Slime Mould Algorithm: A

comprehensive review of recent

variants and applications
Huiling Chen1*, Chenyang Li2, Majdi Mafarja3,4, Ali Asghar Heidari1, Yi Chen1, Zhennao Cai1

1 College of Computer Science and Artificial Intelligence, Wenzhou University, Wenzhou, Zhejiang 325035,
China
[email protected], [email protected], [email protected], [email protected]
2 College of Computer Science, Sichuan University, China
[email protected]
3Department of Computer Science, Birzeit University, POBox 14, West Bank, Palestine
[email protected]
4 Faculty of Computing and Information Technology, King Abdulaziz University, 21589 Jeddah, Saudi
Arabia

*Corresponding Author: Huiling Chen ([email protected])

Abstract:
The slime optimization algorithm (SMA) has recently received much attention from researchers
because of its simple structure, excellent optimization capabilities, and acceptable convergence in
dealing with various types of complex real-world problems. The increase in research enthusiasm has
led to the emergence of many different advanced versions of SMAs based on the existing ones, while
SMAs have been applied to deal with optimization challenges in different domains. Therefore, this study
aims to retrieve, identify, summarize and analyze critical studies related to SMA development. Based on
this, 98 SMA-related studies in the Web of Science were retrieved, selected, and identified. The two
main review vectors were advanced versions of SMAs and application domains. First, we counted and
analyzed various advanced versions of SMAs, summarized, classified, and discussed their improvement
methods and directions. Secondly, we sort out the application domains of SMA and analyze the role,
development status, and shortcomings of SMA in each domain. The survey results based on the existing
literature show that SMA is significantly better than several well-established metaheuristics in terms of
speed and accuracy in handling various benchmark problems and solving multiple real-world
optimization problems. Moreover, SMA can be quickly landed in different application domains due to
its simple structure and powerful scalability. SMA is still in its infancy, and there is still much research
work to be done. This review not only suggests possible future directions in the field but, due to the
inclusion of graphical and tabular comparisons of various properties, also provides future researchers
with a comprehensive source of information about SMA and advanced versions of SAMs and the
scope of adaptation for multiple application domains.
Keywords: Slime mould algorithm; Optimization; Swarm intelligence; Computational intelligence;
Meta heuristic algorithm

1. Introduction
Real-world optimization problems are characterized by complexity, nonlinearity, non-
differentiability, and large-scale variables, which are difficult to be solved effectively by traditional
mathematical optimization techniques [1, 2]. With the improvement of computer computing power, an
algorithm that simulates the cooperative and competitive behaviors of natural clusters of organisms
has received widespread attention, and this algorithm is called the swarm intelligence algorithm [3]. It
belongs to the same metaheuristic algorithm as the evolutionary algorithm, but it has a different
algorithmic structure from the one. Unlike evolutionary algorithms that rely on selection, crossover,
and variation structures, swarm intelligence algorithms rely on cooperation and competition among
individuals in a population, and classical ones such as particle swarm algorithms [4-7]. Algorithm
designers look for inspiration in the survival behavior of natural biological populations to expect to
design efficient operator mechanisms to improve the algorithm's optimality-seeking ability.
One of the latest swarm intelligence algorithms, the Slime Mould Algorithm (SMA), has been
receiving much attention from the community of researchers since it was proposed in 2020 [8].
Therefore, in this paper, SMA is extensively investigated for systematic evaluation and analysis. SMA is
inspired by the food-capturing behavior of the Physarum polycephalum (a type of slime mould). The
Physarum polycephalum covers multiple points in a fan-like manner to search for food and connect
with a venous network and form an optimal path. In addition, the venous network has a contraction
property, which is subject to oscillatory feedback from individual slime bacteria to adjust the area and
region covered. Because of the greedy mechanism of organisms, slime moulds prefer to cover areas
with high food concentrations. SMA simulates the behavioral mechanism of this organism. In its
mathematical description, the variable decision space of the optimization problem represents the
survival environment of slime moulds; the objective function is to evaluate the food concentration of
locations in the survival environment, and the global optimal solution represents the location with the
highest food concentration. SMA is a swarm intelligence algorithm where each candidate solution
represents a slime mould individual. Individuals are connected to each other by a positive and negative
feedback mechanism to establish a venous network for intra-group cooperation and competition.
Finally, in a limited time step, the population continuously searches for optimal food in the survival
environment. Mapping is represented in the optimization problem as the algorithm completes the
search for the optimal solution in iterations.
From the complexity science perspective, SMA is consistent with the five basic principles of
swarm intelligence. Millonas first introduced the concept of swarm intelligence in 1993 and proposed
five principles [9], which are the proximity principle, quality principle, diverse response principle,
stability principle, and adaptability principle. Among them, the proximity principle indicates that the
swarm must be able to achieve simple spatio-temporal computation, and SMA takes the decision space
as the absolute position and the mucilaginous individuals as the relative position to achieve the spatial
position sensing among the swarm. In addition, the update of the swarm position is accomplished
through an iterative step based on the exchange of individual information. The quality principle is that
the swarm can respond quickly to environmental elements. SMA takes the decision space as the
environment and the objective space as the response in the decision space, and SMA can realize the
swarm response in the decision space based on the assessment of the fitness value in the objective
space. The diversity response principle requires that the activity range of the swarm should not be too
narrow, and in SMA, it is expanded by conducting a uniform random search in the search space with a
certain probability. In addition, positive and negative feedback adaptive weights are used to achieve a
search for multiple regional points and maintain swarm diversity. The stability principle requires that
the swarm should not change its pattern with each environmental change. SMA uses positive and
negative feedback adaptive weights to regulate the position of individuals, and the weights are calculated
based on the degree of adaptation of individuals in the new environment, so in the new environment,
SMA will only regulate individuals with poor adaptation and keep individuals with good adaptation.
The adaptability principle requires that the computational model of the swarm should change when
needed, and the SMA can make positive and negative feedback adjustments toward multiple regions
with potentially optimal candidate solutions based on the changing environmental and current
conditions and location distribution of individuals. In summary, the design of SMA follows the five
principles of swarm intelligence, which can be demonstrated from the mathematical description in
Section 3.
SMA algorithms have been used to solve various complex optimization problems, such as
manufacturing design, engineering design, machine learning, deep learning, energy design, feature
selection, image segmentation, etc. In addition, SMA has received a lot of attention from researchers
because of its excellent performance, and a variety of advanced SMAs have been developed, which are
called advanced versions of SMAs in this paper. These advanced versions can be used for solving
single/multi-objective problems in discrete or continuous spaces and problems in search spaces of
unknown types. As of today, the research heat for SMA algorithms is still going on. However, to date,
there is no literature review of research on SMA. We believe that a comprehensive survey for SMA is
important and necessary to help provide a quick introduction for researchers who want to understand
the progress of SMA and also to provide ideas and inspiration for researchers studying SMA.
As a result, the objective of this study is to review, evaluate, and synthesize SMA-related studies.
This study is significant because it has the benefits of (1) identifying, categorizing, and analyzing the
applications of SMA in solving different optimization problems; (2) statistically analyzing variants and
versions of SMA and analyzing the mechanisms of various improved methods; (3) reviewing the
applications of SMA in different practical scenarios and conducting a thorough analytical study; (4)
analyzing the strengths and weaknesses of SMA; and (5) suggested research directions for SMA.
This paper is composed of the following parts. Section 2 presents an overview of the research
methods used in this survey and the different taxonomies of research topics in SMA. We briefly
introduce the basic components and concepts of the SMA in Section 3. This Section aims to familiarize
the reader with the traditional version's structure, logic, equations, and steps. Section 4 provides a
detailed classification of the advanced version of SMA, including single objective/multi-objective for
statistics. In Section 4, we summarize and analyze the development of SMA in different application
domains. In Section 5, the advantages and disadvantages of SMA are summarized and analyzed, and its
comprehensive evaluation is carried out. Section 6 concludes and suggests the future research direction
of SMA.

2. Research Methodology and SMA Survey Taxonomy

2.1 Research Methodology

We used relevant search terms for this paper collection, including swarm intelligence algorithms,
metaheuristic algorithms, multi-objective optimization, optimization algorithms, and evolutionary
algorithms. Then, these search terms were combined with “Slime Mould Algorithm” and “SMA”
(abbreviation for Slime Mould Algorithm) keywords to conduct an extensive search in Web of Science1.
Finally, we got a total of 471 papers. Removing duplicate papers and eliminating papers not related to
the slime algorithm, a total of 111 papers were obtained. Even so, 13 of these papers only use SMA as
the base comparison algorithm. Although mentioning SMA, we believe that these papers cannot be the
target of investigation in this paper because they do not investigate SMA in depth. Finally, based on
the remaining 98 papers, new taxonomies are further proposed for the study, as described in the next
subsection.

Figure 1. Classification of literature based on technical application areas

2.2 SMA Survey Taxonomy

The paper reviews 98 research works related to SMA that were accepted by the Web of Science in
2020-2022. For these 98 papers, we summarize and analyze them from two perspectives. Firstly, we
categorize the 98 papers according to SMA advanced versions, including the single-objective and multi-
objective versions. In addition, because advanced SMA has a rich research work in a single objective,
we subdivide it into three categories depending on the improved approach: mechanism-based, hybrid-
based, and discrete-based methods. Secondly, we analyze and summarize the 98 papers from the
perspective of application domains, counting the development trends of SMA in different application
domains. Based on the content of the 98 papers, there are six application domain categories: energy
optimization, engineering optimization, machine learning (traditional machine learning and deep
learning), image segmentation, scheduling optimization, and others. Energy optimization focuses on
building various models for different energy problems and using SMA to optimize solutions.

1
https://www.webofscience.com/
Engineering optimization [10] includes modeling and optimizing solutions for various industrial
manufacturing and engineering construction problems; machine learning represents the combination
of SMA and various machine learning methods (such as the traditional machine learning models [11]
and the deep learning models [12]) to perform optimization tasks such as feature extraction, parameter
optimization, and prediction; Image segmentation represents the combination of SMA and image
segmentation models to improve the segmentation accuracy of the color or black and white images;
scheduling optimization represents the use of SMA to solve scheduling optimization problems; Where
“Others” represent the use of SMA to optimize only the benchmark numerical functions to propose a
generally new and enhanced optimization algorithm. Figure 1 shows the number of categories in the
98 papers.

3. Overview of the SMA

The SMA is an optimization algorithm that simulates slime mould's diffusion and foraging behavior
[8]. Slime mold has the property of oscillating and contracting when it finds food during the gauge
feeding process. Meanwhile, a network of veins of varying thicknesses is formed between multiple food
sources, and the thickness of the network is related to the quality of the food source. SMA is based on
a biological oscillator and uses adaptive weights to simulate the positive and negative feedback process
generated by the "slime propagation wave" to obtain the optimal connection path of food with good
exploratory capabilities and exploitation tendencies. In addition, there is still a certain probability of
searching for unknown areas when acquiring a food source. Thus, the SMA consists of three main
components: approaching, wrapping, and acquiring food. The schematic diagram of the SMA is given
in Figure 2.

Figure 2. The schematic diagram of the SMA

3.1 Approaching food

The slime bacteria will determine the orientation of the food based on its concentration, which in
turn causes the biological oscillator to produce stronger waves. This contraction pattern close to the
food can be defined as Eq. (1).
 ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
⃗⃗⃗ ⋅ 𝑋
𝑋𝑏 (𝑡) + 𝑣𝑏 ⋅ (𝑊 ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝐴 (𝑡) − 𝑋𝐵 (𝑡)) , 𝑟 < 𝑝
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑋(𝑡 + 1) = { （1）
𝑣𝑐 ⋅ ⃗⃗⃗⃗⃗⃗⃗⃗
𝑋(𝑡) ,𝑟 ≥ 𝑝

where, 𝑋(𝑡) ⃗⃗⃗⃗⃗⃗⃗⃗ is the location vector of the slime bacteria at iteration time 𝑡, 𝑋
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑏 (𝑡) represents the position
vector of the individual with the highest concentration of odors ever found (the best solution so far),
and the two randomly selected individual location vectors from the population are noted as 𝑋 ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝐴 (𝑡) and
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑋𝐵 (𝑡)，𝑣𝑏 stands for a parameter between [−𝑎, 𝑎], 𝑣𝑐 is a parameter that decreases linearly from 1 to
0, and 𝑟 is a random number between 0 and 1 that follows a uniform distribution and changes with
each iteration. 𝑎 and 𝑝 can be defined using Eqs. (2) and (3), respectively.

𝑡
𝑎 = 𝑎𝑟𝑐𝑡𝑎𝑛ℎ (− (𝑀𝐴𝑋 _𝑡) + 1) （2）

⃗⃗⃗⃗⃗⃗⃗⃗ ) −𝐷𝐹|
𝑝 = 𝑡𝑎𝑛ℎ|𝑓(𝑋(𝑡) （3）

⃗⃗⃗⃗⃗⃗⃗⃗ ) denotes the fitness value of 𝑋(𝑡)

where, 𝑡 represents the current iteration number, 𝑓(𝑋(𝑡) ⃗⃗⃗⃗⃗⃗⃗⃗ calculated
by the objective function 𝑓(. ). 𝑀𝐴𝑋 _𝑡 is the maximum iteration number, and 𝐷𝐹 denotes the fitness
value of the optimal solution obtained in the current iteration. 𝑎𝑟𝑐𝑡𝑎𝑛ℎ(. ) denotes the inverse
hyperbolic function and 𝑡𝑎𝑛ℎ (. ) denotes the hyperbolic tangent function, 𝑊 ⃗⃗⃗ is the weight of the
slime mold, which calculation criterion is defined by Eq. (4).

𝑏𝐹−𝑆(ⅈ)
1 + 𝑟 ⋅ 𝑙𝑜𝑔 ( + 1) , 𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ = { 𝑏𝐹−𝑤𝐹
𝑊(𝑖) 𝑏𝐹−𝑆(ⅈ)
（4）
1 − 𝑟 ⋅ 𝑙𝑜𝑔 ( + 1) , 𝑜𝑡ℎ𝑒𝑟
𝑏𝐹−𝑤𝐹

where 𝑟 is a randomly selected number from the uniform distribution of (0, 1). 𝑆 denotes the set of
fitness values (calculated by the objective function) of the population, and 𝑆(𝑖) denotes the fitness value
of the 𝑖-th individual. 𝑆 is divided into two parts according to the ranking of fitness values, the top part
is called the superior set of individuals, and the bottom part is called the inferior set of individuals.
𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛, shown in Eq. (4), represents the 𝑊 ⃗⃗⃗ calculated using the set of dominant individuals;
conversely, 𝑜𝑡ℎ𝑒𝑟 represents the calculation using the set of inferior individuals. Moreover, 𝑏𝐹
represents the best currently obtained fitness value, 𝑤𝐹 denotes the worst fitness value of the current
iteration.

3.2 Wrapping food

The negative and positive feedbacks between the food concentration and the pulse width of the
slime mold are modeled as Eq. (1). In addition, the conditions simulate the method of finding food
selected by the slime mold by recognizing odors in the air. It is known to be given more weight when
the concentration of food is sufficient. In contrast, if there is a low concentration of food in the region,
the weight of that region will decrease, and the mucus will move on to other areas to continue its search.
Based on the above principle, the act of wrapping food is expressed by Eq. (5).
 𝑟𝑎𝑛𝑑 ⋅ (𝑈𝐵 − 𝐿𝐵) + 𝐿𝐵, 𝑟𝑎𝑛𝑑 < 𝑧
⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑋(𝑡 + 1) = { 𝑋 ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ ⃗⃗⃗ ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗ ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝑏 (𝑡) + 𝑣𝑏 ⋅ (𝑊 ⋅ 𝑋𝐴 (𝑡) − 𝑋𝐵 (𝑡)), 𝑟<𝑝 （5）
⃗⃗⃗⃗⃗⃗⃗⃗ ,
𝑣𝑐 ⋅ 𝑋(𝑡) 𝑟≥𝑝

where 𝑟𝑎𝑛𝑑 and 𝑟 are two randomly selected numbers from the uniform distribution of (0, 1), while 𝑧
is a parameter that takes values in the range of [0, 0.1] that the authors determined experimentally in
the basic SMA paper, a good balance between the exploitation and exploration phases can be obtained
when 𝑧 takes a value of 0.03. The same parameter settings are used in this chapter. 𝑈𝐵 and 𝐿𝑈
represent the upper and lower bounds of the decision space, respectively. Eq. (5) is based on Eq. (1)
with the addition of a random detection component, and the slime mold achieves positional adjustment
under the influence of positive and negative feedback through the weighting factor according to the
food concentration. Nevertheless, in reality, there will still be some slime mold that will tend to explore
the area randomly. Therefore, Eq. (5) simulates all the tendency behaviors of slime mold containing
exploration and exploitation.

3.3 Parameters and Principles

⃗⃗⃗ , 𝑣𝑏 and 𝜈𝑐 are used to model the variation in pulse width of the slime mold. 𝑊
𝑊 ⃗⃗⃗ updates the
rate of food by identifying the quality of the food, and simulating the oscillatory frequency of the slime,
thus helping the slime to select the best food source. 𝑣𝑏 and 𝜈𝑐 oscillate randomly in a certain range.
Where 𝑣𝑏 is distributed between [−𝑎, 𝑎]，𝜈𝑐 is distributed between [-1, 1]; then, as the iteration
proceeds, they eventually approach zero. Furthermore, 𝑣𝑏 simulates whether the slime molds would
choose to seek out or approach other food sources when a new one is discovered. The slime molds
separate some organic matter from other places in search of good food, even if a better food source is
found. This behavior provided more possibilities for slime molds to find higher quality food, effectively
improving the local optimal problem. All of these parameters are adaptively adjusted with the iterative
process. However, there is a hyperparameter 𝑧 in SMA that needs to be set in advance. The 𝑧, as shown
in Eq. (5), controls how often the algorithm samples using a uniform random distribution. The larger
the value of 𝑧, the more the algorithm tends to explore randomly. In addition, this random exploration
is also very necessary to increase the diversity of the population and has the chance to help the
population escape the local optimum trap. Therefore, it is very important to control the size of 𝑧. The
SMA paper [8] demonstrates through numerous experiments that 𝑧 = 0.03 is possible for the
population to reach a balance between exploration and exploitation. The pseudo-code of SMA is shown
in the following algorithm 1.

Algorithm 1: SMA’s pseudo-code

Input: The objective function 𝑓(. ); Maximum iteration 𝑀𝐴𝑋_𝑡
Output: 𝒃𝑭, 𝑿𝒃
Initialization of slime mold population: 𝑋ⅈ (𝑖 = 1,2, … , 𝑛);
𝑡 = 0;
While (𝑡 < 𝑀𝐴𝑋 _𝑡)
Compute the fitness of each slime mould (𝑋ⅈ (𝑖 = 1,2, … , 𝑛)) by objective function 𝑓(. );
Update best fitness b𝐹 ad best position 𝑋𝑏 ;
Calculate 𝑊⃗⃗⃗ according to Eq. (4);
For 𝑖 = 1,2, ⋯ , 𝑛(each slime mould)
 Update 𝑝, 𝑤𝑏 and 𝑣𝑐;
Update new location for 𝑋ⅈ according to Eq. (5);
End For
t = t + 1;
End while
Return the b𝐹, 𝑋𝑏 ;
End

4. Advanced version of SMA

The SMA, first proposed in 2020 [8], is the basic SMA version. Researchers modified SMA and
improved its performance to produce more new algorithms called the advanced version of SMA. The
basic version of SMA is proposed to be mainly applied to continuous single-objective optimization
problems. However, due to the diversity and complexity of real-world optimization problems, the basic
version of SMA is difficult to deal with various optimization challenges effectively and comprehensively.
However, considering the excellent performance of SMA in single-objective optimization problems and
its unique algorithmic framework structure, researchers have derived various excellent advanced
versions of SMA based on it. We divided these advanced versions of SMA into three categories on the
existing survey literature: single-objective continuous version, multi-objective continuous version, and
others. The reasons for this division are mainly based on three considerations: (1) the differences
between the categories should be large, (2) the categories should cover a wide range, and (3) the
categories should be clear and simple enough.
Based on the above classification basis, we classified the existing 98 surveyed papers into categories.
Figure 3 shows the proportion of each category in the total literature. In particular, it is important to
note that the advanced version of SMA indicates that the researcher improves algorithm performance
and solves various optimization challenges by adding, removing, or modifying algorithm operators or
structures. The “Other” categories in Figure 3 are not what we need to discuss because they do not
modify or enhance the basic SMA in any way but directly use the algorithm to solve hard problems in
different domains. Of course, this approach is encouraged and has an important role in advancing SMA.
Section 4 will also discuss the existing literature from the application domain perspective.

Figure 3. Classification of the advanced version of SMA

 In Figure 3, we classify the available advanced versions of SMA. Among them, 55 improvements
are single-objective versions, and 7 are multi-objective versions. In addition, the remaining 36 of the 98
papers are not improved SMA versions but directly use SMA to solve various domain problems, which
will be discussed in Section 4. Therefore, we cannot call this literature variant of SMA.

4.1 Single-objective version

Many researchers have worked on improving the optimization performance of SMA on single-
objective problems, and it can be seen from Figure 3 that more than 56% of all the advanced versions
of SMA in the surveyed literature fall under this category. Therefore, we need to further subdivide this
category to present a more apparent lineage to provide a more meaningful reference for the reader.
After organizing the existing literature, we divide the single-objective version of SMA into three classes
from the perspective of algorithm improvement. They are mechanism-based approaches, hybrid-based
methods, and discrete-based methods, respectively. Among them, mechanism-based methods use
various mechanisms to create operators to improve SMA. The hybrid-based approach is to mix SMA
with other metaheuristic algorithms or traditional mathematical optimization algorithms to achieve the
complementary strengths between multiple algorithms. The discrete-based methods discretize SMA
and add discrete mechanisms to improve its ability to solve discrete problems. Figure 4 shows the
amount of literature on these three categories, and in general, researchers are currently working more
on mechanism-based methods (65%) than hybrid-based methods (27.3%). Next, we describe these two
approaches in detail.

Figure 4. Classification of advanced SMA's single objective versions

4.1.1 Mechanisms-based methods

A survey of the existing literature shows that researchers prefer mechanism-based methods to
improve SMA. These mechanisms are operators constructed from different knowledge theories that
have different properties and capabilities according to their knowledge properties, thus compensating
for the inherent defects of the original algorithm. There are two key factors in this approach: first,
whether the properties and roles of the developed mechanism-based operators are suitable for the
algorithm; and second, when to invoke the operators to maximize their performance during the iterative
process of the algorithm. These two critical factors determine whether the mechanism-based approach
can improve the performance of SMA. Both inappropriate operators and untimely invocations can lead
to degenerate improvements in the algorithm.

Table 1. A brief summary of the relevant modified SMA variants

No. Improved methods References
Izci et al. [13], Houssein et al. [14], AlRassas
et al. [15], Yu et al. [16] ,Wazery et al. [17],
1 Opposition-based learning
Vashishtha et al. [18] ,Naik et al. [19] ,
Abualigah et al. [20]
Singh et al. [21], Rizk-Allah et al. [22], Xiao et
2 Chaotic mechanism al. [23], Tang et al. [24], Altay et al. [25] ,Chen
et al. [26]
Yin et al. [27], Gao et al. [28], Alfadhli et al.
3 Multiple swarm
[29]
4 Levy flight operator He et al. [30], Lin et al. [31], Anter et al. [32]
5 Mutation factors Yin et al. [33], Shi et al. [34]
6 Neighborhood dimensional learning Farhat et al. [35]
7 Diffusion mechanism Zhao et al. [36]
8 Quantum computing strategies Zhang et al. [37] , Yu et al. [38]
9 Nelder-Mead simplex mechanism Weng et al. [39], Liu et al. [40]
10 Brownian motion Sun et al. [41]
11 Information feedback model Precup et al. [42], [43]
12 Fractional calculus Ibrahim et al. [44]
13 Quasi-reflection Biswal et al. [45]
14 Others Wei et al. [46], Othman et al. [47], Jia et al. [48]

We surveyed the existing literature to count the mechanisms used to create operators for the
existing SMAs improved by mechanism-based methods. Table 1 shows the names of these mechanisms
and the literature on SMA improvements using this mechanism. As can be seen from the table,
researchers are more likely to use opposition-based learning (OBL) mechanisms to improve the base
version of the SMA. This mechanism was first proposed in 2015 [49]. It is an effective method to
increase population diversity. OBL usually obtains a new set of reverse solutions after reversing the
existing solutions in the search space. Eq. (6) shows the general mathematical description of OBL.

𝑥̌ = 𝑘(𝑢𝑏 − 𝑙𝑏) − 𝑥 (6)

where 𝑥 is a known vector and 𝑥̌ represents the inverse vector; 𝑢𝑏 and 𝑙𝑏 are the upper and lower
bounds of the search space, respectively; and 𝑘 is a random number between 0 and 1, which is
responsible for perturbing the position of 𝑥̌. Of course, researchers can arbitrarily modify the value of
𝑘 to achieve the reverse effect they want.
According to probabilistic principles, the inverse solution has a 50% probability of approaching
the location of the best solution in the search space. OBL can significantly increase the diversity of the
population because the inverse solution and the existing solution have different variable characteristics.
In the existing literature, eight papers attempt to improve the performance of SMA using OBL to
construct operators. In addition, the call time of the operator is also critical. From the findings, OBL
can help SMA to generate more diverse initialized populations and also increase the diversity of
populations by using OBL during the iteration of the algorithm. Vashishtha et al. [18] proposed an
OBL-based SMA algorithm, which mainly uses OBL to obtain initialized populations with diversity to
improve the performance of SMA. Yu et al. [16] use OBL to increase the diversity of the population
both in the initialization phase and in the iterative process of the algorithm. AlRassas et al. [15] added
the OBL method to SMA in each iteration to enhance the diversity and avoid falling into local optimum.
Wazery et al. [17] integrated OBL with SMA in each iteration to form ISMA with higher global search
performance. However, OBL is called in every iteration, which obviously adds additional computational
resources. Abualigah et al. [20] and Naik et al. [19] use an adaptive approach to call OBL, i.e., they
choose to call OBL to introduce population diversity information when a decrease in population
diversity is detected or when they fall into a local optimum late in the iteration. In addition, OBL also
tends to hybridize or cooperate with other mechanisms to generate new operators to enhance the
comprehensive performance of SMA. Houssein et al. [14] did some work to improve SMA by
introducing an orthogonal learning strategy and opposition-based learning mechanism into the SMA
to improve the convergence rate. Izci et al. [13] proposed an opposition-based learning mechanism
and simplex strategy into SMA to strengthen the trend of coordination between global exploration and
local exploitation.
In Table 1, the chaos mechanism is used second only to OBL. Chaos [50] is a disorder that exists
in nonlinear dynamic systems and is widely distributed in nature and social phenomena. Chaos is
characterized by pseudo-randomness, ergodicity, and sensitivity to initial values. In advanced versions
of SMA, researchers usually use a chaotic mechanism to generate a set of pseudo-random numbers to
chaotically perturb the evolutionary process of the algorithm, thereby increasing the traversal search
capability of SMA in a local region. Singh et al. [21] use the chaos mechanism to develop a random
search operator to replace the uniform random search in SMA to improve the search range of the
algorithm in the search space by using the traversal property of chaos.
In the same way, Rizk-Allah et al. [22] and Xiao et al. [23] use chaotic mechanisms to develop a
random search operator to enhance the global search capability of SMA. Besides, integrating chaotic
mechanisms and other mechanisms is also an essential research direction. Tang et al. [24] hybrid chaotic
mechanism and OBL were developed to enhance the diversity of the populations, showing that the
proposed new SMA version had higher diversity and did not easily fall into local optima. Chen et al. [26]
developed a chaotic SMA, showing that chaotic and opposition-based learning mechanisms help
improve SMA. The chaos mechanism has a unique role in improving the search performance of the
SMA algorithm due to its traversal property. Altay et al. [25] applied 10 different chaotic mappings,
generating chaotic values rather than random ones to improve SMA.
The multiple swarm (MS) mechanism and levy flight (LF) mechanism are also widely adopted by
researchers in multiple advanced SMA versions. MC is the algorithm that splits multiple populations in
the iterative evolution process and realizes the communication and cooperation among multiple
populations to improve the algorithm's local mining ability to ensure diversity. Yin et al. [27] presented
an enhanced SMA, which applied dominant groups to improve the convergence speed of the
exploitation capability. Gao et al. [28] designed an improved SMA using MS, which had better
classification ability and stability. Alfadhli et al. [29] proposed a novel SMA version that can adaptively
change the size of the population and illustrated that the method had a stronger optimization
performance. LF mechanism refers to a random walk with a heavy-tailed probability distribution of
step lengths, which means that there is a high probability that a large stride will occur during the random
walk. Using LF to generate pseudo-random numbers to be added to SMA can have a certain probability
of appearing large stride perturbation, thus improving the ability of the algorithm to jump out of the
local optimum. He et al. [30] presented a boosted SMA based on LF, which boosted the parameter
optimization capability of SMA. Lin et al. [31] brought up an enhanced SMA, called ESMA, where the
LF and OBL mechanisms were introduced to strengthen the trend of coordination between local
exploitation and global exploration. An improved SMA based on LF was proposed by Anter et al. [32].
In addition to the mechanisms mentioned above, the researchers also utilized mutation factors
[33][34], Neighborhood dimensional learning [35], Diffusion mechanism [36], Quantum computing
[37][38], Nelder-Mead simplex [39][40], Brownian motion [41], Information feedback model [42],
Fractional calculus [44] and Quasi-reflection [45] to develop operators with different characteristics
to improve the SMA algorithm. The mechanism-based methods are an important research
direction to improve SMA, and their unique advantage is that they can improve the performance
of SMA by adding and replacing operators without modifying the algorithmic framework.
Therefore, this approach can maximize the unique advantages of SMA and reduce its disadvantages
on specific problems.

4.1.2 Hybrid-based methods

Hybrid-based methods combine two metaheuristic algorithms using some fusion mechanism
to form a new algorithm. This approach can achieve complementary strengths between the two
algorithms and has efficient performance in solving problems with complex optimization.
Researchers have also used hybrid-based methods in their studies for SMA, and Table 2 lists the
algorithms that have been hybridized with SMA. Abdel-Basset et al. [45] integrated the whale
optimization algorithm (WOA) with SMA in 2020, the first work to improve SMA using a hybrid-
based method. It is an essential tool for modifying an algorithm by mixing other algorithms and
forming a new one.

Table 2. The brief description of the relevant hybridized SMA variants

No. Hybrid algorithms References
1 Fruit fly optimization algorithm Zhang et al. [51]
2 Marine predator algorithm Yousri et al. [52]
3 Salp swarm algorithm Wang et al. [53]
4 Chimpanzee optimization algorithm Ponlatha et al. [54]
5 Equilibrium optimizer Naik et al. [55]
6 Modified differential evolution Liu et al. [56]
7 Pattern search strategy Krishna et al. [57]
8 Adaptive guided differential evolution algorithm Houssein et al. [58]
9 Simulated annealing algorithm Gures et al. [59]
10 Firefly algorithm Ewees et al.[60]
11 Particle swarm optimizer Al-qaness et al. [61]
12 Arithmetic optimizer algorithm Chen et al. [62]
13 Algorithmic optimization algorithm Chauhan et al. [63]
 14 Whale optimization algorithm Bhandakkar et al. [64],
Abdel-Basset et al. [65]

The hybrid-based methods are classified into algorithmic framework fusion and operator fusion
according to the depth and implementation form of the two algorithms. Algorithmic framework fusion
indicates that two algorithms are used simultaneously to solve the same optimization problem, and the
populations generated by the two algorithms exchange information during the iteration of the
algorithms. This approach has a simple structure, where the algorithms operate independently of each
other but then allow for inter-population information replacement, taking advantage of multiple
algorithms to complete the solution. Zhang et al. [51] combined the fruit fly optimization algorithm
with SMA to develop a hybrid swarm intelligence algorithm with high development efficiency and
stability. Yousri et al. [52] presented a new algorithm combining the marine predator algorithm and
SMA, improving the ability to handle multi-modal and multi-dimensional problems. Ponlatha et al. [54]
presented an enhanced form of SMA by combining SMA with a chimpanzee optimization algorithm,
which optimization performance was more robust than the basic SMA. Krishna et al. [57] proposed a
new hybrid algorithm by hybridizing SMA and pattern search algorithm, and the experimental results
show high search efficiency in a complex search environment. Gures et al. [59] use SMA and simulated
annealing algorithms to solve complex optimization problems. Al-qaness et al. [61] use the particle
swarm algorithm to obtain a high-quality initialized population, and subsequently, SMA is further
optimized on this population to improve the accuracy of the optimization.
Operator fusion represents the combination of different operators of two algorithms to
accomplish a deep fusion between the algorithms. Wang et al. [53] combined SMA with a salp swarm
algorithm and introduced Levy flight strategy and adversarial learning methods into SMA to develop a
hybrid SMA. Naik et al. [55] introduced the equilibrium optimizer concept into SMA, giving SMA a
more vital ability to jump out of the local optimum. Liu et al. [56] use SMA as an operator to enhance
the optimization performance of the different algorithms and fully exploit the ability of multi-point
local search of the sticky bacterium algorithm. Houssein et al. [58] use adaptive guided differential
evolution algorithm's mutation operator to enhance swarm agents' local search, increase the
population's diversity, and help avoid premature convergence. Ewees et al. [60] use the firefly algorithm
(FA) to improve the exploration operator of SMA. This is because the FA algorithm can discover
feasible regions of potential optimal solutions, and the combination of the two can improve the
algorithm's convergence. Chen et al. [62] introduce the position-updating mechanism of individuals in
the artificial bee colony algorithm into the SMA algorithm to realize the complementary advantages of
the two algorithms. Chauhan et al. [63] presented an advanced SMA variant by combining the
algorithmic optimization method with SMA and introducing a learning strategy of lensing
confrontation, which had significantly superior performance compared with other metaheuristics.
Bhandakkar et al. [64] combined the location update mechanism of the whale optimization algorithm
with SMA for a more substantial convergence capability.
The hybrid-based methods can effectively improve the feasibility of SMA on various complex
optimization problems and can effectively take advantage of different algorithms in different
optimization scenarios. The premise of this approach is that the algorithms should have strong
scalability, which is an advantage of SMA from the results. In addition, operator fusion is more
advantageous than algorithmic framework fusion because it can take advantage of the advantageous
properties of operators in different algorithms to recombine into a new hybrid algorithm while reducing
computational resources. However, operator fusion is also tough, which requires the researcher to have
sufficient knowledge of both the algorithm and the mathematical properties of the problem. In general,
both hybrid-based and mechanism-based methods are promising research directions in the study of
improving SMA.

4.1.3 Discrete-based methods

Discrete problems are a very important branch of research in optimization. Unlike continuous
problems, the decision space of discrete problems is discretely distributed, and a small change in a single
dimension of the solution may lead to large fluctuations in the solution in the objective space. Therefore,
for SMA to be useful in discrete problems, its algorithmic framework and encoding methods need to
be improved. At present, researchers have developed several discrete versions of SMA. While keeping
the basic mathematical principles of SMA unchanged, researchers have mainly modified the original
SMA in terms of encoding methods by mapping or transforming the continuous values of each
dimension of the candidate solution into binary values.
Among them, the binary encoding binarizes the candidate solutions without changing the
algorithmic framework of the base version of SMA and modifies the operator formulas to
accommodate the operations on binary data. Zhang et al. [66] designed a discrete SMA for the feature
selection of large-scale data by the binary encoding candidate solutions. Abdollahzadeh et al. [67]
proposed a binary version of SMA by fusing Gaussian perturbation in SMA to solve the famous 0-1
backpack problem. Abdel-Basset et al. [68] developed a discrete version of SMA specifically to solve
the multidimensional multipack problem, using transfer functions to map continuous values to 0 or 1.
Abdel-Basset et al. [69] presented four discrete versions of SMA, in which the first version converted
the standard SMA, the second version integrated SMA with two-phase mutation (TM) to develop better
solutions near the present optimum, the third one combined SMA with a new attacking-feeding strategy
(AF) that strengthen the trend of coordination between global exploration and local exploitation based
on the memory saving per particle, and the fourth one combined TM and AF with SMA to produce a
better solution.
Because of the discrete nature of the decision space, discrete problems often have more complex
properties, such as non-integrability and non-derivability, which are difficult to solve effectively by
traditional mathematical methods. In addition, discrete processing of continuous problems sometimes
has strong advantages, such as increasing the algorithm's speed and reducing the computational memory.
Therefore, it is essential to develop discrete versions of SMA. There are still several problems with the
discrete version of SMA; the first one is that most researchers convert continuous SMA to discrete
SMA only by converting the continuous values of the candidate solutions to discrete binary values
through the conversion function and do not discretize the formula of SMA. Fundamentally, this
method is simple but inefficient. Because continuous space and discrete space are fundamentally
different, the local region in continuous space can be obtained by small perturbations in multiple
dimensions of the candidate solution. In contrast, the local region in discrete space is difficult to define,
and in discrete space, a small perturbation in one dimension of the candidate solution can lead to a
large change in the target value.
4.2 Multi‑objective variants

In contrast to traditional single-objective optimization, multiple objectives must be achieved in a

scenario for multi-objective optimization. It is not easy to have a single optimal solution because of the
tendency for goals to be inherently conflicted, with the optimization of one goal at the expense of the
reduction of others. Instead, they trade off and coordinate to make the overall objective the best
possible. Therefore, multi-objective problem optimization is a challenging task in the optimization field.
The excellent performance of SMA on single-objective optimization has led many researchers to
try to improve SMA to solve multi-objective problems. Houssein et al. [70] proposed a multi-objective
SMA (MOSMA), and the results showed that MOSMA provided better solutions in Pareto set proximity
(PSP) and decision space inverse generation distance (IGDX) metrics. Premkumar et al. [71] presented
the multi-objective SMA (MOSMA), and the proposed method was experimentally shown to have good
performance in solving linear, nonlinear, continuous, and discrete Pareto optimal frontier problems.
Khunkitti et al. [72] proposed that multi-objective SMA can solve optimal power flow cases and
efficiently achieve the Pareto fronts. Based on the Pareto dominance concept and fuzzy decision-making,
Hassan et al. [73] designed an improved SMA (ISMA) and developed a multi-objective ISMA
(MOISMA), in which the positions of the solution were updated according to two equations of the
sine cosine algorithm (SCA) to achieve the best solution. Gush et al. [74] applied SMA to solve a multi-
objective mixed-integer nonlinear optimization problem and maximize the PVHC while minimizing the
voltage deviation.
It can be seen that more and more researchers are using SMA for solving multi-objective
optimization problems. Researchers' experiments show that SMA has better search multi-directionality
and globalization than similar algorithms and is more capable of quickly converging to the Pareto front.
Therefore, using SMA for solving multi-objective optimization problems is a promising research
direction. There is still a lack of research for multi-objective SMA, and more research are needed.

4.3 Discussion

In this Section, we talk about advanced versions of SMA, and by further sorting, we divide this
surveyed literature into single-objective and multi-objective. In addition, since the single-objective
version of advanced SMA has many research works, we subdivide it into hybrid-based, mechanism-
based, and discrete-based methods. In general, the surveyed literature demonstrates that SMA has a
simple structure and strong scalability, which are its main advantages. The basic SMA has been shown
to perform better than particle swarm and difference algorithms, with powerful multipoint local search
capabilities[8]. However, the basic SMA still has many shortcomings. First, SMA does not provide multi-
objective and discrete versions, which is the main reason why it is limited in several fields; second, the
basic SMA prefers to exploit the exploitation ability and reduce the exploration ability in the choice of
balance; thus, SMA can easily fall into local optimum in the face of complex multimodal environments.
Researchers have developed various advanced versions of SMAs to address these shortcomings to solve
various complex optimization problems. These research findings help us advance the research progress
of SMA, but some problems still need to be further addressed.
⚫ Although SMA has a wide range of applications, few scholars have consistently studied the same
SMA variants in depth.
⚫ The research on SMA improvement lacks a unified performance testbed, and the simulation
platforms of many research results differ at the hardware and software environment levels, and
the choice of benchmark function also varies. Although the performance can be quantified
numerically through metrics such as mean and standard deviation, the inherently stochastic nature
of a class of algorithms such as SMA makes it difficult to directly compare the performance of
different variants through the data in the paper without the source code and with different
hardware and software, which hinders the further development of SMA.
⚫ Although there are numerous variants of SMA, the lack of directionality confuses users. For
example, it is difficult to determine which class of problems different versions of SMA are suitable
for solving, whether existing SMA variants need performance improvements when applied to
specific problems, and determine the strengths and weaknesses of different SMA versions and
how their corresponding conclusions are justified.
⚫ The rationale for SMA performance improvement cannot be explained from the perspective of
mathematical argumentation, and the key factors affecting SMA performance cannot be
pinpointed. Although adequate experiments can verify the performance of the corresponding
SMA version, introducing a mechanism often impacts properties such as diversity and convergence.
Thus, it is not easy to define the role of the introduced mechanism, so the performance verification
phase of SMA variants still needs some innovative measures.
⚫ SMA and its variants have limitations in terms of problem size. Since SMA and its variants need
to sample the objective function during the search process, this requires calling the objective
function several times in one iteration, implying that the solving time of SMA and its variants
increases accordingly as the number of decision variables increases, leading to a solution
bottleneck soon.
⚫ The optimization process of SMA lacks a unified tradeoff standard. Many kinds of literature
mention the tradeoff problem in the optimization process; however, they do not specify what kind
of exploration and exploitation ratio is considered to have achieved the tradeoff; rather, the
algorithm is verified to be in tradeoff status by the improvement of the benchmark function
seeking effect.
⚫ The SMA performance verification metrics need further improvement. The validation of the SMA
variant in a large number of works of literature uses the Friedman test and the Wilcoxon signed-
rank test but ignores the CPU time consumption of the SMA variant. Controlling the CPU time
will help the algorithm to solve the large-scale problem better. Therefore, adding a certain weight
to the CPU time is more reasonable to evaluate the algorithm performance comprehensively.
⚫ Multi-objective problems are more challenging than single-objective problems, which is a
promising research direction for researchers to pay more attention to the application of SMA on
multi-objective domains.
Researchers' joint efforts are needed to promote the further development of the SMA community
from the root. In response to the gap in research content above, one is to provide a new research
direction for researchers, and the other is to hope that more research will focus on these challenges,
which can significantly promote the development of SMA if these problems can be broken through.

5. Application domains of the SMA

Because of its excellent performance, SMA has been used to solve optimization challenges in a
wide range of industries. We conduct a comprehensive survey on the application domains of SMA on
the existing survey literature. A classifying the application domains of SMA, based on the 98 surveyed
papers available, is presented in Figure 5, in which SMA research at this stage is focused on specialized
problems such as energy optimization, engineering optimization, and machine learning optimization.

Detailed classification of SMA in application domains

Scheduling
Optimization (4)

Engineering
Optimization (27)

Machine Learning
Optimization (18)

Deep Learning
Optimization (5)

Energy
Optimization (33)

Image Segmentation (9)

Othes (2)

Figure 5. Detailed classification results of the survey literature in application domains.

In general, SMA is used in a wide range of industries for optimization tasks. These include the
energy industry, manufacturing, healthcare, IT services, water transport, agriculture and forestry,
education, finance, and other related industries. Of the 98 SMA-related studies collected in this article,
the energy industry has 37 works, including a variety of work types such as photovoltaic modeling,
optimization of power system models, thermal generators, optimal tidal optimization of distribution
systems, passive mitigation of induced voltages, coordinated scheduling and wind power forecasting;
the manufacturing industry has 16 works, specifically involving automotive components, route planning
for robots, model parameter optimization for controllers, defect identification for rolling bearings,
tower base systems, construction of fuel cell models, optimization of drive elements for optical machine
scanning systems and blind source separation; 13 works in the construction industry, including
identifying soil and classical engineering design cases for railway road substrates; 8 works related to the
medical industry, including the current hot topic of predicting the severity of pneumonia in new
pavilions, and diagnostic work for breast cancer, lung disease and there are four projects in the field of
information service technology, ranging from urban water demand and air quality to next generation
6G applications, and one each in the water transport, agriculture and forestry, education and finance
sectors. In addition, SMA has played a full role in industrial scheduling problems, knapsack problems,
engineering design, and parameter optimization of machine learning models. More details are shown in
the following Sections.

5.1 Scheduling optimization

The scheduling model optimization in position 1 in Figure 5 demonstrates SMA's application in

machine scheduling. The scheduling problem, also known as the combinatorial optimization problem,
consists of an objective function, constraint functions, decision variables, and a range of values. The
mathematical expression is shown in Eq. (7).
max 𝑓(𝑥)
𝑠. 𝑡. 𝑔(𝑥) ≥ 0
𝑥∈𝐷 (7)
where, 𝑓(𝑥) means the objective function, 𝑔(𝑥) means the constraint function, 𝑥 is a decision variable
which range of values is denoted by 𝐷, and it is a finite set of points. If 𝐹 = {𝑥| 𝑥 ∈ 𝐷, 𝑔(𝑥) ≥ 0}, A
combinatorial optimization problem can be represented by a triple (𝐷, 𝐹, 𝑓).
There are a large number of problems of this type with a finite number of feasible solutions,
which could theoretically be achieved by enumeration [75]. However, the scale of practical engineering,
in general, is large, and the number of feasible solutions is enormous, so there is still a need to devise
more efficient ways of using existing knowledge to search the problem space efficiently and thus achieve
the optimum [76-78]. The swarm intelligence algorithm is widely used in this field by virtue of its
superior exploration ability. As a novel algorithm, the application value of SMA has also been explored
by relevant researchers in this field. Generally, a slime individual in SMA is used to represent a feasible
solution, and SMA will give a set of feasible solutions after iteration for the decision-maker to choose
from. SMA and its modified version have been widely used for complex applications, including
communication networks, autonomous mobile robots, and backpack problems.
Specifically, next-generation 6G communication networks are fully connected and covered by user
devices, which will require a shift beyond traditional two-dimensional coverage services to full three-
dimensional services. In order to realize the problem of optimizing the coverage of five man-machine-
mobile base stations with reconfigurable smart surface locations in the three-dimensional space of 6G
architectures, the sticky bacterium algorithm is applied as an effective tool for such constrained three-
dimensional coverage optimization problems [79]. The core problem in designing an autonomous
mobile robot is planning a collision-free path with the minimum processing time and cost under the
beam conditions. Therefore, autonomous mobile robots can solve path planning and obstacle avoidance
problems. SMA is also used to solve such problems due to its excellent search capabilities and
development trends. Agarwal et. al.[80] presented an advanced SMA variant and demonstrated that the
method takes less effort and time to generate the best collision-free path compared to other comparative
algorithms by designing the corresponding application scenarios. Moreover, Abdollahzadeh et al.[67]
discretized the SMA and improved the performance of this discrete SMA using bitwise and Gaussian
mutation mechanisms to handle infeasible solutions using a penalty function and a repair algorithm.
The algorithm performs satisfactorily in 0-1 knapsack problems of different sizes. Otherwise, the
multidimensional knapsack problem has always been considered a difficult task because of its NP-hard
property. To solve this type of problem efficiently, Abdel-Basset et. al.[68] developed a binary form of
SMA and designed two different variants, using a repair mechanism to deal with the constraints and
infeasible solutions in both variants. Further, the clear superiority of the method over the chosen
comparison one is demonstrated on two well-known problems of large scale and small scale.
In general, SMA has several advantages when dealing with scheduling problems. First, SMA has a
simple algorithm structure, and it can be easily modified to be applied to various complex scheduling
tasks; second, SMA has adaptive parameters, and it does not need to spend much time to adjust the
parameters to get performance improvement; third, SMA has a powerful multi-point search capability,
which is a natural advantage of SMA in scheduling problems, and using adaptive positive and negative
feedback weights 𝑊 ⃗⃗⃗ to realize the population through a simultaneous search of multiple points can
improve the efficiency of scheduling tasks. However, existing researchers still have a problem with the
application of SMA in the field of scheduling optimization. Because scheduling tasks are usually discrete,
and researchers' discretization of SMA mainly takes the transform function to map continuous
candidate solutions into binary or integer numbers. It does not modify the formulation of the algorithm
in a discrete way, which can reduce the performance of SMA on discrete problems.

5.2 Engineering optimization

The engineering optimization model in position 2 in Figure 5 demonstrates the application of

SMA in the engineering field. The optimization capabilities of SMA have been demonstrated in
traditional benchmark function tests [29, 55]. In order to further illustrate the application value of SMA,
more and more scholars have explored its performance in practice. Many practical problems can be
reduced to engineering optimization problems in the fields of production planning, control, fault
diagnosis, and reconstruction. There are many ways to solve this problem, and SMA has gradually
developed as an important tool for solving such problems. This Section presents and discusses the
application of SMA and its variants in engineering design problem.
The engineering optimization problem has been widely studied in engineering and scientific
application. The classic engineering design problem includes welding beam, pressure vessel,
compression spring, automobile parts, reinforced concrete beam, pipe string, and so on. Moreover, a
realistic engineering problem is often subject to many practical conditions, for example, how to
maximize benefits at minimum cost. Therefore, it is critical to find a way to deal with these constraints.
Typical engineering design problems can be expressed by nonlinear programming, and its mathematical
model can be expressed as Eq. (8).

𝑚𝑖𝑛 𝑓(𝑋) (8)

𝑠. 𝑡. 𝑔ⅈ (𝑋) ≤ 0; 𝑖 = 1, 2, ...,𝑚
𝑙𝑏𝑗 ≤ 𝑋𝑗 ≤ 𝑢𝑏𝑗 ; 𝑗 = 1, 2, . . . , 𝑛
where 𝑋 denotes the decision variable, 𝑓(𝑋) means the objective function, 𝑔ⅈ (𝑋) is the inequality
constraint, and 𝑙𝑏𝑗 and 𝑢𝑏𝑗 express the upper bound and the lower bound of the decision variable 𝑋𝑗 ,
respectively.
In order to effectively solve this kind of constrained engineering optimization problem, various
optimization methods are used. SMA realizes its application value in this problem by its simple structure
and developable searching ability. Yu et. al.[38] first introduced the operation of the quantum rotation
gate mechanism and water cycle method into SMA, designing a variant of SMA. The algorithm
guarantees a further expansion of the population search space with a more detailed search of the area,
and the algorithm proves its optimization capability in three classical engineering design problems.
Houssein et. al.[58] used mutation operations in the adaptive guided differential evolution
algorithm to enhance the local search ability of SMA population agents and improve population
diversity, effectively avoiding the tendency of SMA to fall into local optimums. Besides, the method has
shown satisfactory optimization capabilities for rolling element bearing, tension/compression spring
design, and pressure vessel. Chauhan et. al.[81] presented a new hybrid algorithm for the basic problems
of SMA, combining an arithmetic optimizer algorithm (AOA), in which the local and global optimum
of SMA initializes the search process of AOA, and the globally optimal SMA obtained by AOA is
initialized again to further develop the search space. The lens opposition-based learning strategy is
further combined with the SMA to diversify the population of the algorithm and thus speed up
convergence. Moreover, several classical engineering optimization problems demonstrate the method's
effectiveness. Yin et al.[33] first adapted the exploration and exploration capabilities of SMA by
replacing the anisotropic search operation of SMA with the search operation of the equilibrium
optimizer, which guides the agents of SMA in their spatial search. Finally, a stochastic difference
mutation operator was added to improve the population diversity of SMA in the late iterations. In
summary, an advanced SMA variant was proposed. The improved SMA demonstrates its advantages in
terms of convergence speed, accuracy, and stability in nine classical engineering model optimization
problems.
Krishna et. al.[57] introduced a pattern search method to the SMA to further alter the randomly
generated position vector, and the altered position vector helped the mucilage to locate food sources.
Finally, nine classical engineering design problems successfully applied the hybrid method. Yin et. al.[27]
proposed an enhanced SMA based on dominant groups with adaptive T-distribution mutations, which
proves its application in eight classical engineering optimization problems. Moreover, satisfactory
optimization capabilities are demonstrated in solving an inverse kinematic problem for a 7-DOF robot
manipulator. Xiao et. al.[23] designed a pinhole imaging learning technique that combines an inverse
incomplete gamma function, tent chaotic mapping, and beta distribution to design an improved SMA.
The method has shown more active performance and robustness in two engineering optimization
problems than other improved meta-heuristics.
Tang et. al.[24] improved the SMA using chaotic opposition-based learning methods, adaptive
parameter control strategies, and a spiral search strategy, and the improved algorithm to solve two classic
engineering optimization cases. In solving such problems, the improved SMA outperformed other
algorithms in terms of convergence accuracy and stability. Altay et. al.[25] designed chaotic SMA by
using ten chaotic mappings to generate chaotic sequences instead of random values in SMA to prove
that the proposed method can solve practical engineering design problems effectively. Abualigah et.
al.[20] developed an SMA variant based on the reasonable integration of opposition-based learning and
levy flight, significantly improving search capability and convergence speed in solving five classic
engineering problems. Houssein et. al.[82] presented an advanced SMA variant based on opposition-
based and orthogonal learning strategies, proving effectiveness in three classic engineering problems.
Moreover, the team used it as an effective maximum power point tracking strategy to improve the
output energy of thermoelectric generators. By introducing Brownian motion, tournament selection
mechanism, and adaptive β-climbing mechanism into SMA, Sun et al.[41] introduced an improved
hybrid algorithm, which has an excellent performance in structural engineering design problems and
multi-layer perceptron design problems. A new hybrid SMA variant was proposed by Wang et. al.[53]
based on the salp swarm algorithm and SMA, and applied it to solve constrained engineering
optimization cases.
In addition, SMA is valuable in several typical engineering design problems and is a developable
tool in other manufacturing areas. Izci et. al.[83] tuned the proportional integral derivative (PID) with
SMA, which favorably regulates the speed of the DC motor and ensures the terminal output of the
automatic voltage regulator system (AVR). Yildiz et. al.[84] worked on a new thin-walled energy
absorber for the design of electric vehicles and enabled the application of SMA in the optimized design
of automotive components, playing a catalytic role in the rapid development of electric bicycles. Guha
et. al.[85] designed a resilient hybrid fractional-order controller by combining a fractional-order
proportional-integral-derivative controller with a fractional-order sliding-mode controller and used
SMA as an effective tool to calculate the optimal yield of the resilient hybrid fractional-order controller.
SMA was used in the parametric evaluation of proton exchange membrane fuel cells, effectively
overcomes the problem of non-linear parameter estimation caused by the process of performing
mathematical simulations, and is a valuable method for the parametric evaluation of models [86]. At
the same time, SMA was used to improve the performance of the three single-input single-output fuzzy
control systems, which play an effective auxiliary function for the control of the tower crane system.
Precupr et.al.[87] provided a fuzzy controller tuning method using SMA for proportional integral
fuzzy controllers represented by Takagi-Sugeno-Kang. Furthermore, in the framework of angular
position control of the laboratory servo systems, the superiority of the fuzzy controller under SMA
augmentation was demonstrated compared to other meta-heuristic algorithms for solving the same
optimization cases. Precup et al. [43] proposed a new version of SMA, denoted SMAF1, by introducing
the insertion information feedback model F1 into SMA and using it to minimize a discrete-time
objective function defined by time and control error, explore the optimal tuning of SMA in interval
type-2 fuzzy controllers. Also, the process of modeling nonlinear servo systems with an integral
component and dead zone plus saturation nonlinearity is explored for position control.
Gures [59] and his team have designed a novel hybrid SMA and simulated annealing algorithm and
applied it to the optimized design of vehicle components for a more in-depth exploration of the weight
reduction aspects of vehicles. The excellent performance of the method in solving shape optimization
problems was demonstrated by solving a design case for shape optimization using HSMA-SA. Izci et.
al.[13] proposed an enhanced SMA based on opposition-based learning with Nelder-Mead simplex and
used it to tune the parameters of a fractional-order proportional integral derivative controller and a
proportional-integral-derivative plus second-order derivative controller, demonstrating that the method
has excellent performance in tributary motor speed and automatic voltage regulator (AVR) control
system. An SMA based on the levy flight operator was used He et. al.[30] to tune the parameters of
the improved auto-disturbance rejection controller (ADRC) to enhance the stability of the ADRC
system, thus reducing the torque pulsation of the brushless DC clicks and further enhancing the driving
original components of the optical machine scanning system. Chen et. al.[88] designed a method for
extracting early bearing failure features based on optimized singular spectrum decomposition (OSSD)
with a 1.5-dimensional symmetric differential energy operator (1.5D -SDEO) demodulation is designed
as a method to extract early bearing failure characteristics. SMA is also used as an effective tool to
optimize the input parameters of the singular spectrum. The experimental results demonstrate the
positive effect of SMA on the extraction of early failure features of rolling bearings. Zhang et. al.[37]
combined the theory of quantum computing to propose an improved SMA and used it to suppress
noise when the unknown source signal is disturbed by impulsive noise with a low signal-to-noise ratio
(SNR). After experimental validation, the improved SMA was shown to have a more stable habit and
significantly improved accuracy compared to the traditional blind source separation method. Tiachacht
et. al.[89] applied SMA to assess the structural loss of a four-stories laboratory beam and a bar plane
truss and to monitor the structural health, demonstrating that SMA can provide excellent results for
the considered structure and, with the introduction of different levels of white Gaussian noise, can also
accurately predict the location and extent of the damage.
Further, SMA has achieved excellent performance in solving single-objective problems, and some
scholars have gone on to explore the performance of SMA when faced with multi-objective engineering
optimization problems. A multi-objective optimization algorithm was devised by Houssein et. al.[70]
based on SMA by designing an external archive to store the obtained Pareto-optimal solutions and used
to simulate social behavior in a multi-objective search space of slime mould. The method has been fully
applied to four constrained multi-objective optimization problems and has shown greater robustness
than the multi-objective particle swarm optimizer, the multi-objective grey wolf optimizer, the multi-
objective whale optimizer, multi-objective slam swarm optimizer, and the second version of the non-
dominated ranking genetic algorithm. Premkumar et. al.[71] developed a multi-objective variant of SMA
and used it in several practical engineering optimizations, such as welded beam design, discrete-gear
train design springs, and speed reducer design, further demonstrating that the SMA can solve multi-
objective engineering problems.
The available survey literature shows that SMAs are widely adopted to obtain better solutions in
engineering optimization tasks. Researchers prefer SMA mainly because of its simple structure and
efficient performance. Moreover, based on the compilation of the existing literature, we further believe
that the most crucial reason for SMA's popularity in engineering optimization is its scalability feature.
SMA can be used to solve various types of engineering problems with simple modifications, and various
techniques can be added to help optimize SMA for that engineering problem. However, the study of
SMA in engineering optimization tasks needs further attention. Because the existing engineering
optimization tasks all contain a large number of constraints, and a large part of the existing literature
only penalizes solutions that do not satisfy the constraints, for example, by changing the fitness value
of the current solution to the worst value when the solution does not satisfy the constraint relationship.
This approach will affect the performance of SMA, and if the constraint relations are strict, there will
be many unavailable solutions, and that kind of population will not be able to evolve further. Therefore,
how to deal with scenarios containing a large number of constraint relations when SMA is applied to
engineering optimization tasks is something that needs to be further studied by researchers.

5.3 Machine learning optimization

The machine model optimization in the third position and the deep learning model optimization
in the fourth position in Figure 5 demonstrate SMA's application in machine learning. In addition to its
effectiveness in common engineering problems, SMA plays a vital role in machine learning model
optimization. For example, when performing feature selection tasks, SMA can find features that are
more important for subsequent learning tasks based on its powerful global and local search capabilities.
In machine learning model parameter optimization problems, SMA is vital for improving the
effectiveness and practicality of machine learning algorithms. In optimizing machine learning model
parameters, SMA is critical to improving the effectiveness and practicality of machine learning
algorithms. In the search for neural network architecture, SMA can also adjust the weights and biases
to make the model have better performance.

Figure 6. The process for feature selection of SMA

Besides, more and more high-dimensional datasets have thousands of features in machine learning
applications, including bioinformatics, image processing, genomics, text classification and other fields
[90, 91]. These datasets often have many redundant and irrelevant features, which we can call noise,
negatively impacting the performance of methods [92]. Thus, feature selection is a key pre-processing
step for selecting related features for many machine learning assignments, such as clustering and
classification [93]. The specific SMA application process for feature selection is presented in Figure 6.
In general, many studies have proved the effectiveness of feature selection in high-dimensional
datasets, but it is still full of challenges because of feature interactions and large search space [69]. SMA
also has particular advantages for the feature selection problem. Ewees et al. [60] presented an enhanced
SMA based on the firefly algorithm named SMAFA. Moreover, 20 UCI datasets were evaluated for
SMAFA, and a comprehensive comparison with some existing MAs was performed. Ibrahim et al. [44]
proposed a fractional-order SMA for extracting features from real data to improve classification
accuracy. To improve the optimizer's performance in processing datasets, rough sets were regarded as
the adaptive functions to deal with the uncertainties in practice. Then, the proposed FOSMA was
compared with many other well-known feature selection methods to investigate their performance. Jia
et al. [48] optimized the SMA by introducing a compound mutation strategy and a restart strategy called
CMSRSSMA and proposed a CMSRSSMA-SVM model with both parameter optimization and feature
selection. The model's performance was tested with 14 datasets from the UCI data repository. The
results showed that the proposed work outperformed other methods in terms of fitness values on most
of the selected datasets, classification accuracy, and the number of features. Wazery et al. [17] presented
a modified SMA combined with a k-nearest neighbor (KNN) classifier based on an opposition-based
learning strategy. ISMA-KNN was superior to other algorithms in accuracy, specificity, sensitivity,
precision, computation time, and feature subset size on 9 UCI benchmark disease datasets of varying
feature sizes.
Despite the rapid development of machine learning methods, traditional machine learning method
parameter optimization has been widely used in many industries. Combining SMA with traditional
machine learning methods will give it new vitality. For example, Chen et al. [26] presented a prediction
method that built on support vector regression (SVR), combining the chaotic SMA (CSMA).and the k-
mean clustering method (KMCM). Using eight low and high-dimensional datasets, suitable crucial
parameters of KMCM and CSMA were obtained, and their computational complexity, prediction
accuracy, and stability were evaluated. Although this method did not perform well in terms of stability,
it showed strong performance in computational complexity and prediction accuracy. Gao et al. [28]
proposed a modified SMA variant (MSMA) using a multiple swarm strategy. A prediction model based
on MSMA and SVM was also proposed. The model enhanced the ability of SVM to optimize parameter
adjustment by identifying more compact feature subsets to gain more suitable feature subsets and
parameters. It can be used as an effective intelligent employment stability forecasting method and
provide a reasonable reference for relevant departments to make employment decisions and policies
for graduates. Vashishtha et al. [18] proposed an intelligent identification scheme for identifying tapered
roller bearing defects using the extreme learning machine (ELM). Moreover, a contrastive SVM was
investigated based on finding the optimal parameters (weights between input and output layers; and
deviations of hidden neurons). Using the optimized ELM parameters, a classification model was
developed. Yan et al. [94] combined a reduced kernel extreme learning machine (RKELM) with a new
deep learning network based on sample entropy for prediction model assignment. The low and high
components of the subsequence were analyzed using the empirical wavelet transform (SE).
Furthermore, for components with high SE values, the multi-parameter optimization of RKELM was
performed using SMA with better convergence speed and accuracy. Zhang et al. [51] designed a
bankruptcy prediction model based on KELM. This model tried to avoid premature convergence and
enhanced global exploration and local exploitation stability by using the SMA algorithm.
Deep learning models and learning reasoning models have a wide range of applications as popular
learning frameworks in recent years [95]. However, due to its excessive parameters and complex process,
it has some defects in some applications [6]. As a newly proposed metaheuristic algorithm, SMA also
has some advantages in optimizing deep learning frameworks. By vectorizing the structural parameters
of the deep learning frameworks and unifying the same ones with different mathematical
representations using binary coding, SMA is adapted to introduce an improvement strategy and apply
the improved SMA to network parameter model search. Al-qaness et al. [61] proposed an advanced
adaptive neuro-fuzzy inference system (ANFIS) to forecast the air quality index in Wuhan, which uses
a hybrid optimization method based on SMA combined with ANFIS, using three years of air quality
index time series data for training, as a way to judge the changes in air quality after the closure in Wuhan.
In addition, AlRassas et al. [15] also used SMA to improve ANFIS, proposed a model for developing
oil production time series forecasts, and evaluated the developed model using different real oil
production data from two fields in two different countries. The results show that the model performs
significantly as an efficient time series prediction model for forecasting. Karuppannan, A. and M.
Muthusamy designed a wavelet neural learning-based fuzzy system controller [96] to effectively control
the speed of a brushless DC motor with set specifications. This controller used SMA to adjust the
weight coefficients in the wavelet neural learning process, thus controlling the delayed convergence.
The results obtained during the simulation proved the superiority and effectiveness of the proposed
work compared to the previous model. SMA optimized an artificial neural network model was also used
by Zubaidi et al. [97] to predict the median of the random signal of monthly urban water demand. This
approach produced exact results with a determination coefficient of 0.9 and an average absolute relative
error of 0.001. The research can help local water managers effectively regulate existing water systems
and plan expansions to accommodate growing water demand. Guo et al. [98] designed a real-time
accurate sonar image recognition method based on hierarchical three-stage deep learning. In the first
step, the automatic feature extractor used a traditional deep convolutional neural network (DCNN). To
reduce tuning and testing time, ELM replaced the last fully connected layer in the second stage. Since
adding ELM to the model introduces uncertainty, in the third stage, SMA would be adopted to adjust
the input weights and deviations of ELM.
Intelligent diagnostics has become more and more common in diagnostic application fields, but
its improvements would be closely related to the reliability of diagnostic applications. However, the
iterative process of experimentation will significantly impact the value of the technology in practice
and the development level to which it can be used in the diagnostic field. Improved SMAs also have
been widely used in many medical areas. Wu et al. [99] proposed a framework consisting of a random
forest and an SVM model optimized by SMA. In this framework, the optimal SVM model is trained
using random forest when identifying key factors, SMA, and applied to identify and clarify the severity
of the clinical diagnosis of COVID-19 infection. Shi et al. [34] designed an evolutionary SVM model
using the improved SMA to identify and distinguish the severity of the COVID-19 disease according
to the basic information and hematological indicators of patients.
SMA is also widely used in the field of machine learning, including feature selection, traditional
machine learning model parameter search, and deep learning hyperparametric adaptive search. Among
them, feature selection is a hot topic of interest for researchers, and by combining it with traditional
machine learning models to form a powerful feature selection framework, key features can be selected
in high-dimensional feature data. Second, SMA has been widely adopted for hyperparametric search in
traditional machine learning and deep learning. The combination of machine learning and SMA is an
exciting project, and there are still many research directions, such as the use of SMA for neural network
architecture search[100], the combination of SMA and deep reinforcement learning [101], etc.

5.4 Energy optimization

Position 5 in Figure 5 demonstrates SMA's application in the energy optimization field. The SMA
has better search capacity than other methods, and many scholars applied it to solve energy optimization
problems. First, many research efforts are aimed at solving grid power dispatch problems in the energy
domain. For example, Abdel-Mawgoud et. [102] adopted a novel SMA to determine the best location
and size for individual battery energy storage in a radial power distribution system and effectively
verified the SMA on IEEE 69 bus systems with different kinds of loads. Kamboj et al. [103] applied
SMA to solve the non-convex, and cost-effective load scheduling issue in power systems, discussed the
effectiveness of SMA in single-region economic load scheduling of small, medium, and large power
systems, and found that differences between other well-known metaheuristics to confirm its results.
Mouassa et al. [104] proposed an improved SMA, which minimized the entire operating cost of the
primary grid by managing the flow between different power generation resources. Wei et al. [105]
adopted an improved SMA to find the optimal reactive power scheduling and verified the superiority
of the other optimization techniques in the literature. In short, SMA has a strong capability to solve the
scheduling problem of the power grid, and the experimental effect is highly superior. SMA for solving
the optimal coordination of directional overcurrent relays was proposed by Draz et al. [106] in a
reticulated power grid. The comparison with the mature water cycle algorithm shows that the SMA is
a competitive algorithm and reliable tool.
 In addition, many researchers have used SMA to solve the optimal flow problem. SMA was applied
by Kouadri et al. [107] to settle the optimal power flow problem in a real power system and verified the
effectiveness by combining a static VAR compensator device and stochastic wind power generation to
obtain the optimal solution of the power flow problem. SMA was applied by Nusair et al. [108] in the
optimal power flow problem in flexible AC transmission systems and stochastic renewable energy
power networks. Biswal et al. [45] designed a quasi-reflection-based SMA to optimize the power flow
distribution of radial power distribution systems. The proposed method was effectively verified through
benchmark function test experiments and IEEE 69, 85, and 118 bus test cases. Amigue et al. [109]
applied the SMA to determine the optimal power point in the photovoltaic power generation system,
and through the IEEE 33 and 69 bus test cases, it was shown that the SMA has the best effect.
Bhandakkar et al. [110] designed an integrated SMA to deal with the allocation issue of hybrid power
flow controllers, and the results showed that using this method can minimize the system's actual power
loss and generation cost. Khunkitti et al. [111] applied the SMA in the multi-objective optimal power
flow problem, tested it on IEEE 30, 57, and 118 buses, and showed that the SMA provided a better
solution than the literature. Farhat et al. [112] introduced the neighborhood dimension learning search
strategy to SMA to solve the optimal power flow problem and compared it against the other two
algorithms. The results showed the advantages of the proposed one.
Some scholars use different SMA versions to solve the parameter estimation problem in
photovoltaic (PV) power generation systems. For example, Mostafa et al. [113] applied the SMA to
identify the parameter of the single-diode model (SDM) and double-diode model (DDM) of PV cells
and compared it comprehensively with existing methods. The results show that the SMA has superior
performance and accuracy. Kumar et al. [114] used SMA to estimate the solar cell parameter, verifying
the performance of SMA. Yousri et al. [115] adopted a variety of modern meta-heuristic algorithms,
such as SMA, atomic search algorithm, etc., to simulate the behavior of photovoltaic power generation
systems under various environmental conditions, providing a variety of alternatives for solving this
problem. EI-Fergany et al. [116] used the improved SMA and Lambert W function to estimate the
unknown parameters of the solar cell system. Comparing with other competing methods, the results
showed that the proposed model has the highest precision in identifying the parameters of the
photovoltaic cell model.
Gush et al. [117] proposed a reactive power control technique for intelligent inverters for battery
energy storage and PV systems, and the SMA was used to improve the PV by carrying capacity of the
distribution network. The results show that the SMA method has an excellent performance in finding
the optimal PVHC of the distribution network. Yousri et al. [118] proposed a novel hybrid algorithm
of marine predator algorithm and SMA to identify the parameters of the three-diode photovoltaic cell
model, indicating that the proposed method obtained the optimum. Mirza et al. [119] employed a novel
SMA and an improved salps swarm algorithm to track the maximum power point of different
photovoltaic arrays, and the results showed that the proposed controller had transient and steady-state
efficiencies as high as 99.9%. Liu et al. [120] proposed an enhanced SMA to identify the parameters of
the single-diode model, the double-diode model, and the three-diode model. By comparing experiments
with other algorithms, the algorithm can be used to extract important unknowns in photovoltaic models.
A reliable and well-developed tool for parameters. Wen et al. [121] put forward an advanced SMA to
extract unknown parameters of the photovoltaic cell model. The implementation results showed that
the proposed method could be reliable for estimating photovoltaic model parameters. Gupta et al. [86]
used the SMA to solve the parameter identification of the proton exchange membrane fuel cell model
and effectively verified that the SMA was superior to other comparison algorithms through comparative
experiments. Lin et al. [122] designed an enhanced SMA based on a double-based best mutation and a
trigonometric-based mutation for solar photovoltaic parameter optimization, which had the highest
accuracy compared with others. From the above results, SMA achieved extremely superior results in
identifying different photovoltaic models.
Many researchers have also tried to use SMA to optimize hybrid power systems consisting of
multiple power sources to minimize resource integration costs and maximize power utilization efficiency.
Ha et al. [123] applied the SMA to gain the optimal power generation scheme between the power system
and thermal power plants, conventional hydropower plants, and pumped storage hydropower plants.
Gupta et al. [124] presented a hybrid optimization model for renewable energy using chaotic SMA and
particle swarm optimization, and the simulation results show that the proposed off-grid hybrid power
system is the most cost-effective option. EI-Sattar et al. [125] proposed an optimization model by
designing and analyzing three different hybrid power system configurations combining biomass systems
with photovoltaic, wind turbine, and battery systems. SMA is then used for model optimization to
reduce energy costs.
Some of the research works also try to use SMA to optimize the parallel design of power. Othman
et al. [126] presented an enhanced SMA to set the operating parameters of multi-stacks solid oxide fuel
cell stacks operating in parallel. The optimal dynamic parameters of the proposed algorithm in various
operating scenarios were verified by numerical simulation effectiveness. Kien et al. [127] A new meta-
heuristic viscous optimization algorithm (SMOA) is used to place shunt capacitors in two 69-node
efficient and 85-node radial distribution grids to reduce the investment cost and energy loss cost of the
capacitors. The main task of SMOA is to find the most suitable location for the shunt capacitors and
determine the optimal shunt capacitor generation for one year at three load levels.
SMA is also used in the field of energy optimization for generator design, power material design,
IoT power distribution, and emission dispatch cases. Singh et al. [128] put forward a chaotic SMA to
optimize the allocation of total power requirements among generators under specific constraints, and
it was effectively verified through five test cases with different power requirements. The algorithm
significantly reduces the total cost of electricity generation. A chaos-opposition-enhanced SMA was
proposed by Rizk-Aliah et al. [129] to minimize the energy cost of wind turbines at high altitudes. Rabah
et al. [130] used the SMA to reduce the influence of the magnetic coupling between metal pipelines and
high-voltage transmission lines. Taking 400 kV single-circuit horizontally configured lines and above-
ground pipelines as examples, the excellent performance of the proposed work was studied. Ponlatha
et al. [131] combined the SMA with the chimpanzee optimization algorithm to optimize the energy
distribution problem of the power distribution management system of the Internet of Things. Hassan
et al. [132] proposed two improved SMA and used them to solve the multi-objective Emission Dispatch
(EED) problems; these two algorithms are applied to minimize the total fuel costs and total emission
with the valve point effect simultaneously. In a word, various improved SMAs are widely used to solve
complex problems in the energy field.
The contribution of SMA in the energy field is very significant. It is applied to various aspects of
energy optimization, including grid optimization scheduling, optimal flow problems, photovoltaic
model optimization, hybrid power system design, power paralleling design, generator design, and
powerful material design. These research results demonstrate the strong competitive ability of SMA in
solving optimization problems in the energy field. Modeling the energy optimization problem and
subsequently solving it using SMA is a general solution; the advantage of SMA is that it does not focus
on the mathematical nature of the modeling but only on defining the scope of the decision space, the
type of variables, and the objective function. The simple structure and excellent SMA performance
make it possible to obtain feasible solutions for complex energy models quickly and in a limited time.

5.5 Image segmentation

The image segmentation model in the sixth position in Figure 5 demonstrates the application of
SMA in the field of image segmentation. It is an important pre-processing for computer vision and
image recognition. Image segmentation refers to separating different regions in an image under some
criteria. The different regions obtained by segmentation are non-intersecting, and each region satisfies
the region-specific consistency. The specific mathematical definition is as follows.
Let the set 𝐼 represent the whole image region, 𝐻denote the predicates with the same properties,
𝑅ⅈ denote the 𝑁 regions decomposed by 𝐼, 𝑖 = 1, 2, ⋯ , 𝑁, 𝑎𝑛𝑑, then the image segmentation region
satisfies the following conditions.
1）⋃𝑁
ⅈ=1 𝑅ⅈ = 𝐼, 𝑅ⅈ ∩ 𝑅𝑗 = ∅, ∀ 𝑖, 𝑗, 𝑖 ≠ 𝑗
2）∀ 𝑖, 𝑖 = 1, 2, ⋯ , 𝑁, 𝐻(𝑅ⅈ ) = 𝑇𝑟𝑢𝑒
3）∀ 𝑖, 𝑗, 𝑖 ≠ 𝑗, 𝐻(𝑅ⅈ ∪ 𝑅𝑗 ) = 𝐹𝑎𝑙𝑠𝑒
1) Indicate that the segmentation region should cover the whole image and each region does not
overlap; 2) Show that each region has the same properties; 3) It indicates that two regions with different
properties cannot be combined into one region. In conclusion, image segmentation's task is to separate
the image into non-overlapping regions with the same properties.
SMA can also correct some errors in grayscale image segmentation by a threshold-oriented
approach. A hybrid approach based on threshold technology was proposed by Abdel-Basset et al. [65]
for the segmentation of COVID-19 chest x-line images. It was that the SMA integrated with WOA to
maximize Kapur's entropy. The modified SMA was evaluated on 12 chest radiograph images, and the
threshold levels were as high as 30. Liu et al. [56] proposed a modified SMA with differential evolution
(MDE) algorithm. In addition, to gain high-quality breast cancer image segmentation results, we
developed a multilevel image segmentation model based on MDE centering on a nonlocal mean 2D
histogram and 2D Kapur-based entropy. Segmentation of dynamic contrast-enhanced magnetic
resonance imaging (DCE-MRI) of the breast was needed to assist radiologists in the inaccurate
detection and diagnosis of breast with DCE-MRI breast cancer. Patra et al. [133] proposed a model
based on Student Psychological Optimizer (SPBO) multilevel threshold segmentation for lesion
detection on breast DCE-MR images. Furthermore, the method was compared with the SMA algorithm
to show that the method can be used for DCE-MRI. An enhanced SMA was proposed by Lin et al. [31]
for multilevel thresholding image segmentation. Moreover, eight grayscale images were selected as the
benchmark images for testing and compared with other algorithms. Besides, experimental indicators
that the Wilcoxon rank sum test, mean fitness (mean), standard deviation (Std), structural similarity
index (SSIM), peak signal-to-noise ratio (PSNR) and feature similarity index (FSIM) were used to
evaluate the segmentation quality. The experimental results showed that the segmentation accuracy of
the proposed algorithm was better than others.
Image segmentation was the core step of image processing, among which multilevel image
segmentation was considered one of the most straightforward and effective methods. A Renyi entropy
multi-threshold image segmentation method based on the improved SMA (DASMA) was designed by
Zhao et al. [36]. The multilevel thresholding segmentation based on DASMA has also been successfully
applied in CT image segmentation of chronic obstructive pulmonary disease (COPD), and which
results verified the performance of DASMA. It can help physicians to quantitatively analyze the diseased
tissues, improve diagnostic accuracy and develop the right treatment plan. Anter et al. [23] proposed a
diagnostic optimization model based on adaptive fuzzy c-means (AFCM) and Le acute accent
distribution SMA named AFCM- LSMA for COVID-19. The FCM algorithm isolated the lung region
from the chest radiograph, and the image intensity histogram was used to reduce the time and
computational effort during the clustering process. A multilevel LN image segmentation technique
based on improved SMA was proposed by Chen et al. [37]. Besides, finding the optimal set of thresholds
is the key to multilevel thresholding image segmentation (MLTIS). The segmentation method based on
the swarm algorithm is easy to get segmentation thresholds of poor quality and fall into local optimum.
Then, the study proposed an improved SMA-based MLTIS method.
Color image thresholding is a famous image segmentation method. An objective function based
on image entropy is defined according to the number and position of thresholds in the color histogram.
Multi-class images are obtained by multi-level thresholding segmentation. The thresholding technique's
main problem is determining the threshold color value for each image. For the image to be segmented,
it is challenging for a manual operator to determine the specific threshold. From this perspective,
multilevel thresholding segmentation of color images is an optimization problem where the algorithm
determines the optimal threshold value to get a satisfactory segmented image. Recently, meta-heuristic
algorithms (MAs), with the advantage of their flexible structure, have been widely used in many fields,
such as image thresholding [134]. However, the MAs still have several drawbacks; for example, they
tend to fall into local optima and converge slowly. Naik et al. [19], after investigating multilevel
thresholding objective functions to minimize the information about entropy dependent on various
categories. In that paper, a new balanced SMA for color image thresholding segmentation was proposed
for analyzing breast thermograms.
With image segmentation as the application background, SMA is used as a segmentation aid,
involving the following aspects of research: (1) SMA is regarded as a means to obtain segmentation
function thresholds to achieve the goal of segmentation efficiency improvement; (2) SMA is used to
solve the shortcomings of inaccurate clustering center finding and slow clustering speed in clustering;
(3) focusing on the improvement of SMA to strengthen its performance, while the advanced SMA
variants are applied in image segmentation, and the segmentation efficiency is improved on the basis
of ensuring an accurate reading.

6. Evaluation of the SMA

SMA is a swarm-based evolutionary algorithm, which belongs to the research field of metaheuristic
algorithms, and it is developed to solve unconstrained and constrained problems in single-objective
continuous. SMA contains two concepts of population and evolution and is an iterative merit-seeking
algorithm. It stimulates the whole process of slime molds predation, builds mathematical models to
construct adaptive weight matrices, and simulates the process of propagation waves of slime molds
based on biological oscillators to generate positive and negative feedback to form the best path to
connect food with good exploration ability and exploitation tendency. SMA is a stochastic optimization
algorithm with a stochastic process that simulates the behavioral rules in slime moulds predation.
Biologically, slime molds are a powerful path planning capability, achieving optimal paths between
multiple points of food through intrinsic biological oscillatory feedback [8].
 SMA has received much attention from researchers upon its release, mainly because of its several
advantages as follows: first, SMA has powerful optimization performance. SMA has been shown to
have better multi-point local search capability than classical difference algorithms, particle swarm
algorithms, and advanced particle swarm algorithms on standard benchmark functions, CEC2014
functions, and classical constrained engineering optimization problems [8]. Secondly, SMA has a simpler
algorithm structure compared to artificial bee colony [135], Harris hawk optimization algorithm [136],
ant colony algorithm [137], etc. (see Section 3 for details). Third, as seen in Tables 1 and 2, it is possible
to combine various mechanisms and algorithms into the algorithmic structure of SMA with beneficial
positive feedback effects, demonstrating the strong scalability of SMA. However, SMA also has its
disadvantages, which are unavoidable because no single algorithm can solve all optimization problems.
Swarm-based evolutionary algorithms mainly contain three processes: exploration, exploitation, and
balancing exploration to exploitation transition. To evaluate SMA comprehensively, we evaluate SMA
comprehensively in terms of four evaluation metrics: exploration, exploitation, balancing, and flexibility.
Exploration: The algorithm can quickly explore multiple unknown regions in a limited number of
time steps, and this capability allows the algorithm to discover regions with potential candidate solutions
quickly. In Eq. (5), the exploration capability of SMA relies on two aspects: one is the random
exploration of the entire decision space with uniformly distributed random number perturbation used
when 𝑟 (random value between 0 and 1) is less than 𝑧 (𝑧 is set to 0.03). This approach tends to be the
most inefficient because as the problem dimension increases and the search space expands, this search
cannot complete the acquisition of valuable information in finite time steps. Second, the multi-point
⃗⃗⃗ (calculated by Eq. (4)). This method
local search brought by positive and negative feedback weights 𝑊
can quickly obtain valuable regional information, but it needs to be built on the premise that the
population has already obtained the useful information. In Table 1, the researchers prefer to improve
the base SMA with OBL, MS, LF, and chaotic mechanisms, all of which have the special property of
improving the exploration ability of the algorithm. Therefore, these results can prove that SMA has the
defect of insufficient exploration ability.
Exploitation: In contrast to exploration, exploitation denotes the ability of an algorithm to
perform a fine-grained search in a local region, an ability that allows the algorithm to discover optimal
candidate solutions in a local search space quickly. In Eq. (5), SMA can use positive and negative
feedback weights w to exploit multiple local regions, a natural advantage of SMA. In addition, Table 2
and Section 4.1.2 indicate that researchers prefer embedding SMA as an exploitation operator into other
algorithms. These results show that SMA has multi-point local exploitation capability and can refine the
search for multiple regions with potential optimal candidate solutions at the one iteration step.
Balancing: It is very important that the algorithm can balance the adaptive transition from
exploration to exploitation. When the algorithm finds a valuable local region through exploration, it
needs to make a refined exploitation of that local region immediately. In addition, when the algorithm
finds a local region that is locally optimal, it needs to mobilize the exploration mechanism to jump out
of the local optimum. In Eq. (5), SMA is mainly used to adjust the local search area by adaptive weights
w for positive and negative feedback, in addition to adding a uniform random search formula and using
the hyperparameter 𝑧 to design the timing of invoking this formula. Meanwhile, as shown in Figure 1,
“Other” indicates that the base SMA (without improvements) is used to deal with complex real-world
problems, accounting for 36.7% of the overall surveyed literature. These results demonstrate that SMA
has a good balance between exploration and exploitation and is widely recognized by researchers as a
competitive optimizer for handling complex real-world problems.
 Flexibility: The algorithm can be used quickly to handle a variety of real-world problems without
much change. Flexibility consists of three main aspects: parameter adaptivity, structural simplicity, and
scalability. In Section 3, SMA contains only the hyperparameter 𝑧 which needs to be defined in advance,
and the other parameters 𝑊 ⃗⃗⃗ , 𝑣𝑏 and 𝜈𝑐 are adaptively adjusted according to the iteration cycle. In
addition, Algorithm 1 shows the process framework of SMA. The overall SMA contains only two-layer
for loops, and the whole process of exploration and development is done through Eq. (5), whose simple
structure is widely favored by researchers (SMA is dealt with optimization challenges in different
application domains in Section 5). In addition, the high scalability can be considered the main advantage
of SMA algorithms. In Section 3.1, both mechanism-based and hybrid-based approaches are required
to extend the structure of SMA to improve its optimization capability. Furthermore, the methods are
effective, mainly due to SMA's strong scalability.
SMA is a powerful optimizer with efficient performance on both single-objective continuous
unconstrained, and constrained problems. It has the advantageous properties of simple structure,
adaptive parameter adjustment, scalability, and multi-point local search, but it also faces the inherent
drawbacks of insufficient global exploration ability and quickly falls into local optimum impact. From
Sections 4 and 5, researchers improve the base SMA from different aspects of the algorithm and use it
to deal with various complex real-world optimization problems. From the results, SMA has potential
research value and good application prospects.

7. Conclusion and future direction

SMA has received much attention from researchers since it was proposed in 2020. In this paper,
we review the research progress of SMA, for which, to our knowledge, there is no research review
available. We first obtained a total of 98 papers after searching and selecting the existing literature on
the Web of Science. Based on this, we used a new taxonomy to provide a comprehensive and structured
summary and analysis of SMA, presenting insightful observations and discussions and supplemented
with clear figures and tables to visualize the research results.
In this paper, we classify the SMA research progress from two perspectives. First, with the
advanced version of SMAs as the first review vein, the existing literature is divided into two categories:
single-objective and multi-objective versions. Further, the single-objective versions are subdivided into
mechanism-based, hybrid-based, and discrete-based methods. A comprehensive review of advanced
versions of SMAs was conducted to examine the research progress. From the findings, it is clear that
advanced versions of SMAs are still mainly single-objective versions and contain only a small number
of multi-objective versions. Besides, the improvement ideas of the single-objective version of advanced
SMAs are mainly mechanism-based methods (e.g., OBL, MS, LF strategies) and hybrid-based methods
(hybridization with other metaheuristic algorithms).
Second, the application domains of SMA are used as the second major review vein. Algorithms
have been currently used to try to solve optimization challenges in five application domains, including
scheduling optimization, engineering optimization, machine learning optimization, energy optimization,
and image segmentation. SMA proves to be an effective tool for optimizing complex real-world
problems and is a promising optimizer.
Finally, the existing studies related to SMA are summarized, analyzed, and discussed mentioned
above. Furthermore, the advantages and disadvantages of SMA are analyzed in Section 6, where we
define four evaluation metrics to assess the optimization performance of SMA comprehensively. In
general, the advantages of SMA focus on four aspects: 1) simple structure, 2) adaptive parameters; 3)
high scalability, and 4) powerful multi-point local search capability. Besides, the main disadvantage of
SMA is its weak global exploration ability. Based on the above analysis, we propose the following
suggestions for SMA's future research work.
⚫ Expand more multi-objective and discrete versions of SMA. In the survey in Section 4, SMA has
only 5 research results in the multi-objective domain, while the discrete SMA has only 4. This
dramatically affects the ability to extend SMA to more practical problems. SMA has a natural
advantage in discrete problems such as combinatorial optimization, path planning, and project
scheduling because of its powerful multi-point search capability, and future researchers can focus
more on this research direction.
⚫ Using machine learning to increase the optimization power of SMA. In Section 5, SMA is used to
help machine learning models for parameter optimization or combined for feature selection. In
fact, machine learning or deep learning models can also help improve the algorithm's performance
[138]. It will be a promising research direction on how to use machine learning to compensate for
the shortcomings of SMA and improve its performance.
⚫ Building an open source community for SMAs. A large number of advanced versions of SMAs
appear in Section 4, as well as SMAs that reuse the same mechanism for improvements. Establish
a unified standard SMA performance verification platform to make the performance comparison
between algorithms more convenient with more reliable experimental results. Gradually establish
an open-source community to facilitate the updates and development of SMA.
⚫ Theoretical studies of SMA are conducted. The current research work has been conducted
experimentally to verify the algorithm's effectiveness and lacks a rigorous theoretical basis.
Although various advanced versions of SMAs have been proposed, suffering from the lack of a
unified theoretical analysis framework, we cannot fundamentally explain the advantages of various
advanced SMAs. A complete theoretical analysis framework would promote the healthy
development of the SMA community.

Acknowledgment
This work was supported in part by the Natural Science Foundation of Zhejiang Province
(LZ22F020005), National Natural Science Foundation of China (62076185, U1809209).

Disclosure statement
No potential conflict of interest was reported by the author.

Data availability statement

The datasets generated and(or) analyzed during the current study are available from the
corresponding author on reasonable request.
References
[1] M. Zhang, Y. Chen, and W. Susilo, "PPO-CPQ: a privacy-preserving optimization of clinical
pathway query for e-healthcare systems," IEEE Internet of Things Journal, vol. 7, no. 10, pp. 10660-
10672, 2020.
[2] K. Zhang et al., "Training effective deep reinforcement learning agents for real-time life-cycle
production optimization," Journal of Petroleum Science and Engineering, vol. 208, p. 109766, 2022.
[3] B. Zhu et al., "A Novel Reconstruction Method for Temperature Distribution Measurement Based
on Ultrasonic Tomography," IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 2022.
[4] X. Luo, Y. Yuan, S. Chen, N. Zeng, and Z. Wang, "Position-transitional particle swarm optimization-
incorporated latent factor analysis," IEEE Transactions on Knowledge and Data Engineering, 2020.
[5] N. Zeng, Z. Wang, W. Liu, H. Zhang, K. Hone, and X. Liu, "A dynamic neighborhood-based
switching particle swarm optimization algorithm," IEEE Transactions on Cybernetics, 2020.
[6] W. Liu, Z. Wang, N. Zeng, F. E. Alsaadi, and X. Liu, "A PSO-based deep learning approach to
classifying patients from emergency departments," International Journal of Machine Learning and
Cybernetics, vol. 12, no. 7, pp. 1939-1948, 2021.
[7] W. Liu, Z. Wang, Y. Yuan, N. Zeng, K. Hone, and X. Liu, "A novel sigmoid-function-based adaptive
weighted particle swarm optimizer," IEEE transactions on cybernetics, vol. 51, no. 2, pp. 1085-1093,
2019.
[8] S. Li, H. Chen, M. Wang, A. A. Heidari, and S. Mirjalili, "Slime mould algorithm: A new method
for stochastic optimization," FUTURE GENERATION COMPUTER SYSTEMS-THE
INTERNATIONAL JOURNAL OF ESCIENCE, Article vol. 111, pp. 300-323, 2020 OCT 2020,
doi: 10.1016/j.future.2020.03.055.
[9] M. M. Millonas, "Swarms, phase transitions, and collective intelligence," arXiv preprint adap-
org/9306002, 1993.
[10] H. Wang, Q. Gao, H. Li, H. Wang, L. Yan, and G. Liu, "A Structural Evolution-Based Anomaly
Detection Method for Generalized Evolving Social Networks," The Computer Journal, vol. 65, no. 5,
pp. 1189-1199, 2022.
[11] N. Huang, S. Wang, R. Wang, G. Cai, Y. Liu, and Q. Dai, "Gated spatial-temporal graph neural
network based short-term load forecasting for wide-area multiple buses," International Journal of
Electrical Power & Energy Systems, vol. 145, p. 108651, 2023.
[12] K. Shang et al., "Haze prediction model using deep recurrent neural network," Atmosphere, vol. 12,
no. 12, p. 1625, 2021.
[13] D. Izci, S. Ekinci, H. L. Zeynelgil, and J. Hedley, "Performance evaluation of a novel improved
slime mould algorithm for direct current motor and automatic voltage regulator systems,"
TRANSACTIONS OF THE INSTITUTE OF MEASUREMENT AND CONTROL, Article
Early Access 2022, doi: 10.1177/01423312211037967.
[14] E. H. Houssein, B. E.-d. Helmy, H. Rezk, and A. M. Nassef, "An efficient orthogonal opposition-
based learning slime mould algorithm for maximum power point tracking," NEURAL
COMPUTING & APPLICATIONS, Article vol. 34, no. 5, pp. 3671-3695, 2022 MAR 2022, doi:
10.1007/s00521-021-06634-y.
[15] A. M. AlRassas et al., "Advance artificial time series forecasting model for oil production using
neuro fuzzy-based slime mould algorithm," JOURNAL OF PETROLEUM EXPLORATION
AND PRODUCTION TECHNOLOGY, Article vol. 12, no. 2, pp. 383-395, 2022 FEB 2022, doi:
 10.1007/s13202-021-01405-w.
[16] K. Yu, L. Liu, and Z. Chen, "An Improved Slime Mould Algorithm for Demand Estimation of
Urban Water Resources," MATHEMATICS, Article vol. 9, no. 12, 2021 JUN 2021, doi:
10.3390/math9121316.
[17] Y. M. Wazery, E. Saber, E. H. Houssein, A. A. Ali, and E. Amer, "An Efficient Slime Mould
Algorithm Combined With K-Nearest Neighbor for Medical Classification Tasks," IEEE ACCESS,
Article vol. 9, pp. 113666-113682, 2021 2021, doi: 10.1109/ACCESS.2021.3105485.
[18] G. Vashishtha, S. Chauhan, M. Singh, and R. Kumar, "Bearing defect identification by swarm
decomposition considering permutation entropy measure and opposition-based slime mould
algorithm," MEASUREMENT, Article vol. 178, 2021 JUN 2021, doi:
10.1016/j.measurement.2021.109389.
[19] M. K. Naik, R. Panda, and A. Abraham, "An entropy minimization based multilevel colour
thresholding technique for analysis of breast thermograms using equilibrium slime mould
algorithm," APPLIED SOFT COMPUTING, Article vol. 113, 2021 DEC 2021, doi:
10.1016/j.asoc.2021.107955.
[20] L. Abualigah, A. Diabat, and M. Abd Elaziz, "Improved slime mould algorithm by opposition-
based learning and Levy flight distribution for global optimization and advances in real-world
engineering problems," JOURNAL OF AMBIENT INTELLIGENCE AND HUMANIZED
COMPUTING, Article
Early Access 2021, doi: 10.1007/s12652-021-03372-w.
[21] T. Singh, "Chaotic slime mould algorithm for economic load dispatch problems," APPLIED
INTELLIGENCE, Article
Early Access 2022, doi: 10.1007/s10489-022-03179-y.
[22] R. M. Rizk-Allah, A. E. Hassanien, and D. Song, "Chaos-opposition-enhanced slime mould
algorithm for minimizing the cost of energy for the wind turbines on high-altitude sites," ISA
TRANSACTIONS, Article vol. 121, pp. 191-205, 2022 FEB 2022, doi:
10.1016/j.isatra.2021.04.011.
[23] Y. Xiao, X. Sun, Y. Zhang, Y. Guo, Y. Wang, and J. Li, "AN IMPROVED SLIME MOULD
ALGORITHM BASED ON TENT CHAOTIC MAPPING AND NONLINEAR INERTIA
WEIGHT," INTERNATIONAL JOURNAL OF INNOVATIVE COMPUTING
INFORMATION AND CONTROL, Article vol. 17, no. 6, pp. 2151-2176, 2021 DEC 2021, doi:
10.24507/ijicic.17.06.2151.
[24] A.-D. Tang, S.-Q. Tang, T. Han, H. Zhou, and L. Xie, "A Modified Slime Mould Algorithm for
Global Optimization," COMPUTATIONAL INTELLIGENCE AND NEUROSCIENCE,
Article vol. 2021, 2021 NOV 24 2021, doi: 10.1155/2021/2298215.
[25] O. Altay, "Chaotic slime mould optimization algorithm for global optimization," ARTIFICIAL
INTELLIGENCE REVIEW, Article
Early Access 2021, doi: 10.1007/s10462-021-10100-5.
[26] Z. Chen and W. Liu, "An Efficient Parameter Adaptive Support Vector Regression Using K-Means
Clustering and Chaotic Slime Mould Algorithm," IEEE ACCESS, Article vol. 8, pp. 156851-
156862, 2020 2020, doi: 10.1109/ACCESS.2020.3018866.
[27] S. Yin, Q. Luo, Y. Du, and Y. Zhou, "p DTSMA: Dominant Swarm with Adaptive T-distribution
Mutation-based Slime Mould Algorithm," MATHEMATICAL BIOSCIENCES AND
ENGINEERING, Article vol. 19, no. 3, pp. 2240-2285, 2022 2022, doi: 10.3934/mbe.2022105.
[28] H. Gao, G. Liang, and H. Chen, "Multi-Population Enhanced Slime Mould Algorithm and with
Application to Postgraduate Employment Stability Prediction," ELECTRONICS, Article vol. 11,
no. 2, 2022 JAN 2022, doi: 10.3390/electronics11020209.
[29] J. Alfadhli, A. Jaragh, M. G. Alfailakawi, and I. Ahmad, "FP-SMA: an adaptive, fluctuant population
strategy for slime mould algorithm," NEURAL COMPUTING & APPLICATIONS, Article
Early Access 2022, doi: 10.1007/s00521-022-07034-6.
[30] J. He, C. Yan, and X. Wang, "Torque Ripple Suppression of Brushless DC Motor Drive System
Based on Improved Harmonic Injection Active Disturbance Rejection Control," SENSORS,
Article vol. 22, no. 3, 2022 FEB 2022, doi: 10.3390/s22031069.
[31] S. Lin, H. Jia, L. Abualigah, and M. Altalhi, "Enhanced Slime Mould Algorithm for Multilevel
Thresholding Image Segmentation Using Entropy Measures," ENTROPY, Article vol. 23, no. 12,
2021 DEC 2021, doi: 10.3390/e23121700.
[32] A. M. Anter, D. Oliva, A. Thakare, and Z. Zhang, "AFCM-LSMA: New intelligent model based on
Levy slime mould algorithm and adaptive fuzzy C-means for identification of COVID-19 infection
from chest X-ray images," ADVANCED ENGINEERING INFORMATICS, Article vol. 49, 2021
AUG 2021, doi: 10.1016/j.aei.2021.101317.
[33] S. Yin, Q. Luo, and Y. Zhou, "EOSMA: An Equilibrium Optimizer Slime Mould Algorithm for
Engineering Design Problems," ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING,
Article
Early Access 2022, doi: 10.1007/s13369-021-06513-7.
[34] B. Shi et al., "Early Recognition and Discrimination of COVID-19 Severity Using Slime Mould
Support Vector Machine for Medical Decision-Making," IEEE ACCESS, Article vol. 9, pp.
121996-122015, 2021 2021, doi: 10.1109/ACCESS.2021.3108447.
[35] M. Farhat, S. Kamel, A. M. Atallah, M. H. Hassan, and A. M. Agwa, "ESMA-OPF: Enhanced Slime
Mould Algorithm for Solving Optimal Power Flow Problem," SUSTAINABILITY, Article vol. 14,
no. 4, 2022 FEB 2022, doi: 10.3390/su14042305.
[36] S. Zhao et al., "Multilevel threshold image segmentation with diffusion association slime mould
algorithm and Renyi's entropy for chronic obstructive pulmonary disease," COMPUTERS IN
BIOLOGY AND MEDICINE, Article vol. 134, 2021 JUL 2021, doi:
10.1016/j.compbiomed.2021.104427.
[37] Z. Zhang, H. Gao, J. Ma, S. Wang, and H. Sun, "Blind Source Separation Based on Quantum Slime
Mould Algorithm in Impulse Noise," MATHEMATICAL PROBLEMS IN ENGINEERING,
Article vol. 2021, 2021 JUL 29 2021, doi: 10.1155/2021/1496156.
[38] C. Yu, A. A. Heidari, X. Xue, L. Zhang, H. Chen, and W. Chen, "Boosting quantum rotation gate
embedded slime mould algorithm," EXPERT SYSTEMS WITH APPLICATIONS, Article vol.
181, 2021 NOV 1 2021, doi: 10.1016/j.eswa.2021.115082.
[39] X. Weng, A. A. Heidari, G. Liang, H. Chen, and X. Ma, "An evolutionary Nelder-Mead slime mould
algorithm with random learning for efficient design of photovoltaic models," ENERGY
REPORTS, Article vol. 7, pp. 8784-8804, 2021 NOV 2021, doi: 10.1016/j.egyr.2021.11.019.
[40] Y. Liu, A. A. Heidari, X. Ye, G. Liang, H. Chen, and C. He, "Boosting slime mould algorithm for
parameter identification of photovoltaic models," ENERGY, Article vol. 234, 2021 NOV 1 2021,
doi: 10.1016/j.energy.2021.121164.
[41] K. Sun, H. Jia, Y. Li, and Z. Jiang, "Hybrid improved slime mould algorithm with adaptive beta hill
climbing for numerical optimization," JOURNAL OF INTELLIGENT & FUZZY SYSTEMS,
 Article vol. 40, no. 1, pp. 1667-1679, 2021 2021, doi: 10.3233/JIFS-201755.
[42] R.-E. Precup, R.-C. David, R.-C. Roman, E. M. Petriu, and A.-I. Szedlak-Stinean, "Slime Mould
Algorithm-Based Tuning of Cost-Effective Fuzzy Controllers for Servo Systems,"
INTERNATIONAL JOURNAL OF COMPUTATIONAL INTELLIGENCE SYSTEMS,
Article vol. 14, no. 1, pp. 1042-1052, 2021 2021, doi: 10.2991/ijcis.d.210309.001.
[43] R.-E. Precup, R.-C. David, R.-C. Roman, A.-I. Szedlak-Stinean, and E. M. Petriu, "Optimal tuning
of interval type-2 fuzzy controllers for nonlinear servo systems using Slime Mould Algorithm,"
INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE, Article
Early Access 2021, doi: 10.1080/00207721.2021.1927236.
[44] R. A. Ibrahim, D. Yousri, M. Abd Elaziz, S. Alshathri, and I. Attiya, "Fractional Calculus-Based
Slime Mould Algorithm for Feature Selection Using Rough Set," IEEE ACCESS, Article vol. 9,
pp. 131625-131636, 2021 2021, doi: 10.1109/ACCESS.2021.3111121.
[45] S. R. Biswal, G. Shankar, R. M. Elavarasan, and L. Mihet-Popa, "Optimal Allocation/Sizing of
DGs/Capacitors in Reconfigured Radial Distribution System Using Quasi-Reflected Slime Mould
Algorithm," IEEE Access, vol. 9, pp. 125658-125677, 2021, doi: 10.1109/ACCESS.2021.3111027.
[46] Y. Wei, Y. Zhou, Q. Luo, and W. Deng, "Optimal reactive power dispatch using an improved slime
mould algorithm," ENERGY REPORTS, Article vol. 7, pp. 8742-8759, 2021 NOV 2021, doi:
10.1016/j.egyr.2021.11.138.
[47] A. M. Othman and A. A. El-Fergany, "Optimal dynamic operation and modeling of parallel
connected multi-stacks fuel cells with improved slime mould algorithm," RENEWABLE
ENERGY, Article vol. 175, pp. 770-782, 2021 SEP 2021, doi: 10.1016/j.renene.2021.04.148.
[48] H. Jia, W. Zhang, R. Zheng, S. Wang, X. Leng, and N. Cao, "Ensemble mutation slime mould
algorithm with restart mechanism for feature selection," INTERNATIONAL JOURNAL OF
INTELLIGENT SYSTEMS, Article
Early Access 2022, doi: 10.1002/int.22776.
[49] H. R. Tizhoosh, "Opposition-based learning: a new scheme for machine intelligence," in
International conference on computational intelligence for modelling, control and automation and international
conference on intelligent agents, web technologies and internet commerce (CIMCA-IAWTIC'06), 2005, vol. 1:
IEEE, pp. 695-701.
[50] C. Li, J. Li, and H. Chen, "A meta-heuristic-based approach for QoS-aware service composition,"
IEEE Access, vol. 8, pp. 69579-69592, 2020.
[51] Y. Zhang et al., "Towards augmented kernel extreme learning models for bankruptcy prediction:
algorithmic behavior and comprehensive analysis," Neurocomputing, vol. 430, pp. 185-212, 2021.
[52] D. Yousri, A. Fathy, H. Rezk, T. S. Babu, and M. R. Berber, "A reliable approach for modeling the
photovoltaic system under partial shading conditions using three diode model and hybrid marine
predators-slime mould algorithm," ENERGY CONVERSION AND MANAGEMENT, Article
vol. 243, 2021 SEP 1 2021, doi: 10.1016/j.enconman.2021.114269.
[53] S. Wang et al., "A Hybrid SSA and SMA with Mutation Opposition-Based Learning for Constrained
Engineering Problems," Computational intelligence and neuroscience, Journal Article vol. 2021, pp.
6379469-6379469, 2021 2021, doi: 10.1155/2021/6379469.
[54] S. Ponlatha, P. Umasankar, P. B. Vadivu, and D. Chitra, "An IOT-based efficient energy management
in smart grid using SMACA technique," INTERNATIONAL TRANSACTIONS ON
ELECTRICAL ENERGY SYSTEMS, Article vol. 31, no. 12, 2021 DEC 2021, doi: 10.1002/2050-
7038.12995.
[55] M. K. Naik, R. Panda, and A. Abraham, "Adaptive opposition slime mould algorithm," SOFT
COMPUTING, Article vol. 25, no. 22, pp. 14297-14313, 2021 NOV 2021, doi: 10.1007/s00500-
021-06140-2.
[56] L. Liu et al., "Performance optimization of differential evolution with slime mould algorithm for
multilevel breast cancer image segmentation," COMPUTERS IN BIOLOGY AND MEDICINE,
Article vol. 138, 2021 NOV 2021, doi: 10.1016/j.compbiomed.2021.104910.
[57] A. B. Krishna, S. Saxena, and V. K. Kamboj, "hSMA-PS: a novel memetic approach for numerical
and engineering design challenges," ENGINEERING WITH COMPUTERS, Article
Early Access 2021, doi: 10.1007/s00366-021-01371-1.
[58] E. H. Houssein, M. A. Mahdy, M. J. Blondin, D. Shebl, and W. M. Mohamed, "Hybrid slime mould
algorithm with adaptive guided differential evolution algorithm for combinatorial and global
optimization problems," EXPERT SYSTEMS WITH APPLICATIONS, Article vol. 174, 2021
JUL 15 2021, doi: 10.1016/j.eswa.2021.114689.
[59] D. Gures, S. Bureerat, S. M. Sait, and A. R. Yildiz, "Comparison of the arithmetic optimization
algorithm, the slime mold optimization algorithm, the marine predators algorithm, the salp swarm
algorithm for real-world engineering applications," MATERIALS TESTING, Article vol. 63, no.
5, pp. 448-452, 2021 2021, doi: 10.1515/mt-2020-0076.
[60] A. A. Ewees et al., "Improved Slime Mould Algorithm based on Firefly Algorithm for feature
selection: A case study on QSAR model," ENGINEERING WITH COMPUTERS, Article
Early Access 2021, doi: 10.1007/s00366-021-01342-6.
[61] M. A. A. Al-qaness, H. Fan, A. A. Ewees, D. Yousri, and M. Abd Elaziz, "Improved ANFIS model
for forecasting Wuhan City Air Quality and analysis COVID-19 lockdown impacts on air quality,"
ENVIRONMENTAL RESEARCH, Article vol. 194, 2021 MAR 2021, doi:
10.1016/j.envres.2020.110607.
[62] X. Chen et al., "An efficient multilevel thresholding image segmentation method based on the slime
mould algorithm with bee foraging mechanism: A real case with lupus nephritis images,"
COMPUTERS IN BIOLOGY AND MEDICINE, Article vol. 142, 2022 MAR 2022, doi:
10.1016/j.compbiomed.2021.105179.
[63] S. Chauhan, G. Vashishtha, and A. Kumar, "A symbiosis of arithmetic optimizer with slime mould
algorithm for improving global optimization and conventional design problem," JOURNAL OF
SUPERCOMPUTING, Article vol. 78, no. 5, pp. 6234-6274, 2022 APR 2022, doi: 10.1007/s11227-
021-04105-8.
[64] A. A. A. Bhandakkar and L. Mathew, "Merging slime mould with whale optimization algorithm for
optimal allocation of hybrid power flow controller in power system," JOURNAL OF
EXPERIMENTAL & THEORETICAL ARTIFICIAL INTELLIGENCE, Article
Early Access 2022, doi: 10.1080/0952813X.2022.2040598.
[65] M. Abdel-Basset, V. Chang, and R. Mohamed, "HSMA_WOA: A hybrid novel Slime mould
algorithm with whale optimization algorithm for tackling the image segmentation problem of chest
X-ray images," APPLIED SOFT COMPUTING, Article vol. 95, 2020 OCT 2020, doi:
10.1016/j.asoc.2020.106642.
[66] D. Zhang et al., "Application of swarm intelligence algorithms to the characteristic wavelength
selection of soil moisture content," INTERNATIONAL JOURNAL OF AGRICULTURAL
AND BIOLOGICAL ENGINEERING, Article vol. 14, no. 6, pp. 153-161, 2021 NOV 2021, doi:
10.25165/j.ijabe.20211406.6629.
[67] B. Abdollahzadeh, S. Barshandeh, H. Javadi, and N. Epicoco, "An enhanced binary slime mould
algorithm for solving the 0-1 knapsack problem," ENGINEERING WITH COMPUTERS, Article
Early Access 2021, doi: 10.1007/s00366-021-01470-z.
[68] M. Abdel-Basset, R. Mohamed, K. M. Sallam, R. K. Chakrabortty, and M. J. Ryan, "BSMA: A novel
metaheuristic algorithm for multi-dimensional knapsack problems: Method and comprehensive
analysis," COMPUTERS & INDUSTRIAL ENGINEERING, Article vol. 159, 2021 SEP 2021,
doi: 10.1016/j.cie.2021.107469.
[69] M. Abdel-Basset, R. Mohamed, R. K. Chakrabortty, M. J. Ryan, and S. Mirjalili, "An efficient binary
slime mould algorithm integrated with a novel attacking-feeding strategy for feature selection,"
COMPUTERS & INDUSTRIAL ENGINEERING, Article vol. 153, 2021 MAR 2021, doi:
10.1016/j.cie.2020.107078.
[70] E. H. Houssein, M. A. Mahdy, D. Shebl, A. Manzoor, R. Sarkar, and W. M. Mohamed, "An efficient
slime mould algorithm for solving multi-objective optimization problems," EXPERT SYSTEMS
WITH APPLICATIONS, Article vol. 187, 2022 JAN 2022, doi: 10.1016/j.eswa.2021.115870.
[71] M. Premkumar, P. Jangir, R. Sowmya, H. H. Alhelou, A. A. Heidari, and H. Chen, "MOSMA: Multi-
Objective Slime Mould Algorithm Based on Elitist Non-Dominated Sorting," IEEE ACCESS,
Article vol. 9, pp. 3229-3248, 2021 2021, doi: 10.1109/ACCESS.2020.3047936.
[72] S. Khunkitti, A. Siritaratiwat, and S. Premrudeepreechacharn, "Multi-Objective Optimal Power
Flow Problems Based on Slime Mould Algorithm," SUSTAINABILITY, Article vol. 13, no. 13,
2021 JUL 2021, doi: 10.3390/su13137448.
[73] M. H. Hassan, S. Kamel, L. Abualigah, and A. Eid, "Development and application of slime mould
algorithm for optimal economic emission dispatch," EXPERT SYSTEMS WITH
APPLICATIONS, Article vol. 182, 2021 NOV 15 2021, doi: 10.1016/j.eswa.2021.115205.
[74] T. Gush, C.-H. Kim, S. Admasie, J.-S. Kim, and J.-S. Song, "Optimal Smart Inverter Control for PV
and BESS to Improve PV Hosting Capacity of Distribution Networks Using Slime Mould
Algorithm," IEEE ACCESS, Article vol. 9, pp. 52164-52176, 2021 2021, doi:
10.1109/ACCESS.2021.3070155.
[75] H. Wu, S. Jin, and W. Yue, "Pricing Policy for a Dynamic Spectrum Allocation Scheme with Batch
Requests and Impatient Packets in Cognitive Radio Networks," Journal of Systems Science and Systems
Engineering, vol. 31, no. 2, pp. 133-149, 2022.
[76] B. Xie, S. Li, F. Lv, C. H. Liu, G. Wang, and D. Wu, "A Collaborative Alignment Framework of
Transferable Knowledge Extraction for Unsupervised Domain Adaptation," IEEE Transactions on
Knowledge and Data Engineering, 2022.
[77] X. Liang, L. Luo, S. Hu, and Y. Li, "Mapping the knowledge frontiers and evolution of decision
making based on agent-based modeling," Knowledge-Based Systems, p. 108982, 2022.
[78] W. Zheng, Y. Xun, X. Wu, Z. Deng, X. Chen, and Y. Sui, "A comparative study of class rebalancing
methods for security bug report classification," IEEE Transactions on Reliability, vol. 70, no. 4, pp.
1658-1670, 2021.
[79] M. Abdel-Basset, L. Abdel-Fatah, K. A. Eldrandaly, and N. M. Abdel-Aziz, "Enhanced
Computational Intelligence Algorithm for Coverage Optimization of 6G Non-Terrestrial
Networks in 3D Space," IEEE ACCESS, Article vol. 9, pp. 70419-70429, 2021 2021, doi:
10.1109/ACCESS.2021.3078585.
[80] D. Agarwal and P. S. Bharti, "Implementing modified swarm intelligence algorithm based on Slime
moulds for path planning and obstacle avoidance problem in mobile robots," Applied Soft Computing,
 no. 4, p. 107372, 2021.
[81] S. Chauhan, G. Vashishtha, and A. Kumar, "A symbiosis of arithmetic optimizer with slime mould
algorithm for improving global optimization and conventional design problem," The Journal of
Supercomputing, vol. 78, no. 5, pp. 6234-6274, 2022.
[82] E. H. ccccc, B. E.-d. Helmy, H. Rezk, and A. M. Nassef, "An efficient orthogonal opposition-based
learning slime mould algorithm for maximum power point tracking," NEURAL COMPUTING &
APPLICATIONS, Article vol. 34, no. 5, pp. 3671-3695, 2022 MAR 2022, doi: 10.1007/s00521-
021-06634-y.
[83] D. Izci and S. Ekinci, "Comparative Performance Analysis of Slime Mould Algorithm For Efficient
Design of Proportional-Integral-Derivative Controller," ELECTRICA, Article vol. 21, no. 1, pp.
151-159, 2021 JAN 2021, doi: 10.5152/electrica.2021.20077.
[84] B. S. Yildiz, "Slime mould algorithm and kriging surrogate model-based approach for enhanced
crashworthiness of electric vehicles," INTERNATIONAL JOURNAL OF VEHICLE DESIGN,
Article vol. 83, no. 1, pp. 54-68, 2020 2020, doi: 10.1504/IJVD.2020.114786.
[85] D. Guha, P. K. Roy, and S. Banerjee, "Observer-aided resilient hybrid fractional-order controller for
frequency regulation of hybrid power system," INTERNATIONAL TRANSACTIONS ON
ELECTRICAL ENERGY SYSTEMS, Article vol. 31, no. 9, 2021 SEP 2021, doi: 10.1002/2050-
7038.13014.
[86] J. Gupta, P. Nijhawan, and S. Ganguli, "Optimal parameter estimation of PEM fuel cell using slime
mould algorithm," INTERNATIONAL JOURNAL OF ENERGY RESEARCH, Article vol. 45,
no. 10, pp. 14732-14744, 2021 AUG 2021, doi: 10.1002/er.6750.
[87] R. E. Precup, R. C. Roman, E. L. Hedrea, C. A. Bojan-Dragos, M. M. Damian, and M. L. Nedelcea,
"Performance Improvement of Low-Cost Iterative Learning-Based Fuzzy Control Systems for
Tower Crane Systems," INTERNATIONAL JOURNAL OF COMPUTERS
COMMUNICATIONS & CONTROL, Article vol. 17, no. 1, 2022 FEB 2022, doi:
10.15837/ijccc.2022.1.4623.
[88] Z. Chen, C. He, Y. Liu, S. Lu, F. Liu, and G. Li, "Incipient Fault Feature Extraction of Rolling
Bearing Based on Optimized Singular Spectrum Decomposition," IEEE SENSORS JOURNAL,
Article vol. 21, no. 18, pp. 20362-20374, 2021 SEP 15 2021, doi: 10.1109/JSEN.2021.3094880.
[89] S. Tiachacht, S. Khatir, T. Cuong Le, R. V. Rao, S. Mirjalili, and M. A. Wahab, "Inverse problem for
dynamic structural health monitoring based on slime mould algorithm," ENGINEERING WITH
COMPUTERS, Article
Early Access 2021, doi: 10.1007/s00366-021-01378-8.
[90] W. Zheng, X. Liu, and L. Yin, "Research on image classification method based on improved multi-
scale relational network," PeerJ Computer Science, vol. 7, p. e613, 2021.
[91] X. Wu, W. Zheng, X. Xia, and D. Lo, "Data Quality Matters: A Case Study on Data Label
Correctness for Security Bug Report Prediction," IEEE Transactions on Software Engineering, pp. 1-1,
2021, doi: 10.1109/TSE.2021.3063727.
[92] J. Chen, L. Du, and Y. Guo, "Label constrained convolutional factor analysis for classification with
limited training samples," Information Sciences, vol. 544, pp. 372-394, 2021.
[93] L. Zhao and L. Wang, "A new lightweight network based on MobileNetV3," KSII Transactions on
Internet and Information Systems (TIIS), vol. 16, no. 1, pp. 1-15, 2022.
[94] H. Yan and Z. Wu, "A Hybrid Short-Term Wind Power Prediction Model Combining Data
Processing, Multiple Parameters Optimization and Multi-Intelligent Models Apportion Strategy,"
 IEEE ACCESS, Article vol. 8, pp. 227126-227140, 2020 2020, doi:
10.1109/ACCESS.2020.3046001.
[95] W. Zheng, X. Liu, X. Ni, L. Yin, and B. Yang, "Improving visual reasoning through semantic
representation," IEEE access, vol. 9, pp. 91476-91486, 2021.
[96] A. Karuppannan and M. Muthusamy, "Wavelet neural learning-based type-2 fuzzy PID controller
for speed regulation in BLDC motor," NEURAL COMPUTING & APPLICATIONS, Article vol.
33, no. 20, pp. 13481-13503, 2021 OCT 2021, doi: 10.1007/s00521-021-05971-2.
[97] S. L. Zubaidi et al., "Hybridised Artificial Neural Network Model with Slime Mould Algorithm: A
Novel Methodology for Prediction of Urban Stochastic Water Demand," WATER, Article vol. 12,
no. 10, 2020 OCT 2020, doi: 10.3390/w12102692.
[98] Y. Guo, M. Khishe, M. Mohammadi, S. Rashidi, and M. S. Nateri, "Evolving Deep Convolutional
Neural Networks by Extreme Learning Machine and Fuzzy Slime Mould Optimizer for Real-Time
Sonar Image Recognition," INTERNATIONAL JOURNAL OF FUZZY SYSTEMS, Article
Early Access 2021, doi: 10.1007/s40815-021-01195-7.
[99] P. Wu et al., "An Effective Machine Learning Approach for Identifying Non-Severe and Severe
Coronavirus Disease 2019 Patients in a Rural Chinese Population: The Wenzhou Retrospective
Study," IEEE ACCESS, Article vol. 9, pp. 45486-45503, 2021 2021, doi:
10.1109/ACCESS.2021.3067311.
[100] Y. Liu, Y. Sun, B. Xue, M. Zhang, G. G. Yen, and K. C. Tan, "A survey on evolutionary neural
architecture search," IEEE transactions on neural networks and learning systems, 2021.
[101] W. Liu, R. Wang, T. Zhang, K. Li, W. Li, and H. Ishibuchi, "Hybridization of evolutionary algorithm
and deep reinforcement learning for multi-objective orienteering optimization," IEEE Transactions
on Evolutionary Computation, 2022.
[102] H. Abdel-Mawgoud, S. Kamel, H. Rezk, T. K. Lone, S. B. J. C. Rhee, Materials, and Continua,
"Optimal Implementation of Photovoltaic and Battery Energy Storage in Distribution Networks,"
vol. 69, no. 2, pp. 1463–1481, 2021.
[103] V. K. Kamboj et al., "A Cost-Effective Solution for Non-Convex Economic Load Dispatch
Problems in Power Systems Using Slime Mould Algorithm," vol. 14, 2022.
[104] S. Mouassa, A. Althobaiti, F. Jurado, and S. S. M. Ghoneim, "Novel Design of Slim Mould
Optimizer for the Solution of Optimal Power Flow Problems Incorporating Intermittent Sources:
A Case Study of Algerian Electricity Grid," IEEE Access, vol. 10, pp. 22646-22661, 2022, doi:
10.1109/ACCESS.2022.3152557.
[105] Y. Wei, Y. Zhou, and Q. J. S. S. E. P. Luo, "Optimal Reactive Power Dispatch Problem Using
Improved Slime Mould Algorithm."
[106] A. Draz, M. M. Elkholy, and A. A. El-Fergany, "Slime mould algorithm constrained by the relay
operating time for optimal coordination of directional overcurrent relays using multiple
standardized tripping curves," 2021.
[107] R. Kouadri, L. Slimani, and T. J. S. S. E. P. Bouktir, "Slime Mould Algorithm for Practical Optimal
Power Flow Solutions Incorporating Stochastic Wind Power and Static VAR Compensator Device."
[108] K. Nusair, F. Alasali, A. Hayajneh, and W. J. I. J. o. E. R. Holderbaum, "Optimal placement of
FACTS devices and power-flow solutions for a power network system integrated with stochastic
renewable energy resources using new metaheuristic optimization techniques," 2021.
[109] F. F. Amigue, S. N. Essiane, S. P. Ngoffe, G. A. Ondoa, G. M. Mengounou, and P. T. Nna Nna,
"Optimal integration of photovoltaic power into the electricity network using Slime mould
 algorithms: Application to the interconnected grid in North Cameroon," Energy Reports, vol. 7, pp.
6292-6307, 2021/11/01/ 2021, doi: https://doi.org/10.1016/j.egyr.2021.09.077.
[110] A. A. Bhandakkar and L. Mathew, "Merging slime mould with whale optimization algorithm for
optimal allocation of hybrid power flow controller in power system," Journal of Experimental &
Theoretical Artificial Intelligence, pp. 1-28, 2022, doi: 10.1080/0952813X.2022.2040598.
[111] S. Khunkitti, A. Siritaratiwat, and S. J. S. Premrudeepreechacharn, "Multi-Objective Optimal Power
Flow Problems Based on Slime Mould Algorithm," vol. 13, 2021.
[112] M. Farhat, S. Kamel, A. M. Atallah, M. H. Hassan, A. M. Agwa, and M. S. Li, "ESMA-OPF:
Enhanced Slime Mould Algorithm for Solving Optimal Power Flow Problem," 2022.
[113] M. Mostafa, H. Rezk, M. Aly, E. M. J. S. E. T. Ahmed, and Assessments, "A new strategy based on
slime mould algorithm to extract the optimal model parameters of solar PV panel," vol. 42, pp. 1-
11, 2020.
[114] C. Kumar, P. Manoharan, T. D. Raj, T. J. O.-I. J. f. L. Dhanesh, and E. Optics, "A New Stochastic
Slime Mould Optimization Algorithm for the Estimation of Solar Photovoltaic Cell Parameters,"
2020.
[115] D. Yousri, M. Aziz, D. Oliva, L. Abualigah, M. J. E. C. Al-Qaness, and Management, "Reliable
applied objective for identifying simple and detailed photovoltaic models using modern
metaheuristics: Comparative study," vol. 223, 2020.
[116] A. A. J. E. R. El-Fergany, "Parameters identification of PV model using improved slime mould
optimizer and Lambert W-function," vol. 7, pp. 875-887, 2021.
[117] T. Gush, C. H. Kim, S. Admasie, J. S. Kim, and J. S. J. I. A. Song, "Optimal Smart Inverter Control
for PV and BESS to Improve PV Hosting Capacity of Distribution Networks Using Slime Mould
Algorithm," 2021.
[118] D. Yousri, A. Fathy, H. Rezk, T. S. Babu, M. R. J. E. C. Berber, and Management, "A reliable
approach for modeling the photovoltaic system under partial shading conditions using three diode
model and hybrid marine predators-slime mould algorithm," p. 114269, 2021.
[119] A. F. Mirza, M. Mansoor, K. Zhan, and Q. J. E. Ling, "High-efficiency swarm intelligent maximum
power point tracking control techniques for varying temperature and irradiance," vol. 228, 2021.
[120] Y. Liu, A. A. Heidari, X. Ye, G. Liang, and C. J. E. He, "Boosting Slime Mould Algorithm for
Parameter Identification of Photovoltaic Models," no. 5, p. 121164, 2021.
[121] X. Weng, A. A. Heidari, G. Liang, H. Chen, and X. Ma, "An evolutionary Nelder–Mead slime mould
algorithm with random learning for efficient design of photovoltaic models," Energy Reports, vol. 7,
pp. 8784-8804, 2021/11/01/ 2021, doi: https://doi.org/10.1016/j.egyr.2021.11.019.
[122] H. Lin et al., "Adaptive slime mould algorithm for optimal design of photovoltaic models," 2022.
[123] P. T. Ha, D. T. Tran, T. T. J. N. C. Nguyen, and Applications, "Electricity generation cost reduction
for hydrothermal systems with the presence of pumped storage hydroelectric plants," vol. 34, no.
12, pp. 9931-9953, 2022.
[124] J. Gupta, P. Nijhawan, S. J. E. S. Ganguli, and P. Research, "Optimal sizing of different configuration
of photovoltaic, fuel cell, and biomass-based hybrid energy system," vol. 29, no. 12, pp. 17425-
17440, 2021.
[125] H. A. El-Sattar, H. M. Sultan, S. Kamel, T. Khurshaid, and C. Rahmann, "Optimal design of stand-
alone hybrid PV/wind/biomass/battery energy storage system in Abu-Monqar, Egypt," Journal of
Energy Storage, vol. 44, p. 103336, 2021/12/01/ 2021, doi:
https://doi.org/10.1016/j.est.2021.103336.
[126] A. M. Othman and A. A. J. R. E. El-Fergany, "Optimal dynamic operation and modeling of parallel
connected multi-stacks fuel cells with improved slime mould algorithm," vol. 175, 2021.
[127] L. C. Kien, T. T. Nguyen, T. D. Pham, T. T. J. N. C. Nguyen, and Applications, "Cost reduction for
energy loss and capacitor investment in radial distribution networks applying novel algorithms," pp.
1-28, 2021.
[128] T. J. A. I. Singh, "Chaotic slime mould algorithm for economic load dispatch problems," 2022.
[129] R. M. Rizk-Allah, A. E. Hassanien, and D. J. I. T. Song, "Chaos-opposition-enhanced slime mould
algorithm for minimizing the cost of energy for the wind turbines on high-altitude sites," no. 15,
2021.
[130] D. Rabah, B. Bentouati, M. S. Javaid, H. Bouchekara, R. A. J. J. o. M. El-Sehiemy, and M. Materials,
"Mitigating the effects of magnetic coupling between HV Transmission Line and Metallic Pipeline
using Slime Mould Algorithm," 2021.
[131] S. Ponlatha, P. Umasankar, P. Balashanmuga Vadivu, and D. Chitra, "An IOT-based efficient energy
management in smart grid using SMACA technique," International Transactions on Electrical Energy
Systems, https://doi.org/10.1002/2050-7038.12995 vol. 31, no. 12, p. e12995, 2021/12/01 2021,
doi: https://doi.org/10.1002/2050-7038.12995.
[132] M. H. Hassan, S. Kamel, L. Abualigah, and A. Eid, "Development and application of slime mould
algorithm for optimal economic emission dispatch," Expert Systems with Applications, vol. 182, p.
115205, 2021/11/15/ 2021, doi: https://doi.org/10.1016/j.eswa.2021.115205.
[133] D. K. Patra, T. Si, S. Mondal, and P. Mukherjee, "Breast DCE-MRI segmentation for lesion
detection by multi-level thresholding using student psychological based optimization,"
BIOMEDICAL SIGNAL PROCESSING AND CONTROL, Article vol. 69, 2021 AUG 2021, doi:
10.1016/j.bspc.2021.102925.
[134] R. Kurban, A. Durmus, and E. Karakose, "A comparison of novel metaheuristic algorithms on
color aerial image multilevel thresholding," ENGINEERING APPLICATIONS OF ARTIFICIAL
INTELLIGENCE, Article vol. 105, 2021 OCT 2021, doi: 10.1016/j.engappai.2021.104410.
[135] D. Karaboga, B. Gorkemli, C. Ozturk, and N. Karaboga, "A comprehensive survey: artificial bee
colony (ABC) algorithm and applications," Artificial Intelligence Review, vol. 42, no. 1, pp. 21-57, 2014.
[136] A. A. Heidari, S. Mirjalili, H. Faris, I. Aljarah, M. Mafarja, and H. Chen, "Harris hawks optimization:
Algorithm and applications," Future Generation Computer Systems, vol. 97, pp. 849-872, 2019/08/01/
2019, doi: https://doi.org/10.1016/j.future.2019.02.028.
[137] C. Blum, "Ant colony optimization: Introduction and recent trends," Physics of Life reviews, vol. 2,
no. 4, pp. 353-373, 2005.
[138] C. He, S. Huang, R. Cheng, K. C. Tan, and Y. Jin, "Evolutionary multiobjective optimization driven
by generative adversarial networks (GANs)," IEEE transactions on cybernetics, vol. 51, no. 6, pp. 3129-
3142, 2020.