ u6), and diameter of shaft two (­ u7) [52].
The analytical expression is as follows: [( )2 ]1
2
745u4 6
M i n i m i z e f1 ( u )   =   0 . 7 8 5 4 + 157.5 × 10
⇀

( )
( ) u2 u3

( 3 − 43.0934 ) − 1.508u1 (u6 + u7 )   +   7 . 4 7 7

⇀
2 2
) 3 + 14.933u
(u1 u32 3.3333u
2 2 g5 u = −1≤0
u6 + u37 + 0.7854 u4 u26 + u5 u27 110.u36
Subject to
( ) [( )2 ]1
⇀ 27 745u4 6
2
g1 u = −1≤0 ( ) + 16.9 × 10
u1 u22 u3 ⇀
u2 u3
g6 u = −1≤0
85.u37
( )
⇀ 397.5
g2 u = −1≤0 ( ) uu
u1 u22 u23
g7 u = 2 3 − 1 ≤ 0
⇀

40
( ) 1.93u3 where 2.6 ≤ u1 ≤ 3.6, 0.7 ≤ u2 ≤ 0.8, 17 ≤ u3 ≤ 28, 7.3
⇀ 4
g3 u = −1≤0 ≤ u4 ≤ 8.3, 7.3 ≤ u5 ≤ 8.3 2 . )9
4 .
u2 u6 u3 ⇀ (
≤ u6 ≤ 3.9 and 5.0 ≤ u7 ≤ 5.5, u = u1 .u2 .u3 .u4 .u5 .u6 .u7
Table 12 summarizes the solutions for this problem with
several optimization algorithms. The results were shown

13
 F. S. Gharehchopogh
1192

Table 7  Investigating the effect of Cauchy mutation, Gaussian mutation, and Lévy flight on the QLGCTSA algorithm
F1 Avg Std F2 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 0.00E + 00 2.56E-10 0.00E + 00 3.11E-10 100 8.17E-217 2.52E-07 0.00E + 00 1.55E-07
500 0.00E + 00 2.10E-03 0.00E + 00 1.37E-02 500 4.73E-213 1.02E-02 0.00E + 00 3.73E-03
1000 0.00E + 00 4.91E + 00 0.00E + 00 2.41E + 00 1000 0.00E + 00 3.30E-02 0.00E + 00 2.71E-02
F3 Avg Std F4 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 0.00E + 00 1.09E + 04 0.00E + 00 8.27E + 03 100 7.05E-222 6.23E + 01 0.00E + 00 1.36E + 01
500 0.00E + 00 1.28E + 06 0.00E + 00 1.57E + 05 500 2.60E-217 9.93E + 01 0.00E + 00 1.72E-01
1000 0.00E + 00 6.11E + 06 0.00E + 00 8.13E + 05 1000 1.01E-216 9.95E + 01 0.00E + 00 1.65E-01
F5 Avg Std F6 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 3.80E-04 9.85E + 01 2.56E-04 2.03E-01 100 2.22E-05 1.44E + 01 1.09E-05 1.10E + 00
500 5.24E-03 9.83E + 04 4.81E-03 9.27E + 04 500 7.59E-04 1.02E + 03 1.26E-03 1.97E + 00
1000 9.77E-03 4.51E + 07 6.93E-03 2.74E + 07 1000 5.44E-02 2.35E + 03 1.61E-01 6.32E + 00
F7 Avg Std F8 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 6.35E-05 5.04E-02 5.03E-05 1.00E-02 100 – 3.97E + 04 – 1.32E + 04 2.39E + 03 1.23E + 03
500 1.07E-04 3.37E + 00 9.64E-05 1.32E + 00 500 – 1.96E + 05 – 3.14E + 04 1.45E + 04 2.69E + 03
1000 4.49E-05 3.87E + 02 3.41E-05 2.40E + 02 1000 – 3.95E + 05 – 4.39E + 04 3.58E + 04 3.68E + 03
F9 Avg Std F10 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 0.00E + 00 9.09E + 02 0.00E + 00 1.47E + 02 100 8.89E-16 2.09E-01 0.00E + 00 8.10E-01
500 0.00E + 00 6.00E + 03 0.00E + 00 4.78E + 02 500 8.89E-16 1.26E-02 0.00E + 00 4.44E-03
1000 0.00E + 00 9.12E + 03 0.00E + 00 1.70E + 03 1000 8.89E-16 9.50E-02 0.00E + 00 4.02E-03
F11 Avg Std F12 Avg Std
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 0.00E + 00 5.78E-04 0.00E + 00 1.21E-02 100 7.42E-08 1.39E + 01 3.34E-08 7.67E + 00
500 0.00E + 00 3.91E-03 0.00E + 00 7.71E-02 500 1.91E-07 3.52E + 06 1.08E-07 5.73E + 06
1000 0.00E + 00 3.59E-01 0.00E + 00 2.34E-01 1000 1.86E-07 6.22E + 08 1.89E-07 2.36E + 08
F13 Avg Std Overall rank Rank
QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA QLGCTSA TSA

100 1.45E-07 2.08E + 01 1.04E-07 3.00E + 01 13 0 1 2

500 6.77E-07 1.17E + 06 7.83E-07 1.02E + 06 13 0 1 2
1000 1.41E-06 4.92E + 08 1.14E-06 1.82E + 08 13 0 1 2

with the parameter setting to compare the QLGCTSA algo-
rithm's performance. The experimental results show the
superiority of the QLGCTSA algorithm over different opti-
mization algorithms.
4.2.2 Cantilever Beam Design Engineering Problem
It is related to optimizing the cantilever beam's weight with
a square cross-section[52]. The analytical expression is as
follows: ( )
Minimize f2 u = 0.0624(u1 + u2 + u3 + u4 + u5 )
⇀

Subject to

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1193

Table 8  CEC2017 benchmark No Name of the function Function Optimum

functions
F24 Unimodal Shifted and Rotated Bent Cigar Function C01 100
F25 Unimodal Shifted and Rotated Zakharov Function C03 300
F26 Multimodal Shifted and Rotated Rosenbrock's Function C04 400
F27 Multimodal Shifted and Rotated Rastrigin's Function C05 500
F28 Multimodal Shifted and Rotated Expanded Schaffer's F6 Function C06 600
F29 Multimodal Shifted and Rotated Lunacek Bi-Rastrigin Function C07 700
F30 Multimodal Shifted and Rotated Non-Continuous Rastrigin's Function C08 800
F31 Multimodal Shifted and Rotated Lévy Function C09 900
F32 Multimodal Shifted and Rotated Schwefel's Function C10 1000
F33 Hybrid Function 1 (N = 3) C11 1100
F34 Hybrid Function 2 (N = 3) C12 1200
F35 Hybrid Function 3 (N = 3) C13 1300
F36 Hybrid Function 4 (N = 4) C14 1400
F37 Hybrid Function 5 (N = 4) C15 1500
F38 Hybrid Function 6 (N = 4) C16 1600
F39 Hybrid Function 7 (N = 5) C17 1700
F40 Hybrid Function 8 (N = 5) C18 1800
F41 Hybrid Function 9 (N = 5) C19 1900
F42 Hybrid Function 10 (N = 6) C20 2000
F43 Composition Function 1 (N = 3) C21 2100
F44 Composition Function 2 (N = 3) C22 2200
F45 Composition Function 3 (N = 4) C23 2300
F46 Composition Function 4 (N = 4) C24 2400
F47 Composition Function 5 (N = 5) C25 2500
F48 Composition Function 6 (N = 5) C26 2600
F49 Composition Function 7 (N = 6) C27 2700
F50 Composition Function 8 (N = 6) C28 2800
F51 Composition Function 9 (N = 3) C29 2900
F52 Composition Function 10 (N = 3) C30 3000

g1 ( u) = 61 37 19
− 1 ≤ 0 where 0.01
7 1
⇀
Table 9  Parameter settings of optimization algorithms u3
+ u32
+ u33
+ u34
+ u35
1 ( ⇀ )
Algorithm Parameter Value ≤ uj ≤ 100(j = 1.2.3.4.5).u u1 .u2. u3 .u4 .u5
The QLGCTSA algorithm solves this problem and gets
SHO Control Parameter (h)
�⃗ [5, 0] optimum output notes, as shown in Table 13.
��⃗ Constant
M [0.5, 1] The experimental result of the QLGCTSA algorithm is
HHO The energy of a rabbit E ∈ [0, 2] compared with SCA [28], CS [52], m-SCA [64], and GCA
TSA Parameter Pmin 1 [65] optimization algorithms. The optimum cost is 1.3401,
Parameter Pmax 4 and the optimum values of the different beams' width (or
SCA Flight length 0.01 height) are 6.01604, 5.30915, 4.4943, 3.5015, and 2.1527.
Awareness probability 0.1 As a result, the efficient execution valence of the QLGCTSA
MVO Wmax(max wep) 1 algorithm is in contrast to the other optimization algorithm.
Wmin(min wep) 0.2
p (Exploitation accuracy) 6 4.2.3 Tension/Compression Spring Design Problem
AVOA L1 0.8
L2 0.2 The objective of this problem is to minimize the tension /
w 2.5 compression spring weight. Its optimum design must sat-
P1 0.6 isfy deflection constraints, shear stress, and surge frequency.
P2 0.4 Wire diameter ­(u1), mean coil diameter ­(u2), and the number
P3 0.6

13
 F. S. Gharehchopogh
1194

Table 10  Results on CEC2017 Algorithm F24 F25 F26

benchmark Functions
AVG STD AVG STD AVG STD

LGCSTA 2.66E + 03 3.23E + 03 3.00E + 02 6.84E-10 4.18E + 02 1.93E + 01

TSA 3.12E + 09 2.20E + 09 1.04E + 04 5.50E + 03 7.96E + 02 6.59E + 02
SCA 9.15E + 08 4.62E + 08 2.73E + 03 1.41E + 03 4.58E + 02 1.61E + 01
SHO 1.57E + 10 5.62E + 09 1.35E + 05 1.99E + 05 4.40E + 03 1.46E + 03
HHO 9.84E + 04 5.18E + 04 3.02E + 02 2.00E + 00 4.19E + 02 3.26E + 01
MVO 7.44E + 05 9.15E + 05 1.35E + 03 5.83E + 02 4.79E + 02 8.06E + 00
AVOA 7.42E + 03 7.83E + 03 8.04E + 02 7.57E + 02 4.27E + 02 2.84E + 01
F27 F28 F29
AVG STD AVG STD AVG STD

LGCSTA 5.37E + 02 1.15E + 01 6.08E + 02 6.03E + 00 7.46E + 02 1.77E + 01

TSA 5.62E + 02 1.29E + 01 6.31E + 02 1.07E + 01 8.12E + 02 3.66E + 01
SCA 5.55E + 02 8.75E + 00 6.21E + 02 5.13E + 00 7.92E + 02 1.23E + 01
SHO 6.44E + 02 2.32E + 01 6.81E + 02 1.97E + 01 9.09E + 02 2.40E + 01
HHO 5.55E + 02 9.80E + 00 6.22E + 02 9.71E + 00 7.90E + 02 2.64E + 01
MVO 6.15E + 02 3.27E + 01 6.29E + 02 1.81E + 01 8.44E + 02 2.48E + 01
AVOA 5.47E + 02 1.81E + 01 6.10E + 02 4.39E + 00 7.53E + 02 1.38E + 01
F30 F31 F32
AVG STD AVG STD AVG STD

LGCSTA 8.35E + 02 10.7859017 1.31E + 03 3.56E + 02 1.70E + 03 2.30E + 02

TSA 8.56E + 02 18.9296206 2.00E + 03 9.03E + 02 1.85E + 03 3.45E + 02
SCA 8.50E + 02 6.82075366 1.24E + 03 1.52E + 02 2.32E + 03 2.70E + 02
SHO 9.17E + 02 15.2027116 4.40E + 03 6.62E + 02 3.33E + 03 1.88E + 02
HHO 8.56E + 02 12.2332981 2.21E + 03 5.27E + 02 1.63E + 03 3.02E + 02
MVO 9.17E + 02 33.1803801 5.94E + 03 3.99E + 03 4.28E + 03 5.88E + 02
AVOA 8.37E + 02 1.55E + 01 1.18E + 03 2.01E + 02 1.89E + 03 3.49E + 02
F33 F34 F35
AVG STD AVG STD AVG STD

LGCSTA 1.13E + 03 2.44E + 01 2.68E + 04 4.00E + 04 6.36E + 03 4.89E + 03

TSA 1.79E + 03 1.21E + 03 3.87E + 07 8.35E + 07 1.83E + 04 8.33E + 03
SCA 1.26E + 03 5.08E + 01 8.52E + 06 4.34E + 06 1.00E + 05 8.72E + 04
SHO 8.64E + 03 6.00E + 03 7.88E + 08 5.08E + 08 1.16E + 08 1.98E + 08
HHO 1.20E + 03 4.94E + 01 4.57E + 05 4.12E + 05 8.47E + 03 6.72E + 03
MVO 1.32E + 03 5.64E + 01 2.86E + 07 2.86E + 07 2.68E + 04 1.71E + 04
AVOA 1.19E + 03 1.81E + 01 2.28E + 05 5.26E + 05 7.60E + 03 4.30E + 03
F36 F37 F38
AVG STD AVG STD AVG STD

LGCSTA 1.67E + 03 2.81E + 02 1.72E + 03 1.31E + 02 1.72E + 03 1.14E + 02

TSA 5.32E + 03 2.71E + 03 3.31E + 04 3.81E + 04 1.80E + 03 1.11E + 02
SCA 1.84E + 03 3.62E + 02 3.07E + 03 6.29E + 02 1.75E + 03 5.42E + 01
SHO 4.01E + 06 7.86E + 06 1.31E + 06 1.64E + 06 2.70E + 03 2.56E + 02
HHO 2.01E + 03 6.48E + 02 2.09E + 03 2.46E + 02 1.80E + 03 1.14E + 02
MVO 4.14E + 04 2.23E + 04 7.21E + 04 6.10E + 04 2.53E + 03 3.38E + 02
AVOA 3.63E + 03 2.70E + 03 4.92E + 03 6.59E + 03 1.76E + 03 1.11E + 02

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1195

Table 10  (continued) F39 F40 F41

AVG STD AVG STD AVG STD

LGCSTA 1.78E + 03 6.12E + 01 2.81E + 04 2.65E + 04 3.77E + 03 2.53E + 03

TSA 1.88E + 03 1.22E + 02 2.53E + 06 6.63E + 06 3.43E + 05 5.30E + 05
SCA 1.80E + 03 3.22E + 01 1.34E + 05 1.36E + 05 7.04E + 03 3.39E + 03
SHO 2.47E + 03 2.17E + 02 3.36E + 08 3.46E + 08 1.81E + 07 2.27E + 07
HHO 1.83E + 03 8.07E + 01 4.70E + 04 2.74E + 04 5.65E + 03 3.09E + 03
MVO 2.19E + 03 2.43E + 02 4.01E + 05 3.51E + 05 4.28E + 05 4.08E + 05
AVOA 1.89E + 03 4.14E + 01 5.09E + 04 3.76E + 04 8.04E + 03 4.86E + 03
F42 F43 F44
AVG STD AVG STD AVG STD

LGCSTA 2.07E + 03 4.34E + 01 2.20E + 03 5.86E-08 2.31E + 03 1.11E + 01

TSA 2.19E + 03 1.16E + 02 2.33E + 03 4.83E + 01 3.14E + 03 5.91E + 02
SCA 2.10E + 03 1.24E + 01 2.23E + 03 3.36E + 01 2.42E + 03 4.30E + 01
SHO 2.46E + 03 1.11E + 02 2.40E + 03 5.55E + 01 4.41E + 03 3.75E + 02
HHO 2.12E + 03 6.27E + 01 2.20E + 03 7.34E-02 2.48E + 03 4.21E + 02
MVO 2.51E + 03 2.15E + 02 2.43E + 03 2.81E + 01 5.11E + 03 1.55E + 03
AVOA 2.09E + 03 5.25E + 01 2.24E + 03 3.55E + 00 2.49E + 03 3.70E + 02
F45 F46 F47
AVG STD AVG STD AVG STD

LGCSTA 2.64E + 03 1.20E + 01 2.77E + 03 1.09E + 02 2.93E + 03 9.18E + 01

TSA 2.70E + 03 3.38E + 01 2.81E + 03 5.96E + 01 3.15E + 03 1.90E + 02
SCA 2.66E + 03 8.90E + 00 2.81E + 03 8.90E + 00 2.98E + 03 2.51E + 01
SHO 2.79E + 03 5.97E + 01 2.94E + 03 4.44E + 01 4.09E + 03 3.78E + 02
HHO 2.67E + 03 2.83E + 01 2.85E + 03 3.70E + 01 2.93E + 03 9.33E + 01
MVO 2.81E + 03 3.31E + 01 2.88E + 03 3.24E + 01 3.01E + 03 6.84E + 00
AVOA 2.65E + 03 1.88E + 01 2.82E + 03 2.43E + 01 2.95E + 03 1.24E + 02
F48 F49 F50
AVG STD AVG STD AVG STD

LGCSTA 3.06E + 03 3.62E + 02 3.07E + 03 2.18E + 01 3.15E + 03 4.78E + 01

TSA 3.45E + 03 3.91E + 02 3.15E + 03 4.35E + 01 3.33E + 03 8.84E + 01
SCA 3.33E + 03 2.74E + 02 3.09E + 03 4.58E + 00 3.22E + 03 1.53E + 01
SHO 4.42E + 03 1.38E + 02 3.42E + 03 1.39E + 02 3.53E + 03 5.91E + 01
HHO 3.44E + 03 5.25E + 02 3.11E + 03 1.97E + 01 3.18E + 03 4.73E + 01
MVO 4.71E + 03 5.95E + 02 3.20E + 03 2.46E + 01 3.23E + 03 3.36E + 01
AVOA 3.15E + 03 3.41E + 02 3.08E + 03 3.15E + 01 3.17E + 03 7.64E + 01
F51 F52 Overall Rank
AVG STD AVG STD Rank ARV

LGCSTA 3.28E + 03 8.52E + 01 9.95E + 04 2.35E + 05 1 2.2374

TSA 3.37E + 03 1.16E + 02 2.00E + 06 3.68E + 06 6 11.2971
SCA 3.25E + 03 2.84E + 01 3.94E + 05 5.05E + 05 4 5.1672
SHO 3.81E + 03 1.66E + 02 6.63E + 07 5.47E + 07 7 11.6913
HHO 3.32E + 03 1.08E + 02 1.18E + 05 3.44E + 05 3 4.9541
MVO 4.27E + 03 3.05E + 02 7.47E + 06 5.73E + 06 5 11.1437
AVOA 3.27E + 03 1.02E + 02 2.38E + 05 2.95E + 05 2 4.2312

13
 F. S. Gharehchopogh
1196

Fig. 3  Convergence diagram on the evaluation of the LGCTSA algorithm and other optimizing algorithms in solving CEC2017 problems

of active coils (u3) are three design variables. The analytical The standard control parameters' parameter setting is the
expression is as follows: same as [53] in the QLGCTSA algorithm. The experimental
Minimize f3 (u) = (u3 + 2)u2 u21 result in Table 14 shows that the QLGCTSA algorithm per-
⇀

Subject
( ) to u3 +u forms better than all other optimization algorithms.
g1 u = 1-71785u
⇀ 2 3
4 ≤ 0
( ) 2 1
4u2 −u1 u2 4.2.4 Three‑Bar Truss Design Problem
2  -1 ≤ 0
⇀ 1
g2 u = 12566(u3 u −u4 ) + 5108u
1 2 1 1
( ) 140.45u1 The objective of this problem is to define the optimum cross-
⇀
g3 u = 1 − ≤0 sectional areas. It involved only two decision parameters that
u22 u3
can be retouched to obtain the minimum weight of stress,
( ) buckling constraints, and deflection. The analytical expres-
where 0.05 ≤ u1 ≤ 2.0 ⇀
≤ u2 ≤ 1.3, 2 ≤ u3 ≤ 15;u = u1 .u2 .u3
sion is as follows:
25
This engineering optimization problem was resolved √
Minimize f4 (u) = (2 2u1 + u2 ) × l1
⇀
using either mathematical techniques such as WOA [53],
GSA [66], GSA [66], HS [67], Improved HS [67], GA [68], Subject to
ES [55], DE [69], and PSO [54]. The QLGCTSA algorithm's √
� � 2u1 + u2
optimized results are compared with the above algorithms g1 u = √
⇀
p−𝜎 ≤0
presented in Table 14 regarding the design variables and 2u21 + 2u1 u2
optimum compression spring weight.

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1197

Table 11  P-values of the No QLGCTSA QLGCTSA QLGCTSA QLGCTSA QLGCTSA QLGCTSA

Wilcoxon rank-sum test for algorithm algorithm algorithm algorithm algorithm algorithm ver-
F24–F52 problems versus TSA versus SCA versus SHO versus HHO versus MVO sus AVOA
p-values R P-values R P-values R P-values R P-values R p-values R

F24 3.02E-11  +  3.02E-11  +  3.02E-11  +  4.50E-11  +  3.02E-11  +  3.81E-02  + 
F25 3.02E-11  +  3.02E-11  +  3.02E-11  +  3.02E-11  +  3.02E-11  +  3.39E-06  + 
F26 5.65E-04  +  1.29E-05  +  3.02E-11  +  8.03E-01 - 3.02E-11  +  2.23E-02  + 
F27 4.81E-05  +  9.66E-05  +  3.02E-11  +  1.89E-04  +  3.02E-11  +  9.01E-01  = 
F28 6.15E-06  +  2.80E-05  +  3.02E-11  +  1.60E-04  +  1.89E-04  +  4.12E-02  + 
F29 3.36E-05  +  3.39E-06  +  3.02E-11  +  5.74E-05  +  3.02E-11  +  3.40E-02  + 
F30 1.87E-03  +  3.08E-04  =  3.02E-11  +  8.13E-05  +  3.02E-11  +  7.40E-02  = 
F31 1.44E-02  +  9.67E-01  =  3.02E-11  +  8.13E-05  +  3.02E-11  +  5.61E-02  = 
F32 1.99E-01  =  4.81E-05  +  3.02E-11  +  2.29E-01  =  3.02E-11  +  1.35E-02  + 
F33 1.33E-05  +  4.14E-06  +  3.02E-11  +  2.23E-04  +  3.02E-11  +  8.95E-01  = 
F34 9.07E-06  +  3.39E-06  +  3.02E-11  +  3.36E-05  +  3.02E-11  +  1.87E-03  + 
F35 7.80E-04  +  4.02E-05  +  3.02E-11  +  9.34E-01  =  5.76E-04  +  3.34E-01  + 
F36 2.62E-04  +  1.28E-02  +  3.02E-11  +  1.15E-01  =  3.02E-11  +  1.28E-02  + 
F37 2.82E-03  +  3.39E-06  +  3.02E-11  +  2.23E-04  +  3.02E-11  +  2.15E-03  + 
F38 7.94E-03  +  2.81E-01  =  3.02E-11  +  4.21E-02  +  3.02E-11  +  2.29E-01  = 
F39 2.15E-03  +  3.44E-02  +  3.02E-11  +  1.81E-02  +  3.02E-11  +  1.47E-01  = 
F40 6.78E-01  =  1.89E-04  +  3.02E-11  +  5.64E-02  =  3.02E-11  +  1.90E-02  + 
F41 1.81E-02  +  2.46E-03  +  3.02E-11  +  7.45E-02  +  3.02E-11  +  1.14E-02  + 
F42 1.40E-03  +  6.80E-02  =  3.02E-11  +  2.25E-02  +  3.02E-11  +  4.81E-02  + 
F43 3.39E-06  +  3.39E-06  +  3.02E-11  +  3.39E-06  +  3.02E-11  +  3.39E-06  + 
F44 9.07E-06  +  3.39E-06  +  3.02E-11  +  5.64E-02  =  3.02E-11  +  4.48E-02  + 
F45 2.33E-05  +  1.87E-03  =  3.02E-11  +  3.23E-03  +  1.40E-03  +  3.81E-02  + 
F46 1.47E-01  =  1.99E-01  =  3.02E-11  +  1.87E-03  +  3.02E-11  +  8.90E-03  + 
F47 9.07E-06  +  5.76E-04  =  3.02E-11  +  7.72E-01  =  3.02E-11  +  3.20E-01  = 
F48 2.79E-02  +  1.58E-01  =  3.02E-11  +  6.20E-02  =  3.02E-11  +  4.46E-01  = 
F49 1.33E-05  +  3.10E-02  +  3.02E-11  +  1.89E-04  +  3.02E-11  +  2.29E-02  = 
F50 3.39E-06  +  3.39E-06  +  3.02E-11  +  4.81E-05  +  3.02E-11  +  3.10E-02  + 
F51 1.61E-02  +  5.07E-01  =  3.02E-11  +  3.20E-01  =  3.02E-11  +  7.40E-01  = 
F52 2.02E-02  +  2.62E-04  +  3.02E-11  +  9.71E-02  =  3.02E-11  +  7.94E-03  + 

� � u2 the QLGCTSA algorithm performs better than other optimi-

g2 u = √
⇀
p − 𝜎9 ≤ 0
2u21 + 2u1 u2 zation algorithms.

� � 4.2.5 I‑Beam Design Engineering Problem

1
g3 u = √
⇀
p−𝜎 ≤0
2u2 + u1 This problem, including four variables, has been tested by
( ) the QLGCTSA algorithm and other optimization algorithms.
where 0 ≤ u1 ≤ 1.0 ≤ u2 ≤ 1;l = 100cm.p = 2kN
cm2
.𝜎 = 2kN
cm2
⇀
.u = u1 .u2 Its goal is to minimize vertical deflection. The stress con-
The obtained solutions by the QLGCTSA and other opti- straints under specified loads and cross-sectional areas sat-
mization algorithms are presented in Table 15. In the lit- isfy this optimization problem. The analytical expression is
erature, this problem tries to solve by several optimization as follows: ( )
algorithms. The experimental results in Table 15 show that
Minimize f5 u = u (u −2u ) u 5000
⇀

the QLGCTSA algorithm performs better than other opti- 3 1

u3
4
+
2
2 4
+2u2 u4 (
u1 −u4
)
12 6 2
mization algorithms. Subject to ( )
In the literature, this problem tries to solve by several g1(u) = 2 u2 u3 + u3 u1 − 2u4 − 300 ≤ 0
⇀

optimization algorithms. The experimental results show that

13
 F. S. Gharehchopogh
1198

Table 12  Solving speed reducer problem with QLGCTSA and another optimization algorithm
Algorithms u1 u2 u3 u4 u5 u6 u7 fmin

QLGCTSA algorithm 3.5 0.7 17 7.3 7.715319 3.350217 5.286655 2994.4711

GWO [26] 3.6 0.8 28 7.3 8.3 2.9 5.0 3020.2331
WOA [53] 3.52111 0.7 17 7.3 7.8 3.35021 5.29533 3010.1480
SCA [28] 3.51889 0.7 17 7.3 8.3 3.35899 5.30519 3028.8657
CS [52] 3.50150 0.7 17 7.6050 7.8181 3.3520 5.2875 3000.9810
PSO [54] 3.58147 0.7 17.8282 7.98445 7.82083 3.15398 5.1873 3005.3248
Montes and Coello [55] 3.50616 0.700831 17 7.46018 7.962143 3.3629 5.3090 3025.005
Sine cosine GWO [56] 3.00576 0.72755 21.8423 7.30835 8.15455 3.36452 5.25164 3027.5130
Ray and Saini [57] 3.51418 0.700005 17 7.497343 7.8346 2.9018 5.0022 2732.9006
Akhtar et al. [58] 3.50612 0.700006 17 7.549126 7.85933 3.36558 5.289773 3008.08
Modified GWO [59] 3.50008 0.7 17 7.3193 7.81168 3.35072 5.28692 2997.085
Modified SCA [60] 3.52394 0.7 17 7.3 7.8 3.36280 5.32467 3033.2845
Moth-flame optimization algorithm [61] 3.59093 0.70554 19.7972 8.08267 7.84181 3.70621 5.48167 3836.2164
Salp Swarm Algorithm [62] 3.50031 0.7 17 7.80001 7.85001 3.35247 5.2867 3002.5678
Improved SCA [63] 3.50081 0.7 17 7.3 7.8 3.35129 5.28698 2997.1295

Table 13  Solving cantilever Algorithm Optimum variables Optimum cost

beam design engineering
problem with LGCTSA and U1 U2 U3 U4 U5
another optimization algorithm
QLGCTSA algorithm 6.01604 5.30915 4.4943 3.5015 2.1527 1.3401
SCA [28] 6.0100 5.3000 4.4901 3.4900 2.1500 1.3400
CS [52] 6.0089 5.3049 4.5023 3.5077 2.1504 1.33999
m-SCA [64] 6.0089 5.3049 4.5023 3.5077 2.1504 1.33999
GCA [65] 6.0100 5.3000 4.4900 3.4900 2.1500 1.3400

Table 14  Solving tension/ Algorithms Optimum Value Optimum Weight

compression spring design
problem with LGCTSA and u1 u2 u3
another optimization algorithm
QLGCTSA algorithm 0.355875 0.051654 11.338558 0.0126654
WOA [53] 0.051207 0.345215 12.004032 0.0126763
GSA [66] 0.050276 0.323680 13.525410 0.0127022
GSA [66] 0.050276 0.323680 13.525410 0.0127022
HS [67] 0 0.053153 0 0.359871 12.078945 0 0.0125799
Improved HS [67] 0 0.051154 0 0.349871 12.076432 0 0.0126706
GA [68] 0.051480 0.351661 11.632201 0.0127048
ES [55] 0.051989 0.363965 10.890522 0.0126810
DE [69] 0.051609 0.354714 11.410831 0.0126702
PSO [54] 0.051728 0.357644 11.244543 0.0126747

13
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization… 1199

Table 15  Solving three-bar Algorithm Decision Variables Optimal Cost

truss design problem with
QLGCTSA and another u1 u2
optimization algorithm
QLGCTSA Algorithm 0.78866378 0.40828041 263.8958435
MVO [50] 0.788602 0.40845307 263.8958499
CS [52] 0.78867 0.40902 263.97156
Ray and Saini [57] 0.795 0.395 264.3
MFO [61] 0.788244770931922 0.40946695784741 263.895979682
PRO [70] 0.7886475 0.4083262 263.8958439
ALO [71] 0.7888975 0.4082831 263.8958434
DEDS [72] 0.78867513 0.40824828 263.8958434
ERaFA [73] 0.78867559 0.40824698 263.8958433
BWOA [74] 0.788666327 0.408273202 263.8958435
OBLGOA [75] 0.78866365 0.408280786 263.895844
DSS-MDE [75] 0.78867513 0.40824828 263.8958434
DSM-ABC [75] 0.682465 0.421956 211.2628
GOA [76] 0.788897556 0.40761957 263.8958815
PSO-DE [77] 0.7886751 0.4082482 263.8958434
Tsai [78] 0.788 0.408 263.68
MBA [79] 0.788565 0.4085597 263.895852
QEA [80] 0.792969 0.396484 263.9339
pFA [81] 0.788676772 0.408243657 263.8958433
Ray and Liew [82] 0.7886210370 0.4084013340 263.8958466
Raj et al. [83] 0.78976441 0.40517605 263.89671

Table 16  Solving I-beam design engineering problem QLGCTSA g2(u) =

⇀ 18u1 ×104
+
15u2 ×103
− 56 ≤ 0
and another optimization algorithm u3 (u1 −2u4 )3 +2u2 u3 (4u24 +3u1 (u1 −2u4 )) (u1 −2u4 )u33 +2u3 u32
Where 10 ≤ h ≤ 80.10 ≤ u2 ≤ 50.0.9 ≤ u3 ≤ 5.u⇀ = (u1 .u2 .u3 .u4 )
Algorithm Optimum variables fmin This paper applies the QLGCTSA algorithm to
u1 u2 u3 u4 solve this problem and compares it with Cuckoo Search
Algorithm(CSA) [52], GWO [26], ARSM [84], and
QLGCTSA 80 50 1.76471 5 0.0071006
algorithm Improved ARSM [84] that shown in Table 16.
GWO [26] 80 42.8154 0.9 2.7179 0.0132020 The design variable's optimum values are 80, 50, 1.76471,
CSA [52] 80 50 0.9 2.3216715 0.0130747 5, and the optimum vertical deflection is 0.0071006.
ARSM [84] 80 37.05 1.71 2.31 0.0157000 Table 16 shows that the QLGCTSA algorithm solves I-beam
Improved 79.99 48.42 0.9 2.40 0.1310000 design optimization problems better than other optimization
ARSM [84] algorithms.

Table 17  Solving pressure Algorithm Optimum Variables Optimum Cost

vessel design problem with
QLGCTSA and another u1 u2 u3 u4
optimization algorithm
QLGCTSA algorithm 13.39411 7.075665 42.09845 176.636596 6059.7143
WOA [53] 0.812500 0.437500 42.09827 176.63899 6059.7410
GSA [66] 1.125000 0.625000 55.98866 84.454203 8538.8359
Evolution strategies [55] 0.812500 0.437500 42.098087 176.640518 6059.7456
PSO [69] 0.812500 0.437500 42.098411 176.637690 6059.7340

13
 F. S. Gharehchopogh
1200
4.2.6 Pressure Vessel Design Problem
This problem minimizes the total material, forming cost, and
single 60° welding cost [85]. The different components are
the diameter of the shell ­(u1), the diameter of the head ­(u2),
the internal radius ­(u3), and the length of the cylindrical part
of the vessel (u4). The analytical expression is as follows:
M i n i m i z e f6 ( u )   =   0 . 6 2 2 4
⇀
u1 u3 u4 + 1.7781u2 u23 + 3.1661u21 u4 + 19.84u21 u3
Subject to
g1(u) = −u1 + 0.0193u3 ≤ 0
⇀
g2(u) = )=−u2 + 0.00954u3 ≤ 0
⇀
g3(u)=−𝜋u23 u4 − 43 𝜋u33 + 1296000 ≤ 0
⇀
(u)=u4 + 240 ≤ 0 g4
⇀
Here, 1 ≤ u1 ≤ 99.1 ≤ u2 ≤ 99.10 ≤ u3 ≤ 200.10 ≤ u4
⇀
≤ 200; u = (u1 .u2 .u3 .u4 )
Table 17 compares the QLGCTSA algorithm with another
optimization algorithm. Experimental result shows that the
QLGCTSA algorithm obtained the best optimum results.
As shown in Table  17, the optimum values of the
QLGCTSA algorithm's decision parameters are 13.39411,
7.075665, 42.09845, 176.636596, and the cylindrical pres-
sure vessel's optimum cost 6059.7143. The QLGCTSA algo-
rithm is superior to another optimization algorithm in terms
of the best solution.
5 Conclusions and Future Work
Tunicate Swarm Algorithm (TSA) was used to solve some
engineering problems. This paper used three mutation opera-
tors to increase the TSA algorithm's global search capability
based on three different distributions: Gaussian mutation,
Cauchy mutation, and lévy flight. Then we evaluated the
impact of each mutation operator on the performance of the
QLGCTSA algorithm. We also assessed the TSA algorithm's
performance using mutation operators in a two-by-two form,
finally presenting an algorithm that applies all three muta-
tion operators for optimization operations. We called this
algorithm the QLGCTSA algorithm. To perform the evalu-
ations, 52 benchmark functions, including three different
UM, MM, and Cauchy mutation groups, were performed,
29 of which include CEC2017. The QLGCTSA algorithm
was also applied to evaluate the QLGCTSA algorithm's per-
formance of six engineering problems of large dimensions
presented in CEC 2017. An experiment was also conducted
to test the QLGCTSA algorithm's scalability benchmarks
of large and various sizes. Implementing comparative
algorithms indicates that the QLGCTSA algorithm out-
performs other competing optimization algorithms and can
be used in the future as a powerful and efficient algorithm
to solve various problems. The paper major considers the
efficiency of the QLGCTSA algorithm using six technical
13
implementation problems. Experimental results show that
the QLGCTSA algorithm has dramatically improved com-
pared with several optimization algorithms. In future work,
the QLGCTSA algorithm can solve different engineering
problems. We intend to use and evaluate the multi-objective
version of this algorithm to solve multi-objective issues con-
cerning excellent and significant performance.
Data availability  Dataset and Source code that support the findings
of this research will be shared when this manuscript is accepted for
publication, or the Editor/reviewer is requested to share.
Declarations 
Conflict of interest  The authors declare that they have no conflict of
interest.
References
1. Gharehchopogh, F.S., I. Maleki, & Z.A. Dizaji. (2021). Chaotic
vortex search algorithm: Metaheuristic algorithm for feature selec-
tion. Evolutionary Intelligence,1–32.
2. Nadimi-Shahraki, M. M., Banaie-Dezfouli, M., Zamani, H.,
Shokooh, T., & Seyedali, M. (2021). B-MFO: A binary moth-
flame optimization for feature selection from medical datasets.
Computers, 10(11), 136.
3. Benyamin, A., Farhad, S. G., & Saeid, B. (2021). Discrete farm-
land fertility optimization algorithm with metropolis acceptance
criterion for traveling salesman problems. International Journal
of Intelligent Systems, 36(3), 1270–1303.
4. Gharehchopogh, F. S., Farnad, B., & Alizadeh, A. (2021). A modi-
fied farmland fertility algorithm for solving constrained engineer-
ing problems. Concurrency and Computation: Practice and Expe-
rience, 33(17), e6310.
5. Houssein, E. H., Emam, M. M., & Ali, A. A. (2021). An efficient
multilevel thresholding segmentation method for thermography
breast cancer imaging based on improved chimp optimization
algorithm. Expert Systems with Applications, 185, 115651.
6. Furugi, A. (2021). A tabu search algorithm for the unrelated
parallel machine scheduling problem with machine availability
constraint and sequence-dependent setup time. Journal of the Fac-
ulty of Engineering and Architecture of Gazi University, 36(3),
1539–1550.
7. Houssein, E. H. (2021). Hybrid slime mould algorithm with adap-
tive guided differential evolution algorithm for combinatorial and
global optimization problems. Expert Systems with Applications,
2021(174), 114689.
8. Nadimi-Shahraki, M. H. (2020). MTDE: An effective multi-trial
vector-based differential evolution algorithm and its applications
for engineering design problems. Applied Soft Computing, 97,
106761.
9. Solimanpur, M., Foroughi, A., & Mohammadi, M. (2016). Opti-
mum route selection in hole-making operations using a dynamic
programming-based method. Cogent Engineering, 3(1), 1201991.
10. Furugi, A. (2022). Sequence-dependent time-and cost-oriented
assembly line balancing problems: A combinatorial Benders’
decomposition approach. Engineering Optimization, 54(1),
170–184.
An Improved Tunicate Swarm Algorithm with Best‑random Mutation Strategy for Global Optimization…
11. Abdollahzadeh, B., Barshandeh S., Javadi H., Epicoco N. (2021).
An enhanced binary slime mould algorithm for solving the 0–1
knapsack problem. Engineering with Computers, 1–22.
12. Essam, H. H., Bahaa, E., Diego, O., & Ahmed, A. E. S. (2021). A
novel black widow optimization algorithm for multilevel thresh-
olding image segmentation. Expert Systems with Applications.,
167, 114159.
13. Taghian, S., & Nadimi-Shahraki M.H. (2019). A binary
metaheuristic algorithm for wrapper feature selection. Interna-
tional Journal of Computer Science Engineering (IJCSE), 8(5),
168–172.
14. Houssein, E. H. (2020). Hybrid harris hawks optimization with
cuckoo search for drug design and discovery in chemoinformatics.
Scientific Reports, 10(1), 1–22.
15. Furugi, A. & Yapici, F. (2021). Optimization of production param-
eters in oriented strand board (osb) manufacturing by using tagu-
chi method. Wood Industry/Drvna Industrija, 72(4).
16. Gharehchopogh, F. S., Shayanfar, H., & Gholizadeh, H. (2020). A
comprehensive survey on symbiotic organisms search algorithms.
Artificial Intelligence Review, 53(3), 2265–2312.
17. Gharehchopogh, F.S. (2022). Advances in tree seed algorithm:
A comprehensive survey. Archives of Computational Methods in
Engineering, 1–24.
18. Ghafori, S., & Gharehchopogh F. S. (2021). Advances in spotted
hyena optimizer: A comprehensive survey. Archives of Compu-
tational Methods in Engineering, 1–22.
19. Gharehchopogh, F. S., & Gholizadeh H. (2019). A comprehensive
survey: Whale optimization algorithm and its applications. Swarm
and Evolutionary Computation, 48, 1–24.
20. Kaur, S., Lalit, K. A., Sangal, A. L., & Gaurav, D. (2020). Tuni-
cate swarm algorithm: a new bio-inspired based metaheuristic
paradigm for global optimization. Engineering Applications of
Artificial Intelligence, 90, 103541.
21. Zamani, H., & Nadimi-Shahraki, M. H. (2016). Feature selection
based on whale optimization algorithm for diseases diagnosis.
International Journal of Computer Science and Information Secu-
rity, 14(9), 1243.
22. Bansal, J. C. H., & S., Shimpi S. J. & Maurice C. (2014). Spi-
der monkey optimization algorithm for numerical optimization.
Memetic Computing, 6(1), 31–47.
23. Cheng, M.Y& Prayogo D. (2014). Symbiotic organisms search:
A new metaheuristic optimization algorithm. Computers & Struc-
tures, 139, 98–112.
24. Tilahun, N. H. S. L., Sathasivam, S., & Choon, O. H. (2013).
Prey-predator algorithm as a new optimization technique using in
radial basis function neural networks. Research Journal of Applied
Sciences, 8(7), 383–387.
25. Muthiah-Nakarajan, V., & Noel, M. M. (2016). Galactic swarm
optimization: A new global optimization metaheuristic inspired
by galactic motion. Applied Soft Computing, 38, 771–787.
26. Mirjalili, S., Mirjalili, S. M., & Lewis, A. (2014). Grey wolf opti-
mizer. Advances in Engineering Software, 69, 46–61.
27. Nadimi-Shahraki, M. H., Taghian, S., & Mirjalili, S. (2021). An
improved grey wolf optimizer for solving engineering problems.
Expert Systems with Applications, 166, 113917.
28. Mirjalili, S. (2016). SCA: A sine cosine algorithm for solving
optimization problems. Knowledge-Based Systems, 96, 120–133.
29. Dhiman, G., & Kumar, V. (2017). Spotted hyena optimizer: A
novel bio-inspired based metaheuristic technique for engineering
applications. Advances in Engineering Software, 114, 48–70.
30. Zamani, H., Nadimi-Shahraki, M. H., & Gandomi A.H. (2022).
Starling murmuration optimizer: A novel bio-inspired algorithm
for global and engineering optimization. Computer Methods in
Applied Mechanics and Engineering, 392, 114616.
31. Kirkpatrick, S., Gelatt, C. D., & Vecchi, M. P. (1983). Optimiza-
tion by simulated annealing. Science, 220(4598), 671–680.
1201
32. Simon, D. (2008). Biogeography-based optimization. IEEE Trans-
actions on Evolutionary Computation, 12(6), 702–713.
33. De Castro L.N., & Von Zuben. F.J. (2000). The clonal selection
algorithm with engineering applications. In Workshop on Artifi-
cial Immune Systems and Their Application, Las Vegas, USA.
34. Abdollahzadeh, B., Soleimanian, G. F., & Mirjalili, S. (2021).
Artificial gorilla troops optimizer: A new nature-inspired
metaheuristic algorithm for global optimization problems. Inter-
national Journal of Intelligent Systems, 36(10), 5887–5958.
35. Zamani, H., Nadimi-Shahraki, M. H., & Gandomi, A. H. (2019).
CCSA: Conscious neighborhood-based crow search algorithm for
solving global optimization problems. Applied Soft Computing,
85, 105583.
36. Shayanfar, H., & Gharehchopogh, F. S. (2018). Farmland fertility:
A new metaheuristic algorithm for solving continuous optimiza-
tion problems. Applied Soft Computing, 71, 728–746.
37. Zamani, H., Nadimi-Shahraki, M. H., & Gandomi, A. H. (2021).
QANA: Quantum-based avian navigation optimizer algorithm.
Engineering Applications of Artificial Intelligence, 104, 104314.
38. Rashedi, E., Nezamabadi-Pour, H., & Saryazdi, S. (2009). GSA:
A gravitational search algorithm. Information Sciences, 179(13),
2232–2248.
39. Abdollahzadeh, B., Gharehchopogh, F. S., & Mirjalili, S. (2021).
African vultures optimization algorithm: A new nature-inspired
metaheuristic algorithm for global optimization problems. Com-
puters & Industrial Engineering, 158, 107408.
40. Kashan, A. H. (2014). League championship algorithm (lca): An
algorithm for global optimization inspired by sport champion-
ships. Applied Soft Computing, 16, 171–200.
41. Fatma, A. H., Essam, H. H., Mai, S. M., Walid, A., & Mirjalili, S.
(2019). Henry gas solubility optimization: A novel physics-based
algorithm. Future Generation Computer Systems, 101, 646–667.
42. Eita, M. & Fahmy, M. (2010). Group counseling optimization:
A novel approach. In: Research and development in intelligent
systems XXVI. Springer, 195–208.
43. Shah-Hosseini, H. (2009). The intelligent water drops algorithm:
A nature-inspired swarm-based optimization algorithm. Interna-
tional Journal of Bio-Inspired Computation, 1(1–2), 71–79.
44. Hegen, X., Zhiyuan, W., Huali, F., Gongfa, L., & Guozhang, J.
(2018). Quantum rotation gate in quantum-inspired evolutionary
algorithm: A review, analysis and comparison study. Swarm and
Evolutionary Computation, 42, 43–57.
45. Salvador, G., Alberto, F., Julian, L., & Francisco, H. (2010).
Advanced nonparametric tests for multiple comparisons in the
design of experiments in computational intelligence and data
mining: Experimental analysis of power. Information sciences,
180(10), 2044–2064.
46. Alcala-Fdez, J., Sanchez, L., Garcia, S., Del, M. J., Ventura, S.,
Garrell, J. M., Otero, J., Romero, C., Bacardit, J., Rivas, V. M.,
Fernandez, J. C., & Herrera, F. (2009). KEEL: A software tool
to assess evolutionary algorithms for data mining problems. Soft
Computing, 13(3), 307–318.
47. Yao, X., Liu, Y., & Lin, G. (1999). Evolutionary programming
made faster. IEEE Transactions on Evolutionary computation,
3(2), 82–102.
48. Digalakis, J. G., & Margaritis, K. G. (2001). On benchmarking
functions for genetic algorithms. International Journal of Com-
puter Mathematics, 77(4), 481–506.
49. Heidari, A. A., Mirjalili, S., Hossam, F., Ibrahim, A., Majdi, M.,
& Huiling, C. (2019). Harris hawks optimization: Algorithm and
applications. Future Generation Computer Systems, 97, 849–872.
50. Mirjalili, S., Mirjalili, S. M., & Hatamlou, A. (2016). Multi-verse
optimizer: A nature-inspired algorithm for global optimization.
Neural Computing and Applications, 27(2), 495–513.
51. Eneko, O., Esther, V.R., Javier, D. S., Antonio, J.N., Daniel, M.,
Antonio, L.T., Ponnuthurai, N.S., Carlos, A.C.C., & Francisco, H.
13
 F. S. Gharehchopogh
1202
(2021). A tutorial on the design, experimentation and application
of metaheuristic algorithms to real-world optimization problems.
Swarm and Evolutionary Computation, 100888.
52. Gandomi, A. H., Yang, X. S., & Alavi, A. H. (2013). Cuckoo
search algorithm: A metaheuristic approach to solve structural
optimization problems. Engineering with Computers, 29(1),
17–35.
53. Mirjalili, S., & Lewis, A. (2016). The whale optimization algo-
rithm. Advances in Engineering Software, 95, 51–67.
54. He, Q., & Wang, L. (2007). An effective co-evolutionary particle
swarm optimization for constrained engineering design problems.
Engineering Applications of Artificial Intelligence, 20(1), 89–99.
55. Mezura-Montes, E., & Coello, C. A. C. (2008). An empirical
study about the usefulness of evolution strategies to solve con-
strained optimization problems. International Journal of General
Systems, 37(4), 443–473.
56. Gupta, S., Deep, K., Moayedi, H., Foong, L.K., & Assif, A.
(2020). Sine cosine grey wolf optimizer to solve engineering
design problems. Engineering with Computers, 1–27.
57. Ray, T., & Saini, P. (2001). Engineering design optimization using
a swarm with an intelligent information sharing among individu-
als. Engineering Optimization, 33(6), 735–748.
58. Akhtar, S., Tai, K., & Ray, T. (2002). A socio-behavioural simu-
lation model for engineering design optimization. Engineering
Optimization, 34(4), 341–354.
59. Mittal, N., Singh, U., & Sohi, B.S. (2016). Modified grey wolf
optimizer for global engineering optimization. Applied Compu-
tational Intelligence and Soft Computing, pp. 1–17.
60. Gupta, S., & Deep, K. (2019). A hybrid self-adaptive sine cosine
algorithm with opposition based learning. Expert Systems with
Applications, 119, 210–230.
61. Mirjalili, S. (2015). Moth-flame optimization algorithm: A novel
nature-inspired heuristic paradigm. Knowledge-Based Systems,
89, 228–249.
62. Mirjalili, S., Gandomi, A. H., Mirjalili, S. Z., Saremi, S., Faris, H.,
& Mirjalili, S. M. (2017). Salp swarm algorithm: A bio-inspired
optimizer for engineering design problems. Advances in Engineer-
ing Software, 114, 163–191.
63. Gupta, S., & Deep, K. (2019). Improved sine cosine algorithm
with crossover scheme for global optimization. Knowledge-Based
Systems, 165, 374–406.
64. Sayed, G. I., Khoriba, G., & Haggag, M. H. (2018). A novel cha-
otic salp swarm algorithm for global optimization and feature
selection. Applied Intelligence, 48(10), 3462–3481.
65. Chickermane, H., & Gea, H. C. (1996). Structural optimization
using a new local approximation method. International Journal
for Numerical Methods in Engineering, 39(5), 829–846.
66. Ashok, D. B., & Jasbir, S. A. (1985). A study of mathematical
programming methods for structural optimization: Part I: Theory.
Numerical Method in Engineering, 21(9), 1583–1599.
67. Mahdavi, M., Fesanghary, M., & Damangir E. (2007). An
improved harmony search algorithm for solving optimization
problems. Applied Mathematics and Computation, 188(2),
1567–1579.
68. Lee, K. S., & Geem, Z. W. (2005). A new meta-heuristic algo-
rithm for continuous engineering optimization: Harmony search
theory and practice. Computer Methods in Applied Mechanics and
Engineering, 194(36–38), 3902–3933.
69. Li, L. J., Huang, Z. B., Liu, F., & Wu, Q. H. (2007). A heuristic
particle swarm optimizer for optimization of pin connected struc-
tures. Computers & Structures, 85(7–8), 340–349.
13
70. Moosavi, S. H. S., & Bardsiri, V. K. (2019). Poor and rich opti-
mization algorithm: A new human-based and multi populations
algorithm. Engineering Applications of Artificial Intelligence, 86,
165–181.
71. Mirjalili, S. (2015). The ant lion optimizer. Advances in Engineer-
ing Software, 83, 80–98.
72. Zhang, M., Luo, W., & Wang, X. (2008). Differential evolution
with dynamic stochastic selection for constrained optimization.
Information Sciences, 178(15), 3043–3074.
73. Wang, C., & Liu, K. (2019). A randomly guided firefly algorithm
based on elitist strategy and its applications. IEEE Access, 7,
130373–130387.
74. Huiling, C., Yueting, X., Mingjing, W., & Xuehua, Z. (2019). A
balanced whale optimization algorithm for constrained engineer-
ing design problems. Applied Mathematical Modelling, 71, 45–59.
75. Chen, D., Ziqi, X., Ximeng, L., Yin, Y., Yang, Y., & Wenzhong,
G. (2019). Dual-search artificial bee colony algorithm for engi-
neering optimization. IEEE Access, 7, 24571–24584.
76. Saremi, S., Mirjalili, S., & Lewis, A. (2017). Grasshopper optimi-
zation algorithm: Theory and application. Advances in Engineer-
ing Software, 105, 30–47.
77. Liu, H., Cai, Z., & Wang, Y. (2010). Hybridizing particle swarm
optimization with differential evolution for constrained numeri-
cal and engineering optimization. Applied Soft Computing, 10(2),
629–640.
78. Tsai, J. F. (2005). Global optimization of nonlinear fractional pro-
gramming problems in engineering design. Engineering Optimi-
zation, 37(4), 399–409.
79. Ali, S., Ardeshir, B., Hadi, E., & Mohd, H. (2013). Mine blast
algorithm: A new population based algorithm for solving con-
strained engineering optimization problems. Applied Soft Comput-
ing, 13(5), 2592–2612.
80. Kamalinejad, M., Arzani, H., & Kaveh, A. (2019). Quantum
evolutionary algorithm with rotational gate and SS Hepsilon SS-
gate updating in real and integer domains for optimization. Acta
Mechanica, 230(8), 2937–2961.
81. Wang, C., & Chu X. (2019). An improved firefly algorithm with
specific probability and its engineering application. IEEE Access,
7, 57424–57439.
82. Ray, T., & Liew, K. M. (2003). Society and civilization: An opti-
mization algorithm based on the simulation of social behavior.
IEEE Transactions on Evolutionary Computation, 7(4), 386–396.
83. Kandikonda, H. R., Sharma, R. S., Mishra, G. S. A., & Patvard-
han, C. (2005). An evolutionary computational technique for
constrained optimisation in engineering design. Journal of the
Institution of Engineers India Part Me Mechanical Engineering
Division, 86, 121–128.
84. Wang, G. G. (2003). Adaptive response surface method using
inherited latin hypercube design points. Journal of Mechanical
Design, 125(2), 210–220.
85. Holland, J.H. (1992). Adaptation in natural and artificial systems:
an introductory analysis with applications to biology, control, and
artificial intelligence. MIT Press.
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.