Future Generation Computer Systems 111 (2020) 300–323

Contents lists available at ScienceDirect

Future Generation Computer Systems

journal homepage: www.elsevier.com/locate/fgcs

Slime mould algorithm: A new method for stochastic optimization

∗
Shimin Li a , Huiling Chen a , , Mingjing Wang e , Ali Asghar Heidari b,c , Seyedali Mirjalili d
a
College of Computer Science and Artificial Intelligence, Wenzhou University, Wenzhou, Zhejiang 325035, China
b
School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Tehran 1439957131, Iran
c
Department of Computer Science, School of Computing, National University of Singapore, Singapore 117417, Singapore
d
Centre for Artificial Intelligence Research and Optimisation, Torrens University Australia, Fortitude Valley, Brisbane, 4006, QLD, Australia
e
Institute of Research and Development, Duy Tan University, Da Nang 550000, Viet Nam

article info a b s t r a c t

Article history: In this paper, a new stochastic optimizer, which is called slime mould algorithm (SMA), is proposed
Received 6 August 2019 based on the oscillation mode of slime mould in nature. The proposed SMA has several new features
Received in revised form 16 February 2020 with a unique mathematical model that uses adaptive weights to simulate the process of producing
Accepted 29 March 2020
positive and negative feedback of the propagation wave of slime mould based on bio-oscillator
Available online 3 April 2020
to form the optimal path for connecting food with excellent exploratory ability and exploitation
Keywords: propensity. The proposed SMA is compared with up-to-date metaheuristics using an extensive set
Slime mould algorithm of benchmarks to verify its efficiency. Moreover, four classical engineering problems are utilized to
Adaptive weight estimate the efficacy of the algorithm in optimizing constrained problems. The results demonstrate
Engineering design problems that the proposed SMA benefits from competitive, often outstanding performance on different search
Constrained optimization landscapes. The source codes of SMA are publicly available at http://www.alimirjalili.com/SMA.html
and https://tinyurl.com/Slime-mould-algorithm.
© 2020 Elsevier B.V. All rights reserved.

1. Introduction the solution space of many problems is often indeterminate

or infinite. It may be infeasible to find optimal solutions by
Metaheuristic algorithms (MAs) have become prevalent in traversing the solution space under current circumstances. MAs
many applied disciplines in recent decades because of higher per- detect the proximate optimal solution of the problem by sampling
formance and lower required computing capacity and time than the enormous solution space randomly in a certain way, to find
deterministic algorithms in various optimization problems [1–4]. or generate better solutions for the optimization problem under
Simple concepts are required to achieve favourable results, and limited circumstances or computational capacity.
it is facile to transplant to different disciplines. Also, the lack of MAs are typically inspired by real-world phenomena to find
randomness in the later stage of some deterministic algorithm better heuristic solutions for optimization problems by simu-
makes it inclined to sink into local optimum, and random factors lating physical rules or biological phenomena. MAs can be di-
in MAs can make the algorithm search for all optimal solutions vided into two main categories: swam-based methods and evo-
in search space, thus effectively avoiding local optimum. In linear lutionary techniques. The first kind mainly simulates physical
problems, some gradient descent algorithms, such as work in [5] phenomena, apply mathematical rules or methodologies includ-
are more efficient than stochastic algorithms for the utilization ing Multi-Verse Optimizer (MVO) [6], Charged System Search
of gradient information. The convergence speed of MAs will be (CSS) [7], Gravitational Search Algorithm (GSA) [8], Sine Co-
less than gradient descent algorithms and can be considered sine Algorithm (SCA) [9], Teaching-Learning-Based Optimization
as a drawback. In non-linear problems, however, MAs typically (TLBO) [10], Central Force Optimization (CFO) [11] and Tabu
commence the optimization process with randomly generated Search (TS) [12]. Nature-inspired methods mainly include two
solutions and do not demand gradient information, which makes types: evolutionary methods and intelligent swarm techniques.
the algorithm eminently suitable for practical problems when The inspiration of the evolutionary algorithm (EA) originates from
the derivative information is unknown. In real-world scenarios, the process of biological evolution in nature. Compared with
the traditional optimization algorithm, it is a global optimization
∗ Corresponding author. method with better robustness and applicability.
Some of the widespread algorithms in the class of EA are
E-mail addresses: simonlishimin@foxmail.com (S. Li),
chenhuiling.jlu@gmail.com (H. Chen), mingjingwang@duytan.edu.vn (M. Wang),
Genetic Algorithm (GA) [13], Genetic Programming (GP) [14],
as_heidari@ut.ac.ir, aliasgha@comp.nus.edu.sg, t0917038@u.nus.edu, Evolution Strategy (ES) [15], Evolutionary Programming (EP) [16]
aliasghar68@gmail.com (A.A. Heidari), ali.mirjalili@gmail.com (S. Mirjalili). and Differential Evolution (DE) [17]. The application of ES and EP

https://doi.org/10.1016/j.future.2020.03.055
0167-739X/© 2020 Elsevier B.V. All rights reserved.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 301

of slime mould. An overall set of 33 benchmarks and four fa-

mous manufacturing design problems has rigorously verified the
effectiveness and robustness of SMA.
The remainder of the paper is structured as below. Section 2 il-
lustrated the concept of the source of slime mould algorithm, and
the mathematical model was established. Section 3 firstly gave a
qualitative analysis of the algorithm and made a comprehensive
comparison of 33 benchmark functions, then tested it on four
engineering design problems. Section 4 summarized the whole
work and put forward some inspirations for future work.

2. Slime mould algorithm

In this section, the basic concept and conduct of slime mould

will be introduced. Then a mathematical model inspired by its
behaviour pattern will be established.

2.1. Originality

Before this article, some scholars have proposed similar nam-

Fig. 1. Classification of evolutionary and SI methods. ing algorithms, but the way of designing the algorithm and us-
age scenarios are quite different from the algorithms proposed
in this paper. Monismith and Mayfield [32] solves the single-
and also swarm-intelligence methods in scientific research and objective optimization problem by simulating the five life cycles
practical problems is also becoming more and more extensive of amoeba Dictyostelium discoideum: a state of vegetative, ag-
[18–20]. Swarm Intelligence (SI) [21] includes a collective or gregative, mound, slug, or dispersal while using ε -ANN to con-
social intelligence that artificially simulates the decentralization struct an initial position-based mesh. Li et al. [33] proposed
of biological clusters in nature or the collective behaviour of self- a method to construct wireless sensor networks by using two
organizing systems. In this class of algorithms, the inspiration forms of slime mould tubular networks to correspond to two
usually comes from biological groups in nature that have col- different regional routing protocols. Qian. et al. [34] combined the
lective behaviour and intelligence to achieve a certain purpose. Physarum network with the ant colony system to improve the
Established and recent algorithms in this class are Particle Swarm algorithm’s competence to avoid local optimal values to handle
Optimization (PSO) [22], Bat-inspired Algorithm (BA) [23], Grey the Travelling Salesman Problem better. Inspired by the diffu-
Wolf Optimization (GWO) [24], Fruit Fly Optimization (FOA) [25], sion of slime mould, Schmickland Crailsheim [35] proposed a
bio-inspired navigation principle designed for swarm robotics.
Moth Flame Optimization (MFO) [26], Ant Colony Optimization
Becker [36] generated inexpensive and fault-tolerant graphs by
(ACO) [27], Harris Hawk Optimizer (HHO) [28], and Artificial Bee
simulating the foraging process of the slime mould Physarum
Colony (ABC) [29]. A schematic design for the classification of
polycephalum. As can be seen from the above discussion, most
evolutionary and SI methods are shown in Fig. 1.
of the modelled slime mould algorithms were used in graph the-
Although different MAs have some distinctness, they all have
ory and generation networks. This living organism also inspired
two identical stages in the search gradation: exploration and
researchers in the area of graph optimization [37]. The algorithm
exploitation [30]. The exploration phase refers to the process
used to optimize the problem [32] simulates the five life cycles
of searching solution space as widely, randomly, and globally
of amoeba Dictyostelium discoideum, but the experiments and
as possible. Exploitation phase refers to the competence of the
proofs in the article are slightly less.
algorithm to search more accurately in the area acquired by
The SMA proposed in this paper mainly simulates the be-
the exploration phase, and its randomness decreases while its
haviour and morphological changes of slime mould Physarum
precision increases. When the exploration ability of the algorithm polycephalum in foraging and does not model its complete life
is dominant, it can search the solution space more randomly cycle. At the same time, the use of weights in SMA is to simulate
and produce more differentiated solution sets to converge fleetly. the positive and negative feedback generated by slime mould
When the exploitative ability of the algorithm is dominant, it during foraging, thus forming three different morphotype, is a
searches more locally to enhance the quality and precision of the brand new idea. This paper also conducted a full experiment
solution sets. However, when the exploration facility is improved, on the characteristics of the algorithm. The results in the next
it will lead to reductions in the exploitation capability, and vice sections demonstrate the superiority of the SMA algorithm.
versa. Another challenge is that the balance of these two abilities
is not necessarily identical to different problems. Therefore, it is 2.2. Concepts
relatively challenging to attain an appropriate balance between
the two phases that are efficient for all optimization problems. The slime mould mentioned in this article generally refers to
Despite the success of conventional and recent MAs, none the Physarum polycephalum. Because it was first classified as a
of them can guarantee to find the global optimum for all op- fungus, thus it was named ‘‘slime mould’’, which its life cycle
timization problems. This has been proven logically by the No- was originally specified by Howard [38] in a paper published in
Free-Lunch (NFL) theory [31]. This theorem motivated numerous 1931. Slime mould is a eukaryote that inhabits cold and humid
researchers to design a new algorithm and solve new classes of places. The main nutritional stage is Plasmodium, the active and
problems more efficiently. With the aspiration of proposing a dynamic stage of slime mould, and also the main research stage
more versatile and efficient algorithm, this paper introduces a of this paper. In this stage, the organic matter in slime mould
new meta-heuristic algorithm: The slime mould algorithm (SMA). seeks food, surrounds it, and secretes enzymes to digest it. During
This method is aroused by the diffusion and foraging conduct the migration process, the front end extends into a fan-shaped,
302 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

thus ensuring that they get the maximum concentration of nu-

trients. Recent studies have also revealed that slime mould has
the competence of making foraging arrangements based on opti-
mization theory [47]. When the quality of various food sources
is different, slime mould can choose the food source with the
highest concentration. However, slime mould also needs to weigh
speed and risk in foraging. For instance, slime mould needs to
make faster decisions in order to avoid environmental damage
to them. Experiments have shown that the quicker the decision-
making speed is, the possibilities of slime mould to find the prime
food source is smaller [48]. Therefore, when deciding the source
of food, slime mould obviously needs to weigh the speed and
accuracy.
Slime mould needs to decide when to leave this area and
search in another area when foraging. When lacking complete
information, the best way for a slime mould to estimate when
to leave the current position is to adopt heuristic or empirical
rules based on the insufficient information currently available.
Experience has shown that when slime mould encounter high-
quality food, the probability of leaving the area is reduced [49].
However, due to its unique biological characteristics, slime mould
can utilize a variety of food sources at the same time. Therefore,
even if the slime mould has found a better source of food, it can
Fig. 2. Foraging morphology of slime mould. still divide a component of the biomass to exploit both resources
simultaneously when higher quality food is found [46].
Slime mould can also dynamically adjust their search pat-
terns according to the quality of foodstuff provenience. When
followed by an interconnected venous network that allows cyto-
the quality of food sources is high, the slime mould will use
plasm to flow inside [39], as shown in Fig. 2. Because of their
the region-limited search method [50], thus focusing the search
unique pattern and characteristic, they can use multiple food
on the food sources that have been found. If the denseness of
sources at the same time to form a venous network connecting
the food provenience initially found is low, the slime mould will
them. If there is enough food in the environment, slime mould
leave the food source to explore other alternative food sources
can even grow to more than 900 square centimetres [39].
in the region [51]. This adaptive search strategy can be more
Owing to the feature of slime mould can be easily cultured on
reflected when different quality food blocks are dispersed in a
agar and oatmeal [40], they were widely used as model organ- region. Some of the mechanisms and characteristics of the slime
isms. Kamiya and his colleagues [41] was the first team to study mould mentioned above will be mathematically modelled in the
the detailed process of the cytoplasmic flow of slime mould. Their subsequent sections.
work is of great help to our subsequent understanding of the
way slime mould move and connects food in the environment. 2.3. Mathematical model
We now cognize that when a vein approaches a food source, the
bio-oscillator produces a propagating wave [42] that increases the In this part, the mathematical model and method proposed
cytoplasmic flow through the vein, and the faster the cytoplasm will be described in details.
flows, the thicker the vein. Through this combination of positive-
negative feedback, the slime can establish the optimal path to 2.3.1. Approach food
connect food in a relatively superior way. Therefore, slime mould The slime mould can approach food according to the odour
was also mathematically modelled and applied in graph theory in the air. To express its approaching behaviour in mathematical
and path networks [43–45]. formulae, the following formulae are proposed to imitate the
The venous structure of slime mould develops along with the contraction mode:
phase difference of the contraction mode [42], so there are three −−→ − → (− → −−→ −−→)
⎧
−−−−−→ ⎨ X b (t ) + v b · W · XA (t ) − XB (t ) , r < p
correlations between the morphological changes of the venous X (t + 1) = (2.1)
structure and the contraction mode of slime mould. −
→ −−→
v c · X (t ), r ≥ p
⎩
(1) Thick veins form roughly along the radius when the con-
−
→ −
→
traction frequencies vary from outside to inside. where v b is a parameter with a range of [−a, a], v c decreases
(2) When the contraction mode is unstable, anisotropy begins linearly from one to zero. t represents the current iteration,
to appear. −
→
Xb represents the individual location with the highest odour
(3) When the contraction pattern of slime mould is no longer concentration currently found, X⃗ represents the location of slime
ordered with time and space, the venous structure is no longer −
→ −→
mould, XA and XB represent two individuals randomly selected
present. −
→
from slime mould, W represents the weight of slime mould.
Therefore, the relationship between venous structure and con-
The formula of p is as follows:
traction pattern of slime mould is consistent with the shape
of naturally formed cells. The thickness of each vein is deter- p = tanh |S (i) − DF | (2.2)
mined by the flow feedback of the cytoplasm in the Physarum
where i ∈ 1, 2, . . . , n, S (i) represents the fitness of X
⃗ , DF repre-
solver [46]. The rise in the flow of cytoplasm leads to an in-
sents the best fitness
−→ obtained in all iterations.
crease in the diameter of veins. As the flow decreases, the veins
The formula of v b is as follows:
contract because of the decrease of the diameter. Slime mould
−
→
can build a stronger route where food concentration is higher, v b = [−a, a] (2.3)
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 303

Fig. 3. Possible locations in 2-dimension and 3-dimension..

( )
t
a = arctanh(− + 1) (2.4)
max _t
−
→
The formula of W is listed as follows:
bF − S (i)
⎧ ( )
⎪1 + r · log + 1 , condition
bF − w F
⎪
−−−−−−−−−−−→ ⎨
W (SmellIndex(i)) =
bF − S (i)
( )
+ 1 , others
⎪
⎩1 − r · log
⎪
bF − w F
(2.5)
SmellIndex = sort(S) (2.6)
where condition indicates that S (i) ranks first half of the popula-
tion, r denotes the random value in the interval of [0, 1], maxt
shows maximum iteration, bF denotes the optimal fitness ob-
tained in the current iterative process, w F denotes the worst fit-
ness value obtained in the iterative process currently, SmellIndex
denotes the sequence of fitness values sorted(ascends in the
minimum value problem).
Fig. 3 visualizes the effects of Eq. (2.1). The location of search- Fig. 4. Assessment of fitness.

ing individual X⃗ can be updated according to the best location − →

Xb
−
→ − →
currently obtained, and the fine-tuning of parameters v b, v c and
−
→ frequency does not change too much. condition simulates the
W can change the location of the individual. Fig. 3 is also used to
slime mould to adjust their search patterns according to the
illustrate the position change of the searching individual in three-
quality of food. When the food concentration is content, the
dimensional space. randin the formula can make individuals form
weight near the region is bigger; when the food concentration
search vectors at any angle, that is, search solution space in any
is low, the weight of the region will be reduced, thus turning
direction, so that the algorithm has the possibility of finding
to explore other regions. Fig. 4 shows the process of evaluating
the optimum solution. Therefore, Eq. (2.1) enables the searching
fitness values for slime mould.
individuals to search in all possible directions near the optimal
solution, thus simulating the circular sector structure of slime Based on the above principle, the mathematical formula for
mould when approaching food. It is also applicable to extend this updating the location of slime mould is as follows:
concept to Hyper-dimensional space. ⎪rand · (UB − LB ) + LB, rand < z )
⎧
→∗ ⎨−−→ −
− → ( −−→ −−→
2.3.2. Wrap food X = Xb (t ) + v b · W · XA (t ) − XB (t ) , r < p (2.7)
⎩−
⎪
→ −−→
This part simulates the contraction mode of venous tissue v c · X (t ), r ≥ p
structure of slime mould mathematically when searching. The
higher the concentration of food contacted by the vein, the where LB and UB denote the lower and upper boundaries of
stronger the wave generated by the bio-oscillator, the faster search range, rand and r denote the random value in [0,1]. The
the cytoplasm flows, and the thicker the vein. Eq. (2.5) mathe- value of z will be discussed in the parameter setting experiment.
matically simulated the positive and negative feedback between
the vein width of the slime mould and the food concentration 2.3.3. Oscillation
that was explored. The component rin Eq. (2.5) simulates the Slime mould mainly depends on the propagation wave pro-
uncertainty of venous contraction mode. log is used to alleviate duced by the biological oscillator to change the cytoplasmic flow
the change rate of numerical value so that the value of contraction in veins, so that they tend to be in a better position of food
304 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Table 1
Unimodal and multimodal test functions of 23 standard benchmarks.
Functions Dim Range fmin
∑n
f1 (x) = 2
i=1 xi n [−100,100] 0
∑n
f2 (x) = | xi | + ni=1 | xi |
∏
i=1 n [−10,10] 0
∑n (∑ )2
i
f3 (x) = i=1 j−1 xj n [−100,100] 0

f4 (x) = maxi {|[xi |, 1 ≤ i ≤ n} ] n [−100,100] 0

∑n−1 )2
f5 (x) = + (xi − 1)2
(
i=1 100 xi+1 − x2i n [−30,30] 0
∑n
f6 (x) = i=1 ([xi + 0.5])
2
n [−100,100] 0
∑n
f7 (x) = i=1 ixi + random [0, 1]
4
n [−128,128] 0
∑n (√
f8 (x) =
)
i=1 −xi sin | xi | n [−500,500] −418.9829*n
∑n [ 2
f9 (x) = i=1 xi − 10 cos (2π xi ) + 10
]
n [−5.12,5.12] 0
n
( 1 ∑n
f10 (x) = −20 exp(−0.2( 1n i=1 xi ) 0.5) − exp n i=1 cos (2π xi ) + 20 + e
∑ 2 ∧
)
n [−32,32] 0
∑n 2 ∏n ( )
x
f11 (x) = 4000
1
i=1 xi − i=1 cos
√i +1 n [−600,600] 0
i
{ }
f12 (x) = πn 10 sin (π y1 ) + i=1 (yi − 1)2 1 + 10 sin2 (π yi+1 ) + (yn − 1)2
n− 1
∑ [ ]

+ ni=1 u (xi , 10, 100, 4) , yi = 1 + xi + 1

∑
⎧ 4

⎪k (xi − a)m xi > a

⎪
⎪ n [−50,50] 0
⎨
u (xi , a, k, m) = 0 − a < xi < a
⎪
⎪
⎪
k (−xi − a)m xi < a
⎩
∑n
f13 (x) = 0.1 sin (3π x1 ) + i=1 (xi − 1)2 1 + sin2 (3π xi + 1) + (xn − 1)2 [1+
{ 2 [ ]
]} ∑n n [−50,50] 0
sin (2π xn ) + i=1 u (xi , 5, 100, 4)
2

concentration. For the purpose of simulating the variations of

−
→ −
→ −
→
venous width of slime mould, we used W , v b and v c to realize
the variations.
−
→
W mathematically simulates the oscillation frequency of slime
mould near one at different food concentration, so that slime
mould can approach food more quickly when they find high-
quality food, while approach food more slowly when the food
concentration is lower in individual position, thus improving the
efficiency of slime mould in choosing the optimal food source.
−
→
The value of v boscillates randomly between [−a, a] and grad-
ually approaches zero as the increasement of iterations. The value
−
→
of v c oscillates between [−1,1] and tends to zero eventually. The
trend of the two values is shown as Fig. 5. Synergistic interaction
−
→ −
→
between v b and v c mimics the selective behaviour of slime
mould. In order to find a better source of food, even if slime
mould has found a better source of food, it will still separate some
organic matter for exploring other areas in an attempt to find a 2.3.4. Computational complexity analysis
higher quality source of food, rather than investing all of it in one SAM mainly consists of the subsequent components: initial-
source. ization, fitness evaluation, sorting, weight update, and location
−→
Moreover, the oscillation process of v bsimulates the state of update. Among them, N denotes the number of cells of slime
slime mould deciding whether to approach the food source or mould, D denotes the dimension of functions, and T denotes
find other food sources. Meanwhile, the process of probing food the maximum number of iterations. The computation complexity
is not smooth. During this period, there may be various obstacles, of initialization is O(D), the computation complexity of fitness
such as light and dry environment, which restrict the spread of evaluation and sorting is O(N + N log N), the computational
slime mould. However, it also improves the possibility of slime complexity of weight update is O(N × D), the complexity of
location update is O (N × D). Therefore, the total complexity of
mould to find higher quality food and evades the trapping of local
SMA is O(D + T ∗ N ∗ (1 + log N + D)).
optimum.
The pseudo-code of the SMA is shown in Algorithm 1. The
3. Experimental results and analyses
intuitive and detailed process of SMA is shown in Fig. 6. The
General logic of SMA is also shown in Fig. 7.
In this sector, we compared the SMA with some competitive
There are still many mechanisms that can be added to the MAs on an all-inclusive set of 33 benchmark cases. The experi-
algorithm, or more comprehensive simulation of the life cycle of mentations were ran on the operating system of Windows Server
slime mould. However, to enhance the extensibility of the algo- 2012 R2 Datacenter with 128 GB RAM and CPU of Intel (R) Xeon
rithm, we simplify the process and operators of the algorithm, (R) E5−2650 v4 (2.20 GHz). The algorithms for comparison were
leaving only the simplest algorithm as possible. coded by MATLAB R2018b.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 305

Table 2
Unimodal and simple multimodal functions of CEC2014.
Functions Dim Range fmin
f14 (x) = Rotated High Conditioned Elliptic Function n [−100,100] 100
f15 (x) = Rotated Bent Cigar Function n [−100,100] 200
f16 (x) = Shifted and Rotated(SR) Ackley’s Function n [−100,100] 500
f17 (x) = SR Weierstrass Function n [−100,100] 600
f18 (x) = SR HappyCat Function n [−100,100] 1300
f19 (x) = SR HGBat Function n [−100,100] 1400
f20 (x) = SR Expanded Griewank’splus Rosenbrock‘s Function n [−100,100] 1500
f21 (x) = SR Expanded Scaffer’sF6 Function n [−100,100] 1600

From the search history subplot, the slime mould in different

benchmark functions put up a similar cross-type search trajectory
clustered near the optimal value, thus accurately searching in
reliable search areas and reflecting fast convergence. Meanwhile,
the distribution of slime mould is mainly concentrated in multiple
regions with local optimum, which shows the tradeoff of slime
mould between multiple local optimums.
The trajectory of the first slime mould in the first dimension
can be used as a representative of other parts of slime mould,
revealing the primary exploratory behaviour of the slime mould.
The fast oscillation in the prophase and the slight oscillation in
the anaphase can ensure the fast convergence of slime mould
and the accurate search near the optimal solution [52]. As can
be perceived from the figure, the position curve of slime mould
−
→ −
→
Fig. 5. Trends of v b and v c . has a very large amplitude in the early iteration process, even up
to 50% of the exploration space. In the later iteration period, if the
function is smooth, the amplitude of the position of slime mould
3.1. Qualitative analysis begins to decrease; if the amplitude of the function changes sig-
nificantly, the position amplitude also changes much. This reflects
The qualitative analysis results of SMA in handling unimodal the high adaptability and robustness of slime mould in different
functions and multimodal functions are presented in Fig. 8 to in- functions.
tuitively analyse the position and fitness changes of slime mould By observing the average fitness curve, the variation tendency
during foraging. The figure is comprised of four concernment of the fitness of slime mould during the iterative procedure can
indicators: search history, the trajectory of the slime mould in be visually observed. Although the average fitness curve of slime
the 1st dimension, the average fitness of slime mould, and con- mould is oscillating, the average fitness value tends to decrease,
vergence curve. Search history represents the location and dis- and the oscillation frequency decreases inversely proportional to
tribution of slime mould in the iteration process. The trajectory iterations, thus ensuring the rapid convergence of slime mould in
of slime mould reveals the behaviour of the position change of the prophase and the precise search in the anaphase.
slime mould in the first part of the first dimension. Average The convergence curve reveals the average fitness of the opti-
fitness indicates the variation trend of the average fitness of mal fitness value searched by slime mould varies with iterations.
the slime mould colony changes with the iteration process. The By observing the downtrend of the curve, we can observe the
convergence curve shows the optimal fitness value in the slime convergence rate of slime mould and the time when it switches
mould during the iteration process. between the exploration and exploration gradation.

Fig. 6. Flowchart of SMA.

306 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Fig. 7. The steps of SMA.

Table 3 Table 4
Hybrid and composition functions of CEC 2014. Parameter settings of counterparts.
Functions Dim Range fmin Algorithm Parameter settings
f22 (x) = Hybrid Function 1 n [−100,100] 1700 WOA a1 = [2, 0] ; a2 = [−2, −1] ; b = 1
f23 (x) = Hybrid Function 2 n [−100,100] 1800 GWO a = [2, 0]
f24 (x) = Hybrid Function 3 n [−100,100] 1900 MFO b = 1; t = [−1, 1] ; a ∈ [−1, −2]
f25 (x) = Hybrid Function 4 n [−100,100] 2000 BA A = 0.5; r = 0.5
f26 (x) = Hybrid Function 5 n [−100,100] 2100 SCA A=2
f27 (x) = Hybrid Function 6 n [−100,100] 2200 FA α = 0.5; β = 0.2; γ = 1
f28 (x) = Composite function 1 n [−100,100] 2300 PSO c1 = 2; c2 = 2; v Max = 6
f29 (x) = Composite function 2 n [−100,100] 2400 SSA c1 ∈ [01] ; c2 ∈ [01] ;
f30 (x) = Composite function 3 n [−100,100] 2500 MVO Existence probability ∈ [0.21] ; trav elling distance rate ∈
f31 (x) = Composite function 4 n [−100,100] 2600 [0.61]
f32 (x) = Composite function 5 n [−100,100] 2700 ALO k = 500
f33 (x) = Composite function 6 n [−100,100] 2800 PBIL Learning rate = 0.05; elitism parameter = 1;
probability v ector mutation rate = 0
DE Scaling factor = 0.5; crossov er probability = 0.5

3.2. Benchmark function validation

In this section, SMA was assessed on a comprehensive set of Note that the best results will be bolded (take one in the case of
functions from 23 benchmarks and CEC 2014. These functions juxtaposition).
cover unimodal, multimodal, hybrid, and composite functions, as
shown in Tables 1–3. Some composite functions of CEC 2014 are
3.2.1. Exploitation competence analysis
shown in Fig. 9. Dim denotes the dimension of function; Range
denotes the definition domain of the function, and fmin denotes The data in Table 5 demonstrates that SMA ranked first or tied
the optimal value of the function. The MAs used for comparison first on average when solving F1-5, F7, and F14. The convergence
include well-regarded and recent ones: WOA [53], GWO [24], curves of F2 and F5 in Fig. 10 can be visually observed that SMA
MFO [26], BA [23], SCA [9], FA [54], PSO [22], SSA [55], MVO [6], has the fastest convergence trend among all the comparative
ALO [56], PBIL [57], DE [58] and advanced MAs: AGA [59], BLPSO functions. The data in Table 6 demonstrates that SMA can still
[60], CLPSO [61], CBA [62], RCBA [63], CDLOBA [64], m_SCA [65], exhibit significant advantages even when compared to a modi-
IWOA [66], LWOA [67], and CSSA [68]. The parameter setup of fied MA, such as ranking first among other unimodal functions
traditional MAs is detailed in Table 4. The parameter selection other than F5 and F14. These functions are unimodal functions in
was based on the parameters used by the original author in the the benchmarks, reflecting SMA’s efficient exploration capability.
article or the parameters widely used by various researchers. Moreover, in order to more fairly evaluate the local search effi-
All algorithms were performed under the same conditions to
ciency of the algorithm, an evaluation version of the experiment
achieve fairness in comparative experiments. Among them, the
has been added. The data in Table 7 demonstrate the experimen-
population was set to 30, the dimension and the iteration time
was set to 30 and 1000 respectively. To reduce the impacts of tal results obtained by 300,000 evaluations of the SMA with 10
random factors in the algorithm on the results, all the compared other participants on the unimodal functions. In the experimental
algorithms were run individually 30 times in each function and results, the values obtained by SMA were still better than those
averaged as the final running result. On the purpose of measuring of other algorithms on F1-5 and F7. At the same time, the median
experiment results, Standard deviation (STD), Average results values of the solutions were also consistent with the ranking of
(AVG), and Median (MED) were employed to evaluate the results. the optimal values, indicating the stability of the SMA.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 307

Fig. 8. Qualitative analysis.

3.2.2. Exploration competence analysis be seen that the results of SMA in F9–F11, F17, and F20–21 are
The data in Table 8 represents that SMA is still competi- optimal, and only slightly lower than other algorithms in F8,
tive in multimodal functions. In F8–F11 and F20–21, the AVG of F18, and F19, which indicates that SMA can still maintain its
SMA was the smallest or the smallest in parallel compared with advantages over advanced algorithms and reflect SMA’s capability
other algorithms. From the convergence curves of F8 and F21 in to avoid local optimum solutions. Fig. 11 also shows that SMA can
Fig. 10, it can be observed that SMA can search for the highest find a superior solution at a relatively fast convergence tendency
accuracy fitness value in these two multimodal functions, while in multimodal functions such as F9–11, F17, and F21. Table 10
some algorithms fail to obtain a superior solution after a certain illustrates the experimental results of SMA with 10 other com-
amount of iterations. This is due to local optima stagnation, which parators on the multimodal function. Among them, SMA obtained
illustrates that SMA can still show better exploration ability in the best average and median results on F8–F11 compared with
case of preferable exploration. From the data in Table 9, it can other algorithms, and AGA obtained the best average and median
308 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Fig. 9. Illustration of CEC 2014 composite functions.

Table 5
Comparison results on unimodal functions with traditional algorithms during 1000 iterations.
Algorithm F1 F2 F3
AVG STD MED AVG STD MED AVG STD MED
SMA 0.000000 0.000000 1.08E−64 5.330E−207 0.000000 5.93E−58 0.00000 0.00000 8.22E−02
SCA 0.015244 0.029989 9.36E+01 1.150E−05 2.743E−05 8.06E−03 3261.99676 2935.03792 2.75E+04
SSA 1.231E−08 3.536E−09 1.83E+02 0.848146 0.941518 8.90E+00 236.62194 155.54710 2.94E+03
GWO 4.223E−59 1.081E−58 4.39E−46 1.128E−34 9.149E−35 7.07E−28 4.027E−15 1.418E−14 1.50E−09
MFO 2000.0006 4068.3807 2.04E+03 33.666839 20.253973 3.42E+01 24 900.5554 14 138.0477 2.91E+04
WOA 4.322E−153 2.276E−152 2.34E−54 5.032E−104 1.591E−103 3.42E−34 20 802.2782 10 554.3925 5.30E+04
GOA 7.670196 6.676643 1.27E+03 9.540510 14.128406 3.09E+01 1794.1195 1103.3922 7.64E+03
DA 1158.4940 600.8920 1.19E+03 14.313148 5.649106 1.45E+01 9612.3629 6188.5858 9.64E+03
ALO 1.050E−05 7.825E−06 7.10E+00 28.698940 42.100743 3.02E+01 1275.7431 596.2918 1.73E+03
MVO 0.318998 0.112060 9.40E+02 0.388930 0.137834 1.39E+01 48.11246 21.77526 4.61E+03
PBIL 46 908.0000 4218.6045 4.84E+04 95.200000 5.892134 9.80E+01 54 824.1 6552.855378 6.02E+04
PSO 128.803704 15.368375 1.42E+02 86.075426 65.298810 1.12E+02 406.96260 71.30926 6.06E+02
DE 3.030E−12 3.454E−12 4.01E−04 3.723E−08 1.196E−08 2.24E−03 24 230.5748 4174.3788 3.00E+04
Algorithm F4 F5 F6
AVG STD MED AVG STD MED AVG STD MED
SMA 2.301E−197 0.000000 1.31E−25 0.42779 0.63700 9.89E+00 0.000879 0.000415 5.97E−01
SCA 20.532489 11.046644 7.53E+01 532.7126 1907.4456 1.58E+06 4.550121 0.357049 3.37E+01
SSA 8.254602 3.287966 1.62E+01 135.5698 174.1213 7.77E+03 0.000000 0.000000 2.04E+02
GWO 1.776E−14 2.228E−14 9.01E−12 27.10029 0.86432 2.73E+01 0.726058 0.278337 9.75E−01
MFO 64.420279 8.689356 6.47E+01 5 348 258 20 289 785 5.35E+06 1656.708 5277.651 1.68E+03
WOA 45.706343 26.935040 4.61E+01 27.26543 0.57447 2.73E+01 0.100557 0.110525 1.01E−01
GOA 12.596514 4.317304 2.35E+01 1631.1583 2241.1368 2.58E+05 4.884661 4.512327 1.36E+03
DA 23.631736 8.191777 2.37E+01 127 371 96 386 1.31E+05 1330.292 632.470 1.34E+03
ALO 12.133214 3.585375 1.32E+01 298.8031 431.1446 5.00E+02 0.000012 0.000011 7.49E+00
MVO 1.076968 0.310884 1.40E+01 407.9465 615.3290 8.63E+04 0.323756 0.097394 9.34E+02
PBIL 79.666667 4.088110 8.00E+01 143 346 156 31 547 349 1.51E+08 45 881.833 4850.932 4.77E+04
PSO 4.498158 0.329339 4.79E+00 154 736 36 039 1.85E+05 132.779 15.189 1.45E+02
DE 1.965929 0.430531 1.32E+01 46.12942 27.29727 1.40E+02 3.096E−12 1.461E−12 4.11E−04
Algorithm F7 F14 F15
AVG STD MED AVG STD MED AVG STD MED
SMA 8.839E−05 7.118E−05 4.08E−04 9 549 563 6 529 870 2.97E+07 22 233.8245 14 144.9575 5.47E+07
SCA 0.024382 0.020732 6.04E−01 425 718 766 116 756 947 7.06E+08 2.689E+10 5.427E+09 3.97E+10
SSA 0.095541 0.050530 1.59E−01 20 297 116 8 153 518 6.91E+07 11 222.8121 11 173.7583 3.37E+08
GWO 0.000869 0.000435 1.46E−03 88 751 868 66 700 399 1.29E+08 2.254E+09 1.759E+09 3.98E+09
MFO 4.620163 13.076256 4.77E+00 87 010 749 137 363 574 1.00E+08 1.341E+10 7.685E+09 1.35E+10
WOA 0.000986 0.001147 2.66E−03 160 431 438 69 271 930 1.62E+08 2.154E+09 1.086E+09 2.17E+09
GOA 0.024028 0.011253 2.96E−02 33 807 500 14 819 986 1.28E+08 17 667 580 11 032 455 2.34E+09
DA 0.326978 0.138556 3.31E−01 305 164 519 121 919 102 3.05E+08 6.363E+09 2.751E+09 6.37E+09
ALO 0.103373 0.034257 1.06E−01 12 505 761 5 184 932 1.69E+07 12 378 9058 1.25E+07
MVO 0.020859 0.009584 1.42E−01 14 860 094 6 244 884 5.89E+07 566 570 210 025 1.45E+09
PBIL 282.1349 43.2693 2.93E+02 574 020 990 128 317 251 7.02E+08 4.961E+10 5.107E+09 5.32E+10
PSO 111.0068 21.5378 1.11E+02 17 174 833 5 483 990 2.16E+07 191 733 286 23 903 821 2.09E+08
DE 0.026937 0.006322 5.44E−02 100 597 441 31 636 302 1.78E+08 1601.8022 3314.1727 1.97E+05

on F16–21. Compared with AGA, SMA has a more significant from the optimum curve of F28–33 in Fig. 10 that SMA achieves
advantage in unimodal functions, while AGA has a preferable superior solutions faster than other counterparts, thus well coor-
performance in multimodal functions.
dinating the ability of exploration and exploitation. The statistics
3.2.3. Analysis of avoiding locally optimal solutions of Tables 13–14 illustrate that SMA can also maintain certain ad-
All functions in Tables 11–12, as fix-dimension multimodal vantages in composition functions compared with the advanced
functions, have multiple local optima, which are challenging for algorithm, which further reflects that SMA can avert falling into
MAs, thus can discriminate the overall efficacy of algorithms in
local optimum with fast convergence. F25, F32, and F33 in Fig. 11
exploration and exploration. According to the data in Tables 11–
12, SMA ranked first in AVG on F28, F29, F30, F32, and F33, which also intuitively incarnate the performance preponderance of SMA
show a very potential comprehensive ability. It can also be seen in composition functions.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 309

Table 6
Comparison results on the unimodal functions with advanced algorithms.
Algorithm F1 F2 F3
AVG STD MED AVG STD MED AVG STD MED
SMA 0.000000 0.000000 4.72E−37 4.20E−187 0.000000 1.24E−66 0.000000 0.000000 1.19E−02
BLPSO 2208.3313 397.7883 5.00E+03 17.665054 1.905407 3.35E+01 13 540.48 1672.45 1.82E+04
CLPSO 596.7364 150.3595 5.15E+03 11.846531 1.669288 4.09E+01 16 836.42 3085.75 2.71E+04
CBA 0.113583 0.454545 4.38E−01 305 804 1 652 847 5.73E+05 73.709725 31.029467 2.54E+02
RCBA 0.201488 0.052889 5.31E−01 10.958358 28.471304 2.77E+01 95.544912 43.376020 7.44E+02
CDLOBA 0.005957 0.002133 1.88E−02 3781.932 15 086.168 1.24E+04 1.791342 6.166318 3.50E+02
m_SCA 2.521E−46 1.378E−45 8.14E−04 3.478E−33 1.420E−32 2.01E−06 8.991E−16 3.188E−15 5.82E+03
IWOA 8.130E−146 4.370E−145 1.00E−53 2.385E−102 6.585E−102 1.44E−33 15 410.3 7420.1 3.62E+04
LWOA 6.743E−07 7.589E−07 1.55E−01 2.801E−07 3.833E−07 6.54E−02 43 293.10 13 505.91 9.25E+04
CSSA 0.017344 0.027805 1.74E−02 0.061732 0.027609 6.21E−02 2.926441 3.133898 2.95E+00
Algorithm F4 F5 F6
AVG STD MED AVG STD MED AVG STD MED
SMA 8.84E−183 0.00000 1.80E−36 1.27571 4.90297 1.22E+01 0.000880 0.000407 9.26E−01
BLPSO 27.66310 2.40967 3.54E+01 520 889 178 483 2.75E+06 2207.564 410.182 5.20E+03
CLPSO 42.44490 4.41014 5.61E+01 113 820 39 571 2.95E+06 563.251 138.054 5.26E+03
CBA 17.03820 7.72324 2.20E+01 197.6163 360.2440 2.58E+02 0.001823 0.007886 1.16E−01
RCBA 9.00594 3.41186 1.49E+01 148.2466 122.4613 2.29E+02 0.187352 0.054118 4.62E−01
CDLOBA 46.10460 7.48538 4.81E+01 138.1210 178.6248 2.29E+02 0.005940 0.001899 1.79E−02
m_SCA 2.248E−13 1.223E−12 1.53E+01 27.62609 0.84321 3.34E+01 2.540097 0.499546 4.06E+00
IWOA 13.12456 16.19609 2.26E+01 26.57003 0.66075 2.70E+01 0.036361 0.069578 6.17E−02
LWOA 11.12439 14.63066 2.69E+01 25.63874 6.59153 2.90E+01 0.009637 0.002992 4.25E−01
CSSA 0.03301 0.01983 3.45E−02 0.17508 0.16603 1.76E−01 0.030982 0.062573 3.11E−02
Algorithm F7 F14 F15
AVG STD MED AVG STD MED AVG STD MED
SMA 8.21E−05 7.16E−05 3.24E−04 9 689 581 7 904 687 3.20E+07 15 808.97 10 533.48 5.40E+07
BLPSO 0.59346 0.17290 1.50E+00 1.72E+08 3.74E+07 2.98E+08 3.718E+09 5.932E+08 8.78E+09
CLPSO 0.26201 0.05157 1.74E+00 1.77E+08 6.19E+07 4.28E+08 1.985E+09 4.391E+08 1.47E+10
CBA 0.47023 0.31242 7.47E−01 1.15E+07 5 802 441 1.80E+07 513 564.79 1 056 309.50 2.80E+06
RCBA 0.61360 0.25709 1.02E+00 5 943 596 2 275 351 1.06E+07 372 942.94 107 512.69 8.44E+05
CDLOBA 26.93780 39.54585 6.71E+01 4 469 831 2 849 244 1.07E+07 18 462.13 9920.05 3.57E+04
m_SCA 0.00071 0.00053 2.02E−02 1.15E+08 6.69E+07 3.52E+08 1.048E+10 4.703E+09 2.38E+10
IWOA 0.00185 0.00236 3.92E−03 9.34E+07 4.72E+07 1.19E+08 1.047E+09 8.576E+08 1.43E+09
LWOA 0.00650 0.00439 3.44E−02 8.81E+07 3.31E+07 4.11E+08 3.334E+08 1.326E+08 2.21E+10
CSSA 0.00019 0.00016 6.78E−04 1.68E+09 2.36E+08 1.68E+09 8.837E+10 6.958E+09 8.84E+10

3.2.4. Significance of superiority analysis by the algorithm are unequal. When inequality exists, then per-
Wilcoxon sign-rank test method [69] was exerted to verify form post-hoc analysis to know which algorithms have significant
whether SMA has obvious advantages over pairwise comparison. differences. Therefore, non-parametric Friedman’s test [71] was
If the p-value produced by the comparison is below the significant utilized. Table 16 illustrates the average ranking of the results
level of 0.05 in this case, it means that the achievements of the of the algorithms on the benchmarks based on three sets of
algorithm in pairwise comparison have obvious superiority in the experiments. In a non-hypothesis, there is equality between all
statistical sense. Otherwise, it is considered that the discrepancies algorithms, so if the hypothesis is reversed, it means that there
between the two contestants are inconspicuous in a statistical are differences between the algorithms being compared. Then
sense. In order to draw further comprehensive conclusions and we chose Holm’s test [72] as the method of post hoc analysis,
control the family-wise error rate (FWER), the true statistical sig- which is a multiple comparison method that can be used for
nificance (#TSS) of the combined pairwise comparison is shown control algorithms. Using the z-value obtained in Table 17 to
in Eq. (3.1) [70]: find the corresponding p-value from the normal distribution ta-
k−1
∏ ble and compare it with the corrected α value. Take SMA as a
p=1− 1 − pH1 (3.1) control algorithm and compared it with other algorithms. The
i=1 p-values have been sorted according to their significance. If the
The p value achieved from this expression is shown in Ta- p-value is lower than the corresponding significant level α , the
ble 15, where the TSS in F1-8, F10, F12, F15, F28–30, and F32–33 corresponding hypothesis is reversed, that is, the algorithm is
were all less than 0.05 when compared with traditional algo- significantly different. This paper selected two significant level
rithms. Therefore, SMA has significant differences in these func- α = 0.10 and α = 0.05, which indicate that there are marginal
tions compared to the traditional algorithms. TSS in Table 15, and significant differences between the two methods. As can be
when compared with advanced algorithms indicates that SMA seen from Table 17, compared with the traditional algorithms
outperforms other algorithms in F1-8, F10, F17, F19–22, F25, other than DE, the z-value is smaller than the corrected value
F28–30, F32–33. with α = 0.05 as the significant level, that is, there are significant
Although pairwise comparisons can be used for comparisons differences in benchmarks. Compared with advanced algorithms
between algorithms, the FWER generated during the experiment other than LWOA, there are significant differences among the
cannot be corrected in advance, and the choice of algorithms in benchmark functions, and slightly different from LWOA. In the
multiple comparisons can greatly affect the results of the analysis. experiments with other algorithms in the evaluation version, SMA
In order to reduce the effect of algorithm selection in each result is slightly different compared to GWO and WOA, not significantly
set, multiple comparison processes are used to modify FWER. In different from AGA and DE, while significantly different from the
multiple comparisons, first, check whether the results obtained remaining algorithms.
310 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Table 7
Comparison results on unimodal functions during 3E5 evaluations.
Algorithm F1 F2 F3
AVG STD MED AVG STD MED AVG STD MED
SMA 0.00000 0.00000 2.150E−268 0.00000 0.00000 1.999E−141 0.00000 0.00000 7.427E−244
SCA 5.33E−52 2.92E−51 1.325E−19 3.28E−60 9.54E−60 1.256E−28 2.65E+00 1.03E+01 2.763E+03
SSA 3.97E−09 7.20E−10 6.629E+01 2.20E−01 5.24E−01 4.818E+00 6.21E−08 1.97E−08 5.697E+02
GWO 0.00000 0.00000 0.000E+00 0.00000 0.00000 1.002E−286 8.62E−174 0.00000 1.908E−125
MFO 1.67E+03 3.79E+03 1.667E+03 3.53E+01 2.45E+01 3.533E+01 1.58E+04 1.08E+04 1.579E+04
WOA 0.00000 0.00000 0.000E+00 0.00000 0.00000 0.000E+00 2.15E+01 5.44E+01 1.755E+03
GOA 1.37E−03 7.51E−04 7.244E+02 4.93E−01 5.10E−01 1.954E+01 1.15E+02 3.94E+02 2.836E+03
MVO 3.11E−03 7.04E−04 5.957E+02 3.84E−02 1.30E−02 1.113E+01 3.70E−01 1.10E−01 1.613E+03
PSO 1.01E+02 1.43E+01 1.113E+02 4.69E+01 3.54E+00 5.156E+01 1.85E+02 2.76E+01 2.205E+02
DE 1.46E−159 3.86E−159 4.314E−76 2.02E−94 2.33E−94 1.359E−45 1.39E+03 7.73E+02 6.275E+03
AGA 2.38E−02 2.48E−02 5.567E−02 1.18E−02 3.99E−03 1.701E−02 4.51E−02 4.92E−02 8.333E−02
Algorithm F4 F5 F6
AVG STD MED AVG STD MED AVG STD MED
SMA 0.00000 0.00000 2.648E−131 2.22E−03 9.67E−04 1.837E−01 9.61E−06 4.23E−06 1.583E−02
SCA 4.46E−03 1.34E−02 1.490E+01 2.73E+01 6.99E−01 2.793E+01 3.70E+00 2.72E−01 4.367E+00
SSA 3.72E−01 7.06E−01 7.726E+00 7.27E+01 9.68E+01 2.160E+03 3.86E−09 9.08E−10 6.799E+01
GWO 1.79E−152 8.68E−152 2.593E−126 2.61E+01 9.13E−01 2.632E+01 4.64E−01 2.81E−01 6.100E−01
MFO 6.54E+01 1.03E+01 6.536E+01 2.69E+06 1.46E+07 2.686E+06 2.99E+03 7.91E+03 2.990E+03
WOA 3.68E+00 7.91E+00 4.832E+00 2.44E+01 3.14E−01 2.437E+01 5.89E−06 2.44E−06 5.896E−06
GOA 2.45E+00 2.03E+00 1.366E+01 1.52E+02 3.50E+02 6.639E+04 1.52E−03 7.49E−04 7.702E+02
MVO 8.89E−02 3.43E−02 9.891E+00 6.68E+01 9.45E+01 3.591E+04 3.05E−03 7.30E−04 6.130E+02
PSO 3.81E+00 2.16E−01 3.993E+00 8.98E+04 1.83E+04 1.085E+05 9.85E+01 8.65E+00 1.094E+02
DE 3.54E−15 5.37E−15 7.076E−07 3.08E+01 1.81E+01 3.259E+01 0.00000 0.00000 0.000E+00
AGA 3.17E−02 2.19E−02 6.531E−02 5.10E−02 6.04E−02 1.262E−01 1.58E−02 1.69E−02 1.145E−01
Algorithm F7 F14 F15
AVG STD MED AVG STD MED AVG STD MED
SMA 9.53E−06 8.25E−06 5.830E−05 2.15E+06 7.66E+05 9.335E+06 1.09E+04 1.28E+04 5.209E+06
SCA 2.43E−03 2.30E−03 1.570E−02 2.35E+08 5.63E+07 3.955E+08 1.65E+10 3.59E+09 2.586E+10
SSA 8.58E−03 4.21E−03 2.034E−02 1.72E+06 6.73E+05 2.440E+07 1.21E+04 9.72E+03 1.130E+08
GWO 6.07E−05 4.25E−05 9.191E−05 5.78E+07 3.28E+07 8.364E+07 2.18E+09 2.05E+09 3.621E+09
MFO 3.64E+00 5.34E+00 3.660E+00 9.51E+07 1.18E+08 9.580E+07 1.05E+10 7.21E+09 1.054E+10
WOA 1.38E−04 1.36E−04 3.663E−04 2.67E+07 1.08E+07 2.686E+07 4.45E+06 7.57E+06 4.481E+06
GOA 1.70E−03 9.63E−04 2.530E−03 1.31E+07 9.07E+06 4.304E+07 2.27E+07 1.24E+08 1.157E+09
MVO 2.99E−03 1.04E−03 6.692E−02 2.78E+06 1.07E+06 2.863E+07 1.55E+04 1.05E+04 9.453E+08
PSO 1.02E+02 2.89E+01 1.022E+02 8.12E+06 2.06E+06 1.019E+07 1.51E+08 1.61E+07 1.643E+08
DE 2.48E−03 6.04E−04 4.437E−03 2.05E+07 6.27E+06 3.310E+07 8.91E+02 1.81E+03 9.373E+02
AGA 1.77E−04 1.22E−04 3.056E−04 1.73E+02 8.34E+01 2.952E+02 2.40E+02 5.14E+01 2.971E+02

3.3. Wall-clock time analysis be recognized that the result of the algorithm is superior when z
was taken as 0.03, because the probability maintains the balance
In this section of the experimentations, SMA was compared between exploration and exploitation. Experimenter can also take
with the other 11 participants in the calculation of time-consuming different values for z according to specific problems.
experiments in the 33 benchmarks mentioned above. The time- To explore the influence of populations and iterations on
consuming calculation approach is that all participants indepen- the algorithm, we chose F13 to test the synergistic effect of
dently run 10 times on each function and recorded the results the two parameters on the algorithm. As can be seen visually
in Table 18. As can be observed from the data in the table, from Fig. 12, as the population size and iterations increased,
the computation of SMA takes relatively more extended time, the average became better. The reason is that the increase in
because the calculation of the oscillation factor requires more the number of populations improves search efficiency, and the
computing power. However, SMA can still outperform some al- increase in iterations leads to an incensement in the times of
gorithms while taking less time, such as GOA, DA, and ALO. In searches and the accuracy of subsequent searches. However, the
general, even if it is relatively time-consuming, SMA still possess results were not increased proportionally when the population
tremendous effectiveness advantages over other algorithms, so size and iterations continue to grow due to the global approxi-
the time results are expected. mate optimal solution has been roughly discovered. Researchers
can select the appropriate populations and iterations based on
3.4. Parameter sensitivity analysis specific questions.

In this section, parameter sensitivity test was utilized to eval- 3.5. Experiments on engineering design problems
uate the impacts of population size, iterations and parameter z
on the algorithm. The range of parameter z is [0,0.1], and there Methods based on different logics should be solved using
are 11 values at intervals of [0, 0.01]. The population size was set at least a proper numerical validation [73–78]. Most problems
to 5,10,30,50,100 and 200. The number of iterations was set to have constraints in the real production environment. The pro-
50,100,200,500,1000 and 2000. Under other conditions remained, cess of considering constraints of equality and inequality during
different values of parameter z were tested on F1–13 and the optimization is called constraints processing. The candidate so-
results are shown in Table 19. SMA0 indicates that z takes a value lutions of the heuristic algorithm can be divided into feasible
of 0, SMA1 indicates that z takes a value of 0.01, and so on. The and infeasible according to the constraints. There are currently
values in the table are ranking. From the results in Table 20, it can several types of constraint methods: death penalty, annealing,
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 311

Table 8
Results on multimodal functions with traditional algorithms during 1000 iterations.
Algorithm F8 F9 F10
AVG STD MED AVG STD MED AVG STD MED
SMA −12 569.4 0.1 −1.26E+04 0.00000 0.00000 9.96E−01 8.882E−16 0.00000 8.88E−16
SCA −3886.1 225.6 −3.82E+03 18.35521 21.43693 7.22E+01 11.32308 9.66101 1.42E+01
SSA −7816.8 842.3 −6.98E+03 56.61307 12.89967 1.38E+02 2.25688 0.72068 5.03E+00
GWO −6088.7 859.4 −3.83E+03 0.06990 0.38287 1.12E−01 0.00000 0.00000 1.62E−14
MFO −8711.6 827.4 −8.71E+03 162.06619 49.63022 1.63E+02 15.79421 6.91218 1.60E+01
WOA −11 630.6 1277.5 −1.15E+04 0.00000 0.00000 0.00000 3.967E−15 2.030E−15 4.09E−15
GOA −7430.4 761.2 −5.33E+03 86.74360 31.98704 2.35E+02 4.63913 1.06742 9.76E+00
DA −5631.8 590.7 −5.62E+03 155.13449 38.31121 1.56E+02 8.64831 1.22491 8.72E+00
ALO −5610.1 438.7 −5.61E+03 80.88997 20.29005 8.49E+01 2.00733 0.77081 2.90E+00
MVO −7744.9 693.4 −5.59E+03 112.71842 24.57189 2.33E+02 1.14572 0.70341 7.70E+00
PBIL −4046.4 331.0 −3.87E+03 150.36667 19.01267 1.55E+02 18.44223 0.19901 1.85E+01
PSO −6728.1 650.2 −6.72E+03 369.24464 18.68261 3.73E+02 8.41508 0.41051 8.75E+00
DE −12 409.8 149.2 −9.93E+03 59.28367 6.07679 8.60E+01 4.638E−07 1.383E−07 5.66E−03
Algorithm F11 F12 F13
AVG STD MED AVG STD MED AVG STD MED
SMA 0.00000 0.00000 0.00000 0.001195 0.001422 1.42E−02 0.001577 0.003000 1.45E−01
SCA 0.23534 0.22480 1.29E+00 2.290194 2.958865 3.48E+07 518.6869 2782.8453 1.78E+07
SSA 0.01009 0.01067 2.75E+00 5.542545 3.122247 2.17E+01 1.010473 4.701096 9.51E+01
GWO 0.00028 0.00156 3.30E−04 0.037303 0.019955 5.70E−02 0.488377 0.174343 6.85E−01
MFO 22.63478 42.31343 2.82E+01 0.470607 0.782326 3.78E+02 6792.354 37 201.162 8.22E+03
WOA 0.00000 0.00000 0.00000 0.005205 0.003512 5.21E−03 0.181197 0.166955 1.81E−01
GOA 0.83124 0.15983 1.29E+01 6.489011 2.717562 4.07E+03 26.3886 16.5919 1.36E+05
DA 9.87794 4.37600 1.00E+01 306.688 1096.994 3.10E+02 4.571E+04 1.022E+05 4.73E+04
ALO 0.00994 0.01271 1.07E+00 9.456697 3.198074 1.28E+01 2.193406 7.919110 3.25E+00
MVO 0.57543 0.08747 8.98E+00 1.294524 1.103471 1.27E+01 0.081286 0.043182 1.78E+03
PBIL 416.755 48.474 4.25E+02 2.667E+08 7.771E+07 2.99E+08 5.860E+08 9.982E+07 6.40E+08
PSO 1.03228 0.00489 1.04E+00 4.80322 0.86670 5.16E+00 23.191583 4.195613 2.88E+01
DE 9.761E−11 2.126E−10 7.56E−03 3.633E−13 3.399E−13 5.03E−05 1.691E−12 1.165E−12 2.44E−04
Algorithm F16 F17 F18
AVG STD MED AVG STD MED AVG STD MED
SMA 521.0056 0.109097 5.21E+02 618.2822 3.265441 6.23E+02 1300.6543 0.117872 1.30E+03
SCA 521.0427 0.053484 5.21E+02 636.9826 2.244227 6.40E+02 1303.9293 0.374149 1.30E+03
SSA 520.0584 0.107997 5.21E+02 622.8313 4.728569 6.28E+02 1300.5756 0.148959 1.30E+03
GWO 521.0410 0.054652 5.21E+02 616.6474 2.512406 6.24E+02 1300.6905 0.549189 1.30E+03
MFO 520.2870 0.170908 5.20E+02 622.7437 2.701796 6.23E+02 1301.3678 1.019364 1.30E+03
WOA 520.7787 0.119860 5.21E+02 637.7305 2.887311 6.38E+02 1300.5741 0.260727 1.30E+03
GOA 520.1390 0.082631 5.21E+02 622.1088 4.176909 6.30E+02 1300.5707 0.149671 1.30E+03
DA 520.9891 0.094995 5.21E+02 637.2321 2.789804 6.37E+02 1301.4935 1.087595 1.30E+03
ALO 520.0494 0.093898 5.21E+02 626.0851 3.620101 6.27E+02 1300.4614 0.100828 1.30E+03
MVO 520.5350 0.102963 5.21E+02 614.4619 3.437751 6.25E+02 1300.6110 0.114900 1.30E+03
PBIL 521.0393 0.043185 5.21E+02 640.6707 1.407127 6.41E+02 1305.2666 0.311548 1.31E+03
PSO 521.0618 0.054837 5.21E+02 624.8413 3.071015 6.26E+02 1300.5438 0.095901 1.30E+03
DE 520.7948 0.090515 5.21E+02 629.2747 1.350482 6.32E+02 1300.5363 0.050040 1.30E+03
Algorithm F19 F20 F21
AVG STD MED AVG STD MED AVG STD MED
SMA 1400.6670 0.361757 1.40E+03 1510.9564 3.012250 1.52E+03 1611.4845 0.567778 1.61E+03
SCA 1473.0029 15.520309 1.51E+03 16 869 13 476.33 1.26E+05 1613.2141 0.241155 1.61E+03
SSA 1400.4157 0.238649 1.40E+03 1513.1155 4.171347 1.53E+03 1612.2034 0.537832 1.61E+03
GWO 1407.2551 8.107508 1.42E+03 1949.1287 920.5966 2.05E+03 1611.7755 0.656408 1.61E+03
MFO 1430.1235 20.716796 1.43E+03 208 671 416 720.09 2.17E+05 1612.6679 0.536141 1.61E+03
WOA 1405.0142 6.261895 1.41E+03 1727.0908 122.1192 1.73E+03 1612.8485 0.463174 1.61E+03
GOA 1400.4834 0.331069 1.40E+03 1519.1245 6.359294 2.07E+03 1612.5397 0.510917 1.61E+03
DA 1422.6359 10.796483 1.42E+03 9188.8893 11 460.10 9.19E+03 1613.1921 0.298363 1.61E+03
ALO 1400.2530 0.047583 1.40E+03 1513.5362 4.828335 1.52E+03 1612.6442 0.572926 1.61E+03
MVO 1400.5551 0.403115 1.40E+03 1512.5460 3.700993 1.54E+03 1612.2971 0.526756 1.61E+03
PBIL 1525.2857 13.420862 1.54E+03 1 435 558 748 053.04 1.65E+06 1613.3661 0.212279 1.61E+03
PSO 1400.3217 0.095276 1.40E+03 1519.8378 1.631079 1.52E+03 1612.5422 0.412383 1.61E+03
DE 1400.4031 0.089745 1.40E+03 1517.1531 1.278695 1.52E+03 1612.5367 0.196986 1.61E+03

static, dynamic, co-evolutionary, and adaptive. Although useful 3.5.1. Welded beam structure problem
information may be lost in the process of abandonment, we still The main purpose of the problem is to constrain side con-
adopted a relatively simple method of the death penalty with low straints, end deflection of the beam (δ), buckling load on the bar
computational cost to deal with search individuals who violated (Pc ), bending stress in the beam (θ), moreover, shear stress (τ )
constraints and then re-assigned them a relatively large target with the least economic cost of welded beams.
value. There are four variables, for instance, the thickness of the weld
In the following sections, SMA has been tested on four (h), length of the attached bar (l), the height of the bar (t ), the
engineering-constrained design problems: a welded beam prob- thickness of the bar (b). The design diagram for this problem is
lem, a pressure vessel problem, a cantilever, and I-beam. shown in Fig. 13. The formulations were listed below:
312 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Table 9
Comparison results on the multimodal functions with advanced algorithms.
Algorithm F8 F9 F10
AVG STD MED AVG STD MED AVG STD MED
SMA −12 569.4 0.068790 −1.25E+04 0.00000 0.00000 0.00000 8.88E−16 0.00000 8.88E−16
BLPSO −4544.5 400.3510 −3.87E+03 207.3039 17.0015 2.30E+02 10.22852 0.69752 1.30E+01
CLPSO −8295.7 351.9193 −6.10E+03 139.7601 15.8072 2.17E+02 8.16910 0.64983 1.43E+01
CBA −7355.4 720.5161 −7.32E+03 133.1773 40.7382 1.44E+02 14.91852 3.56105 1.50E+01
RCBA −7248.6 814.7588 −7.24E+03 77.4955 14.5193 1.07E+02 6.76084 6.62622 9.76E+00
CDLOBA −7236.3 600.1951 −7.23E+03 243.8551 62.2823 2.72E+02 19.57830 0.77234 1.97E+01
m_SCA −5925.7 986.2730 −3.94E+03 0.00000 0.00000 1.11E+01 5.35800 9.03538 1.34E+01
IWOA −11 252.0 1780.6529 −1.12E+04 0.00000 0.00000 0.00000 3.73E−15 2.17E−15 3.73E−15
LWOA −10 775.8 1141.9779 −1.02E+04 5.12692 18.79066 2.12E+01 4.81E−05 2.84E−05 1.03E−01
CSSA −12 569.5 0.000239 −1.26E+04 7.14583 39.06861 7.15E+00 0.03173 0.03027 3.21E−02
Algorithm F11 F12 F13
AVG STD MED AVG STD MED AVG STD MED
SMA 0.00000 0.00000 0.00000 0.00095 0.00101 2.68E−02 0.00135 0.00211 1.16E−01
BLPSO 21.49704 3.65806 4.49E+01 4441.072 7073.234 3.24E+05 378 616.22 235 965.32 3.39E+06
CLPSO 6.33968 0.91129 4.95E+01 20.05685 8.11078 5.40E+05 11 963.83 13 926.90 4.89E+06
CBA 0.22145 0.11045 7.77E−01 15.33572 7.52799 1.59E+01 43.5008 21.1814 4.59E+01
RCBA 0.02800 0.00947 6.72E−02 13.56632 4.54840 1.47E+01 0.09299 0.03609 2.19E−01
CDLOBA 145.5030 96.9037 1.74E+02 20.17146 6.03281 2.08E+01 35.8588 11.9314 3.85E+01
m_SCA 0.00000 0.00000 5.52E−02 0.19369 0.16449 9.82E−01 1.58065 0.19641 2.40E+00
IWOA 0.00264 0.01100 3.70E−03 0.00930 0.02578 1.18E−02 0.16079 0.13761 2.07E−01
LWOA 0.02455 0.04926 4.54E−01 0.00063 0.00024 1.78E−02 0.01660 0.01442 2.05E−01
CSSA 0.02723 0.03762 2.74E−02 5.98E−05 5.33E−05 6.03E−05 0.00090 0.00086 9.06E−04
Algorithm F16 F17 F18
AVG STD MED AVG STD MED AVG STD MED
SMA 521.0127 0.069163 5.21E+02 619.4282 2.915833 6.24E+02 1300.6589 0.145401 1.30E+03
BLPSO 521.0920 0.070988 5.21E+02 629.3125 1.805214 6.34E+02 1300.9286 0.138697 1.30E+03
CLPSO 521.0176 0.059879 5.21E+02 629.7237 1.356299 6.35E+02 1300.6655 0.089057 1.30E+03
CBA 520.3188 0.287026 5.20E+02 641.6516 3.410418 6.42E+02 1300.5091 0.134277 1.30E+03
RCBA 520.3774 0.123562 5.21E+02 640.2023 3.196174 6.41E+02 1300.4976 0.123416 1.30E+03
CDLOBA 521.0056 0.064721 5.21E+02 636.2815 2.936580 6.37E+02 1300.5098 0.146951 1.30E+03
m_SCA 520.9230 0.085023 5.21E+02 625.2555 2.906023 6.37E+02 1301.7144 0.980372 1.30E+03
IWOA 520.7061 0.096424 5.21E+02 634.7725 3.121824 6.36E+02 1300.5275 0.096831 1.30E+03
LWOA 520.7827 0.071113 5.21E+02 633.6692 3.853306 6.40E+02 1300.6093 0.123410 1.30E+03
CSSA 521.0604 0.088972 5.21E+02 644.9713 1.825103 6.45E+02 1309.5241 0.830936 1.31E+03
Algorithm F19 F20 F21
AVG STD MED AVG STD MED AVG STD MED
SMA 1400.6565 0.361610 1.40E+03 1510.5477 2.46585 1.52E+03 1611.5995 0.70239 1.61E+03
BLPSO 1410.4409 2.902210 1.43E+03 1802.5795 180.2212 4.48E+03 1613.0067 0.23416 1.61E+03
CLPSO 1403.5324 2.812311 1.45E+03 1952.4155 304.9825 4.26E+04 1613.0049 0.22798 1.61E+03
CBA 1400.3048 0.092093 1.40E+03 1562.3666 18.85652 1.56E+03 1613.5381 0.36317 1.61E+03
RCBA 1400.2943 0.060668 1.40E+03 1538.9490 7.61211 1.54E+03 1613.6523 0.32500 1.61E+03
CDLOBA 1400.3181 0.058475 1.40E+03 1753.9951 117.6904 1.76E+03 1613.5741 0.25668 1.61E+03
m_SCA 1426.1725 10.27231 1.46E+03 4997.7533 4929.0634 1.55E+04 1612.5383 0.51908 1.61E+03
IWOA 1400.2787 0.143274 1.40E+03 1625.8982 78.1816 1.67E+03 1612.9124 0.55626 1.61E+03
LWOA 1400.3289 0.095342 1.47E+03 1572.8452 27.80344 1.26E+04 1612.8272 0.52137 1.61E+03
CSSA 1680.8338 17.75465 1.68E+03 232 677.12 39 953.5 2.33E+05 1613.1690 0.24750 1.61E+03

Consider: where

X = [x1 , x2 , x3 , x4 ] = [h l t b] √
x2
τ −
( →)
x = (τ ′ )2 + 2τ ′ τ ′′ + (τ ′′ )2
Minimize: 2R
P MR x2
F (X ) = 1.10471x21 x2 + 0.04811x3 x4 (14.0 + x2 ) τ′ = √ , τ ′′ = , M = P(L + )
2x1 x2 J 2
√
Subject to: 2
x22
( )
x1 + x3
R= +
g1 (X ) = τ (X ) − τmax ≤ 0 4 2
g2 (X ) = σ (X ) − σmax ≤ 0
{ [ )2 ]}
√ x22
(
x1 + x3
J =2 2x1 x2 +
g3 (X ) = δ (X ) − δmax ≤ 0 4 2
g4 (X ) = x1 − x4 ≤ 0 6PL 6PL3
σ −
( →) (−→
, δ
)
x = x =
g5 (X ) = P − PC (X ) ≤ 0 x4 x23 Ex23 x4
g6 (X ) = 0.125 − x1 ≤ 0
√
x23 x64
→) 4.013E 36 (1 − x3 E )
√
(−
g7 (X ) = 1.10471x21 + 0.04811x3 x4 (14.0 + x2 ) − 5.0 ≤ 0 PC x = 2 L 2L 4G
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 313

Table 10
Comparison results on multimodal functions during 3E5 evaluations.
Algorithm F8 F9 F10
AVG STD MED AVG STD MED AVG STD MED
SMA −1.26E+04 2.48E−04 −1.257E+04 0.00000 0.00000 0.000E+00 8.88E−16 0.00000 8.882E−16
SCA −4.41E+03 2.15E+02 −4.288E+03 0.00000 0.00000 3.499E+00 1.26E+01 9.43E+00 1.610E+01
SSA −7.79E+03 7.06E+02 −7.419E+03 6.54E+01 1.50E+01 9.676E+01 1.81E+00 8.07E−01 3.901E+00
GWO −6.38E+03 7.23E+02 −4.403E+03 0.00000 0.00000 0.000E+00 7.64E−15 1.08E−15 7.638E−15
MFO −8.37E+03 7.59E+02 −8.366E+03 1.65E+02 3.28E+01 1.651E+02 1.58E+01 7.02E+00 1.576E+01
WOA −1.21E+04 9.04E+02 −1.207E+04 0.00000 0.00000 0.000E+00 3.38E−15 2.12E−15 3.375E−15
GOA −7.56E+03 6.06E+02 −6.158E+03 1.04E+02 4.22E+01 1.742E+02 2.71E+00 8.89E−01 7.415E+00
MVO −8.18E+03 7.17E+02 −6.424E+03 8.27E+01 2.44E+01 1.772E+02 1.08E−01 3.58E−01 6.771E+00
PSO −7.07E+03 8.27E+02 −7.067E+03 3.43E+02 1.69E+01 3.469E+02 7.78E+00 2.41E−01 8.041E+00
DE −1.24E+04 1.31E+02 −1.243E+04 3.32E−02 1.82E−01 3.317E−02 7.64E−15 1.08E−15 7.994E−15
AGA −8.38E+02 9.72E−03 −8.379E+02 9.94E−03 0.00000 1.655E−02 1.64E−02 0.00000 1.644E−02
Algorithm F11 F12 F13
AVG STD MED AVG STD MED AVG STD MED
SMA 0.00000 0.00000 0.000E+00 7.55E−06 8.36E−06 2.780E−04 6.77E−06 3.68E−06 2.418E−03
SCA 8.03E−11 4.36E−10 6.453E−02 3.27E−01 5.08E−02 6.351E+03 1.98E+00 1.11E−01 2.375E+00
SSA 1.18E−02 1.10E−02 1.577E+00 1.41E+00 1.70E+00 6.100E+00 5.06E−03 6.75E−03 3.688E+00
GWO 2.49E−04 1.36E−03 2.514E−04 2.56E−02 1.20E−02 3.778E−02 4.01E−01 1.95E−01 5.442E−01
MFO 3.31E+01 5.55E+01 3.312E+01 2.29E−01 4.75E−01 2.288E−01 6.15E−01 1.11E+00 6.152E−01
WOA 6.58E−04 2.52E−03 6.577E−04 1.09E−06 4.07E−07 1.087E−06 3.84E−04 2.00E−03 3.836E−04
GOA 1.81E−02 1.51E−02 7.615E+00 1.93E+00 1.50E+00 1.380E+01 9.33E−01 3.86E+00 5.700E+03
MVO 2.76E−02 1.33E−02 6.603E+00 1.64E−01 5.09E−01 7.007E+00 4.06E−03 5.30E−03 3.389E+01
PSO 1.02E+00 1.27E−02 1.022E+00 3.38E+00 3.70E−01 3.822E+00 1.57E+01 1.83E+00 1.729E+01
DE 0.00000 0.00000 0.000E+00 1.57E−32 5.57E−48 1.571E−32 1.35E−32 5.57E−48 1.350E−32
AGA 2.14E−02 1.37E−02 3.063E−02 2.17E−02 2.82E−02 5.744E−02 1.13E−02 9.89E−03 1.987E−02
Algorithm F16 F17 F18
AVG STD MED AVG STD MED AVG STD MED
SMA 5.21E+02 2.27E−01 5.210E+02 6.15E+02 3.06E+00 6.188E+02 1.30E+03 1.26E−01 1.301E+03
SCA 5.21E+02 5.60E−02 5.210E+02 6.33E+02 2.39E+00 6.364E+02 1.30E+03 3.71E−01 1.304E+03
SSA 5.20E+02 1.07E−01 5.210E+02 6.19E+02 4.24E+00 6.234E+02 1.30E+03 1.45E−01 1.301E+03
GWO 5.21E+02 5.11E−02 5.210E+02 6.14E+02 3.27E+00 6.210E+02 1.30E+03 3.11E−01 1.301E+03
MFO 5.20E+02 1.73E−01 5.203E+02 6.23E+02 3.53E+00 6.231E+02 1.30E+03 1.26E+00 1.302E+03
WOA 5.20E+02 1.61E−01 5.204E+02 6.36E+02 4.15E+00 6.363E+02 1.30E+03 1.05E−01 1.301E+03
GOA 5.20E+02 7.96E−02 5.210E+02 6.17E+02 3.63E+00 6.250E+02 1.30E+03 6.95E−02 1.301E+03
MVO 5.20E+02 4.14E−02 5.210E+02 6.10E+02 3.97E+00 6.214E+02 1.30E+03 1.24E−01 1.301E+03
PSO 5.21E+02 4.59E−02 5.210E+02 6.23E+02 3.42E+00 6.231E+02 1.30E+03 7.31E−02 1.300E+03
DE 5.21E+02 4.46E−02 5.206E+02 6.20E+02 2.07E+00 6.226E+02 1.30E+03 4.04E−02 1.300E+03
AGA 5.00E+02 4.82E−01 5.005E+02 6.00E+02 1.68E−02 6.000E+02 1.30E+03 2.53E−02 1.300E+03
Algorithm F19 F20 F21
AVG STD MED AVG STD MED AVG STD MED
SMA 1.40E+03 3.13E−01 1.401E+03 1.51E+03 1.83E+00 1.517E+03 1.61E+03 7.14E−01 1.612E+03
SCA 1.44E+03 7.88E+00 1.466E+03 4.96E+03 4.20E+03 2.681E+04 1.61E+03 2.17E−01 1.613E+03
SSA 1.40E+03 2.22E−01 1.400E+03 1.51E+03 2.10E+00 1.523E+03 1.61E+03 6.27E−01 1.612E+03
GWO 1.40E+03 7.60E+00 1.410E+03 1.89E+03 7.48E+02 1.960E+03 1.61E+03 6.66E−01 1.612E+03
MFO 1.43E+03 2.55E+01 1.435E+03 3.14E+05 5.20E+05 3.141E+05 1.61E+03 5.10E−01 1.613E+03
WOA 1.40E+03 1.23E−01 1.400E+03 1.57E+03 2.49E+01 1.575E+03 1.61E+03 5.51E−01 1.613E+03
GOA 1.40E+03 3.31E−01 1.401E+03 1.51E+03 2.20E+00 1.531E+03 1.61E+03 7.46E−01 1.612E+03
MVO 1.40E+03 3.32E−01 1.401E+03 1.51E+03 2.26E+00 1.527E+03 1.61E+03 5.89E−01 1.612E+03
PSO 1.40E+03 9.78E−02 1.400E+03 1.52E+03 1.38E+00 1.517E+03 1.61E+03 4.48E−01 1.612E+03
DE 1.40E+03 1.24E−01 1.400E+03 1.51E+03 1.10E+00 1.513E+03 1.61E+03 2.18E−01 1.612E+03
AGA 1.40E+03 1.21E−02 1.400E+03 1.50E+03 7.70E−03 1.500E+03 1.60E+03 9.19E−03 1.600E+03

P = 60001 b, L = 14 in. δmax = 0.25 in. E = 30 × 16 psi, thickness of the head (Th ) and the length of the cylindrical por-
G = 12 × 10 psi6 tion. Both ends of the container are covered with a hemispherical
shell at one end. Fig. 14 illustrates the design of the object and its
τmax = 13600 psi, σmax = 30000 psi corresponding parameters.
In this problem, SMA was compared with MFO [26], SSA [55], The formulations of four constraints are listed as follow:
Random [79], Siddall [80], Ragsdell [79], Coello and Montes [81], Consider:
GWO [24], WOA [53], GSA, Simplex [79] and David [79]. Table 20 X = [x1 x2 x3 x4 ] = [Ts Th R L]
illustrates that SMA can obtain the optimal value among the
competitors. Objective:
f (X )min = 0.6224x1 x3 x4 + 1.7781x3 x21 + 3.1661x4 x21 + 19.84x3 x21
3.5.2. Pressure vessel structure problem
The intention of the problem is to find the parameters of Subject to:
cylindrical pressure vessels which can minimize the total cost of
production and meet the pressure requirements. The parameters g1 (X ) = −x1 + 0.0193x3 ≤ 0
including the thickness of the shell (Ts ), inner radius (Th ), the g2 (X ) = −x3 + 0.00954x3 ≤ 0
314 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Table 11
Comparison results on the Hybrid functions of CEC 2014 with traditional algorithms.
Algorithm F22 F23 F24
AVG STD MED AVG STD MED AVG STD MED
SMA 1 981 009 955 714 3.35E+06 23 768.042 9648.796 2.41E+05 1916.4612 20.81879 1.92E+03
SCA 1.475E+07 7 203 070 2.36E+07 2.767E+08 1.768E+08 7.00E+08 2025.9911 29.94193 2.08E+03
SSA 1 105 845 643 830 3.04E+06 10 164.216 8416.726 5.28E+06 1920.6469 18.54917 1.93E+03
GWO 3 134 418 3 888 996 4.44E+06 1.721E+07 2.683E+07 4.25E+07 1959.2308 39.30239 1.97E+03
MFO 3 685 312 5 224 753 4.56E+06 3.014E+07 1.146E+08 3.10E+07 1971.9869 47.63181 1.97E+03
WOA 1.704E+07 1.559E+07 1.73E+07 435 824 298 836 4.45E+05 1996.2181 42.52503 2.00E+03
GOA 1 438 154 1 067 917 4.32E+06 6928.741 5474.188 5.04E+07 1916.6548 2.61108 1.94E+03
DA 1.179E+07 8 700 286 1.18E+07 1.213E+07 1.867E+07 1.23E+07 1998.5264 63.45143 2.00E+03
ALO 1 218 376 902 036 1.58E+06 3771.512 1977.571 1.54E+05 1922.1569 20.15628 1.92E+03
MVO 648 113 423 330 2.31E+06 11 057.398 8877.483 3.04E+07 1913.1472 2.24974 1.93E+03
PBIL 1.987E+07 5 897 960 2.61E+07 1.090E+09 4.562E+08 1.33E+09 2153.5319 42.97329 2.18E+03
PSO 721 353 340 828 1.11E+06 3 733 762 1 011 652 4.15E+06 1917.9437 2.40252 1.92E+03
DE 5 502 647 2 468 774 8.31E+06 199 823 164 718 6.71E+05 1911.6762 2.02980 1.92E+03
Algorithm F25 F26 F27
AVG STD MED AVG STD MED AVG STD MED
SMA 26 151.83 14 592.587 3.50E+04 691 037.74 447 817.80 1.12E+06 2785.9438 200.10440 2.82E+03
SCA 42 036.30 23 681.495 1.14E+05 3 774 544 2 456 159 6.71E+06 3295.8542 153.63191 3.52E+03
SSA 28 121.80 15 931.602 3.37E+04 378 039 364 201 8.50E+05 2733.8257 195.24839 2.83E+03
GWO 26 371.89 17 760.819 3.15E+04 1 401 869 3 144 566 1.93E+06 2681.9022 164.02295 2.83E+03
MFO 68 807.98 34 415.152 7.04E+04 909 632.7 874 394.6 1.07E+06 2988.3818 304.13153 2.99E+03
WOA 141 181.82 150 095.222 1.42E+05 9 190 469 11 782 812 9.53E+06 3196.0473 259.04790 3.20E+03
GOA 15 217.92 11 721.455 2.85E+04 386 067 274 805 1.22E+06 2721.2340 196.75397 2.99E+03
DA 170 066.67 177 342.58 1.70E+05 4 046 744 4 943 178 4.05E+06 3238.7712 336.86424 3.24E+03
ALO 44 576.07 18 719.138 4.76E+04 416 024 304 543 5.44E+05 3023.7395 188.96878 3.03E+03
MVO 7113.7952 3345.696 2.22E+04 233 142 208 201 6.68E+05 2636.7226 181.32671 2.82E+03
PBIL 100 886.52 83 601.365 1.64E+05 6 032 360 2 197 809 8.28E+06 3545.4177 245.28854 3.70E+03
PSO 19 948.16 7799.1853 3.10E+04 324 137 191 521 4.32E+05 2934.5667 216.35151 2.97E+03
DE 12 477.50 4695.8815 1.57E+04 880 050 356 197 1.61E+06 2594.1841 115.53194 2.77E+03

Table 12
Comparison results on composite functions of CEC2014 with traditional algorithms.
Algorithm F28 F29 F30
AVG STD MED AVG STD MED AVG STD MED
SMA 2500.0000 0.00000 2.50E+03 2600.0000 0.00000 2.60E+03 2700.0000 0.00000 2.70E+03
SCA 2712.2094 24.49225 2.80E+03 2612.5376 16.11479 2.66E+03 2734.7826 10.58302 2.76E+03
SSA 2632.2624 11.99882 2.66E+03 2644.8211 7.39515 2.65E+03 2717.7099 4.57908 2.72E+03
GWO 2644.1677 15.52070 2.65E+03 2600.0269 0.00904 2.60E+03 2709.8706 6.60434 2.71E+03
MFO 2672.0010 55.23865 2.67E+03 2678.9946 29.86466 2.68E+03 2718.0092 9.84490 2.72E+03
WOA 2680.3948 54.85251 2.68E+03 2611.4760 7.24474 2.61E+03 2717.4700 20.81026 2.72E+03
GOA 2636.9870 8.96575 2.69E+03 2645.5150 5.40226 2.67E+03 2717.0876 4.91578 2.73E+03
DA 2721.4771 44.95443 2.72E+03 2661.0711 14.44841 2.66E+03 2743.0074 14.56249 2.74E+03
ALO 2629.0815 8.44167 2.63E+03 2651.7113 7.91597 2.65E+03 2726.5079 7.20090 2.73E+03
MVO 2624.2212 6.19267 2.65E+03 2635.7205 6.61514 2.66E+03 2708.2693 1.65495 2.72E+03
PBIL 3031.2435 81.13515 3.08E+03 2827.7418 25.17373 2.83E+03 2760.6456 11.08699 2.77E+03
PSO 2619.6398 1.46590 2.62E+03 2631.7808 6.40174 2.63E+03 2718.6208 5.72742 2.72E+03
DE 2615.2456 0.00132 2.62E+03 2628.8256 2.71282 2.63E+03 2722.3177 3.27296 2.73E+03
Algorithm F31 F32 F33
AVG STD MED AVG STD MED AVG STD MED
SMA 2700.7493 0.11128 2.70E+03 2900.0000 0.00000 2.90E+03 3000.0000 0.00000 3.00E+03
SCA 2703.5386 0.42894 2.70E+03 3824.0615 291.2901 3.95E+03 5546.0181 481.5436 5.59E+03
SSA 2700.5370 0.14086 2.70E+03 3530.3707 215.6535 3.62E+03 4170.5366 395.2774 4.47E+03
GWO 2766.6830 68.52197 2.77E+03 3401.0954 122.9196 3.59E+03 4250.4859 411.0486 4.75E+03
MFO 2702.0134 1.45934 2.70E+03 3622.4093 196.0596 3.62E+03 3955.0668 198.3093 3.96E+03
WOA 2717.0896 37.72566 2.72E+03 3902.6457 357.3971 3.90E+03 5395.1711 768.3897 5.41E+03
GOA 2772.4391 72.70589 2.78E+03 3534.0606 204.6178 3.68E+03 4454.6468 563.7650 4.87E+03
DA 2744.5288 66.71856 2.74E+03 3906.7882 348.4316 3.91E+03 6418.4644 759.2642 6.42E+03
ALO 2720.5156 40.82834 2.72E+03 3570.0454 291.1912 3.59E+03 5699.9910 503.7231 5.77E+03
MVO 2743.7298 71.22152 2.75E+03 3390.4461 148.2171 3.60E+03 4137.0810 330.4390 4.68E+03
PBIL 2704.7340 0.37452 2.71E+03 3931.9655 192.1610 3.97E+03 4535.2650 415.0320 4.62E+03
PSO 2790.8757 30.60922 2.79E+03 3487.2364 302.7920 3.51E+03 7526.0776 944.8788 7.80E+03
DE 2700.5562 0.06268 2.70E+03 3439.1447 117.3664 3.56E+03 3727.8590 34.29086 3.78E+03

4 From the data of Table 21, it is obvious that SMA can obtain
g3 (X ) = −π x4 x23 − π x33 + 1296000 ≤ 0
3
g4 (X ) = x4 − 240 ≤ 0 fairly superior optimal values compared with MFO [26], BA [82],

Variable ranges: HPSO [83], CSS [7], CPSO [84], ACO [85], GWO [24], WOA [53],

0 ≤ x1 ≤ 99, 0 ≤ x2 ≤ 99, 10 ≤ x3 ≤ 200, 10 ≤ x4 ≤ 200 MDDE [86], Lagrangian multiplier [87] and Branch-bound [88].
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 315

Table 13
Comparison results on the Hybrid functions of CEC 2014 with advanced algorithms.
Algorithm F22 F23 F24
AVG STD MED AVG STD MED AVG STD MED
SMA 1 804 495 975 279 2.92E+06 19 731.57 12 366.188 2.85E+05 1924.4016 29.9388 1.93E+03
BLPSO 5.95E+06 2.79E+06 1.01E+07 2.03E+07 7.71E+06 9.85E+07 1940.2723 7.4647 1.98E+03
CLPSO 9.36E+06 4.32E+06 2.15E+07 2.35E+07 1.66E+07 2.67E+08 1973.9754 18.5819 2.07E+03
CBA 875 955 682 022 1.41E+06 19 590.62 55 097.347 4.37E+04 1930.5601 26.3854 1.93E+03
RCBA 536 209 287 301 9.14E+05 27 035.80 48 215.716 4.05E+04 1929.3577 27.4688 1.93E+03
CDLOBA 253 257 159 357 5.29E+05 14 436.39 8148.226 2.05E+04 1983.8732 37.8347 1.99E+03
m_SCA 3 405 875 2.39E+06 1.58E+07 5.42E+07 6.44E+07 3.95E+08 1974.7719 29.9192 2.03E+03
IWOA 1.11E+07 6.76E+06 1.42E+07 232 413.82 5.40E+05 5.39E+05 1974.5125 48.0377 1.98E+03
LWOA 1.13E+07 7 452 363 2.92E+07 588 224.18 2.66E+06 7.46E+07 1954.6903 42.1280 2.12E+03
CSSA 2.34E+08 7.53E+07 2.34E+08 7.83E+09 2.67E+09 7.83E+09 2599.1453 124.8102 2.60E+03
Algorithm F25 F26 F27
AVG STD MED AVG STD MED AVG STD MED
SMA 20 184.423 11 536.23 2.74E+04 566 429.1 440 779.47 8.85E+05 2853.3837 275.30887 2.88E+03
BLPSO 40 020.222 17 226.38 6.37E+04 1 443 418 637 791.49 3.44E+06 3039.7382 166.89957 3.27E+03
CLPSO 35 837.420 13 506.64 8.72E+04 1 678 372 755 401.12 5.09E+06 2852.1935 155.80947 3.31E+03
CBA 47 394.400 21 444.35 6.74E+04 425 055.3 394 927.78 6.04E+05 3470.7820 389.88490 3.47E+03
RCBA 31 728.076 17 810.98 5.27E+04 353 013.1 266 894.62 5.93E+05 3385.7049 345.09736 3.40E+03
CDLOBA 49 593.385 25 881.81 6.28E+04 158 313.0 1.68E+05 2.95E+05 3280.2697 252.07423 3.28E+03
m_SCA 25 033.409 11 116.37 6.57E+04 903 524.0 975 927 3.32E+06 2710.7381 185.83183 3.26E+03
IWOA 54 586.150 24 987.96 6.40E+04 3 798 947 3 041 196 5.53E+06 3025.7726 243.33062 3.05E+03
LWOA 45 846.654 23 789.30 2.51E+05 2 379 875 1 411 692 1.57E+07 3021.9632 251.61117 3.37E+03
CSSA 4.00E+06 4.34E+06 4.00E+06 1.47E+08 1.33E+08 1.47E+08 55 417.05 53 372.36 5.54E+04

Table 14
Comparison results on composite functions of CEC2014 with advanced algorithms.
Algorithm F28 F29 F30
AVG STD MED AVG STD MED AVG STD MED
SMA 2500.0000 0.00000 2.50E+03 2600.0000 0.00000 2.60E+03 2700.0000 0.00000 2.70E+03
BLPSO 2642.6104 5.72836 2.68E+03 2667.7295 2.33812 2.68E+03 2729.1590 3.76596 2.74E+03
CLPSO 2641.0387 6.17469 2.73E+03 2660.4964 3.06098 2.69E+03 2723.7881 4.81229 2.74E+03
CBA 2618.5576 2.10126 2.62E+03 2672.7287 32.2058 2.67E+03 2738.8447 16.6010 2.74E+03
RCBA 2616.5451 1.59102 2.62E+03 2671.7927 31.5474 2.67E+03 2734.0323 12.9174 2.73E+03
CDLOBA 2619.8273 5.91180 2.63E+03 2701.7423 41.3677 2.70E+03 2724.5749 12.1083 2.73E+03
m_SCA 2657.6041 18.7651 2.70E+03 2600.0082 0.00530 2.60E+03 2717.9183 3.76490 2.74E+03
IWOA 2653.0880 16.8268 2.66E+03 2607.2619 5.28645 2.61E+03 2714.2469 16.8627 2.72E+03
LWOA 2635.7620 26.8390 2.81E+03 2610.4768 6.01548 2.61E+03 2712.9723 16.0958 2.72E+03
CSSA 2525.7607 121.6063 2.53E+03 2600.4211 0.21279 2.60E+03 2700.0573 0.03311 2.70E+03
Algorithm F31 F32 F33
AVG STD MED AVG STD MED AVG STD MED
SMA 2700.7690 0.14584 2.70E+03 2900.0000 0.000000 2.90E+03 3000.0000 0.000000 3.03E+03
BLPSO 2719.8189 40.7556 2.73E+03 3639.7610 104.8236 3.78E+03 4689.9702 351.2075 5.46E+03
CLPSO 2703.1872 0.86410 2.71E+03 3255.3722 59.2992 3.45E+03 5394.7830 503.6760 6.86E+03
CBA 2714.7665 60.7392 2.72E+03 3992.4341 466.0520 4.00E+03 5749.7099 1087.8530 5.77E+03
RCBA 2731.0975 67.5372 2.75E+03 4088.5512 406.7701 4.11E+03 5820.2491 1076.0463 5.86E+03
CDLOBA 2715.7761 56.3032 2.72E+03 3902.9353 371.3979 3.91E+03 5346.9885 837.0278 5.36E+03
m_SCA 2701.3185 0.81232 2.70E+03 3324.8813 254.9219 3.52E+03 4280.9832 270.1009 4.87E+03
IWOA 2732.9019 77.0976 2.73E+03 3800.5003 342.5802 3.83E+03 5181.1074 676.4584 5.28E+03
LWOA 2700.5873 0.13596 2.70E+03 3865.9583 237.8645 4.00E+03 4457.1988 270.3753 4.91E+03
CSSA 2792.4644 23.3249 2.79E+03 4836.8934 344.1577 4.84E+03 8555.0615 3476.6646 8.56E+03

Table 15
True p-value obtained from comparison on thirty-three benchmarks.
F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11
#TSS 2.08E−05 2.08E−05 2.08E−05 2.08E−05 2.08E−05 4.13E−03 2.68E−05 1.34E−04 1.00E+00 2.31E−05 1.00E+00
Traditional F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22
MAs #TSS 3.71E−05 5.23E−02 5.53E−02 6.34E−03 7.95E−01 1.32E−01 3.26E−01 2.24E−01 1.98E−01 6.88E−02 7.01E−01
F23 F24 F25 F26 F27 F28 F29 F30 F31 F32 F33
#TSS 4.53E−01 7.59E−01 9.96E−01 5.34E−01 4.96E−01 2.29E−05 2.08E−05 2.63E−04 4.36E−01 2.08E−05 2.16E−05
F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11
#TSS 1.56E−05 1.56E−05 1.56E−05 1.56E−05 1.85E−02 1.40E−03 2.60E−03 1.60E−05 1.00E+00 2.74E−05 1.00E+00
Advanced F12 F13 F14 F15 F16 F17 F18 F19 F20 F21 F22
MAs #TSS 1.83E−01 4.91E−01 1.25E−01 1.91E−01 9.93E−01 1.86E−05 7.57E−01 5.59E−04 1.56E−05 3.22E−05 2.04E−03
F23 F24 F25 F26 F27 F28 F29 F30 F31 F32 F33
#TSS 3.37E−01 5.58E−02 3.91E−02 3.62E−01 8.15E−01 1.56E−05 1.56E−05 7.49E−05 7.75E−01 1.56E−05 1.64E−05
316 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Fig. 10. Comparisons between SMA and traditional MAs.

3.5.3. Cantilever structure problem Consider:

The cantilever beam is made up of five hollow square cross-
sections, as exhibited in Fig. 15. Since the thickness is fixed, only X = [x1 x2 x3 x4 x5 ]
six parameters identified in the figure need to be considered.
The intention of the problem is to dwindle the total mass of Minimize:
the cantilever beam when the bearing capacity is satisfied. The
formulae of this optimization problem are listed as follow: F (X ) = 0.6224(x1 + x2 + x3 + x4 + x5 )
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 317

Fig. 11. Comparisons between SMA and advanced MAs.

Subject to: Compared to MFO [26], SOS [89], CS [90], MMA [91] and
GCA [91], SMA can achieve better results in the cantilever design
61 37 19 7 1 problem, as shown in Table 22.
G (X) = + + + + ≤1
x31 x32 x33 x34 x35
3.5.4. I-beam structure problem
Variable ranges: The intention of this engineering problem is to minimize the
vertical deviation of I-beam by adjusting four parameters as
0.01 ≤ x1 .x2 , x3 , x4 , x5 ≤ 100 shown in Fig. 16.
318 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Fig. 12. The influence of populations and iterations.

Fig. 13. Structure of welded beam design based on [26].

Fig. 14. Structure of pressure vessel based on [26].

Table 23 lists the optimization results for SMA compared to The morphology of the slime mould also changes with three
ARSM [92], SOS [89], CS [90] and IARSM [92]. The data reveals different contraction patterns.
that SMA can obtain excellent optimal values in this engineer- To qualitatively analyse the algorithm, four metrics (search
ing problem, reflecting the applicability of SMA to engineering history, the trajectory of the first dimension, average fitness,
problems. and convergence curve) were applied. Then, the algorithm was
evaluated in 33 benchmark functions consisting of unimodal,
4. Conclusions and future perspectives multimodal, fix-dimension multimodal, and composite functions.
Most of the functions tested are composite functions. Wilcoxon
This paper proposed an effective optmizer motivated by slime sign-rank test and Friedman test were applied to estimate the
behaviour to tackle the optimization problems. The algorithm efficacy of the algorithm more scientifically. The experimental re-
mainly uses the weights to simulate the positive and negative sults illustrate that SMA can guarantee the performance of explo-
feedback of the bio-oscillator during the foraging to the food rations while achieving superior exploitations, thus maintaining
source to form a different thickness of the feeding vein network. an outstanding balance between exploitations and explorations,
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 319

Table 16
Results of Friedman test of iterative version and function evaluation version.
Iterative version on F1–33
SMA SCA SSA GWO MFO WOA GOA DA ALO MVO PBIL PSO DE
Avg 3.057 9.396 5.180 5.280 8.037 6.735 6.690 10.580 6.124 5.013 12.228 7.865 4.815
Rank 1 11 4 5 10 8 7 12 6 3 13 9 2
Iterative version on F1-33
SMA BLPSO CLPSO CBA RCBA CDLOBA m_SCA IWOA LWOA CSSA
Avg 2.297 7.578 6.996 5.507 5.408 6.100 5.000 4.710 4.907 6.497
Rank 1 10 9 6 5 7 4 2 3 8
Evaluation version on F1–21
SMA SCA SSA GWO MFO WOA GOA MVO PSO DE AGA
Avg 3.189 8.103 5.668 5.369 8.201 5.135 6.805 5.895 8.970 4.292 4.373
Rank 1 9 6 5 10 4 8 7 11 2 3

Table 17
Holms’ test (take SMA as the control algorithm).
SMAVS . z − v alue p-value α/i, α = 0.05 α/i, α = 0.10
PBIL 9.9878 8.6010E−24 0.0042 0.0083
DA 8.2494 7.9860E−17 0.0045 0.0091
SCA 6.9851 1.4240E−12 0.005 0.01
MFO 6.7639 6.7120E−12 0.0056 0.0111
PSO 4.9307 4.0900E−07 0.0063 0.0125
Traditional algorithms WOA 4.2669 9.9060E−06 0.0071 0.0143
GOA 3.8877 5.0540E−05 0.0083 0.0167
ALO 3.7296 9.5740E−05 0.01 0.02
GWO 2.9394 1.6460E−03 0.0125 0.025
SSA 2.3389 9.6680E−03 0.0167 0.0333
MVO 1.7384 0.04111 0.025 0.05
DE 1.6436 0.05009 0.05 0.1
BLPSO 12.0748 7.1760E−34 0.00556 0.01111
CLPSO 10.7331 3.5580E−27 0.00625 0.0125
CSSA 9.3915 2.9580E−21 0.00714 0.01429
CDLOBA 8.0498 4.1460E−16 0.00833 0.01667
Advanced algorithms
CBA 6.7082 9.8520E−12 0.01 0.02
RCBA 5.3666 4.0120E−08 0.0125 0.025
IWOA 4.0249 2.8500E−05 0.01667 0.03333
m_SCA 2.6833 0.00365 0.025 0.05
LWOA 1.3416 0.08986 0.05 0.1
PSO 5.648039 8.1160E−09 0.005 0.01
MFO 4.896673 4.8730E−07 0.0056 0.0111
SCA 4.801299 7.8820E−07 0.00625 0.0125
GOA 3.532738 2.0570E−04 0.0071 0.0143
Evaluation MVO 2.644126 0.0041 0.0083 0.0167
SSA 2.422361 0.0077 0.01 0.02
GWO 2.130033 0.0166 0.0125 0.025
WOA 1.901289 0.0286 0.0167 0.0333
AGA 1.156902 0.1237 0.025 0.05
DE 1.077811 0.1406 0.05 0.1

Fig. 15. Structure of cantilever beam based on [26].

which reflects the superior performance of the algorithm in a SMA is also applicable to engineering optimization problems in
statistical sense compared with other algorithms. real life with satisfactory optimization results.
Meanwhile, SMA was used in four classical engineering struc- The accounts for the satisfactory performance of SMA in main-
tural problems, including welded beam, pressure vessel, can- taining the balance of exploitation and exploration inclinations
tilever, and I-beam design problems. The results demonstrate that can be theoretically attributed to the following points:
320 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

Table 18
Wall-clock time costs of SMA and other candidates on 33 benchmarks.
SMA SCA SSA GWO MFO WOA GOA DA ALO MVO PSO DE
F1 14.040 1.310 0.811 1.825 1.513 0.562 119.466 90.262 218.355 3.822 0.967 5.054
F2 13.291 1.139 0.796 1.622 1.342 0.577 118.108 110.886 216.748 3.806 0.920 4.446
F3 13.478 2.480 2.215 2.839 2.636 2.278 110.012 116.268 207.606 4.165 1.919 5.288
F4 12.776 1.123 0.796 1.560 1.123 0.546 117.188 86.659 207.513 2.761 0.640 3.838
F5 12.995 1.404 1.030 1.950 1.466 0.780 123.646 115.768 233.112 3.401 0.874 4.976
F6 15.241 1.513 1.045 2.090 1.544 0.686 146.376 122.929 260.163 3.760 0.889 4.914
F7 16.037 2.356 1.919 2.855 2.465 1.638 141.665 118.857 257.995 4.883 1.794 5.678
F8 15.709 1.763 1.295 2.434 1.778 0.983 141.119 172.038 255.467 2.824 1.264 5.491
F9 16.115 1.576 1.123 2.028 1.669 0.764 142.133 136.657 259.305 3.994 1.139 4.742
F10 14.726 1.794 1.279 2.090 1.997 0.936 143.209 111.151 251.099 4.025 1.092 5.444
F11 16.115 2.215 1.607 2.309 1.981 0.998 144.878 121.338 264.063 4.040 1.217 5.600
F12 19.032 4.602 4.134 5.023 4.836 3.900 149.917 124.598 265.561 6.880 4.134 8.596
F13 18.939 4.852 4.274 5.101 4.243 3.604 149.153 126.205 266.684 6.833 4.087 8.518
F14 16.411 2.090 1.732 2.855 2.340 1.326 140.807 145.549 256.341 4.399 1.544 5.647
F15 15.725 2.106 1.498 2.465 1.950 1.108 144.894 136.080 261.583 4.524 1.373 5.990
F16 15.803 2.106 1.638 2.652 2.044 1.295 146.766 152.351 261.161 4.243 1.342 5.351
F17 31.715 16.677 15.943 17.254 16.224 15.678 163.832 185.080 281.145 19.438 16.021 20.686
F18 16.177 2.090 1.544 2.621 2.246 1.279 146.376 170.306 263.080 4.508 1.342 5.288
F19 16.006 2.075 1.560 2.434 1.872 1.092 142.554 130.791 263.767 4.446 1.435 5.850
F20 2.387 1.700 2.465 2.153 1.295 147.062 143.318 264.250 4.493 1.529 5.366
F21 16.177 2.246 1.716 2.683 2.278 1.295 145.674 171.820 262.238 4.680 1.451 5.444
F22 15.897 2.231 1.654 2.808 2.340 1.513 145.097 172.787 258.665 4.446 1.654 5.881
F23 15.413 1.919 1.576 2.480 2.200 1.357 140.183 144.192 253.080 4.555 1.404 5.928
F24 19.531 5.054 4.633 5.491 4.976 4.072 150.635 128.264 268.993 7.940 4.586 9.064
F25 16.552 2.262 1.638 2.730 2.184 1.310 145.112 120.901 265.670 4.384 1.513 5.710
F26 13.057 2.168 1.529 2.168 1.763 1.154 118.015 119.450 206.592 3.697 1.498 4.836
F27 14.414 2.122 1.763 2.371 2.059 1.373 115.644 115.862 210.289 3.962 1.466 5.179
F28 16.833 4.477 3.572 4.914 4.618 3.385 125.331 128.810 216.873 5.632 3.182 7.784
F29 15.663 3.526 2.933 3.635 2.980 2.278 117.656 89.155 216.046 5.429 2.668 6.942
F30 17.831 4.274 3.448 4.711 4.228 3.292 141.462 165.829 248.775 6.646 3.869 7.847
F31 33.665 20.296 20.327 19.890 19.563 19.781 162.257 187.373 277.261 22.464 19.095 24.633
F32 33.915 20.530 19.656 20.155 19.438 19.251 164.488 187.700 280.474 22.729 19.594 23.681
F33 18.705 5.850 5.554 6.334 5.834 4.586 150.042 124.411 269.461 8.237 4.898 9.547

Table 19
Ranking of results with varied values of parameter z.
Function SMA0 SMA1 SMA2 SMA3 SMA4 SMA5 SMA6 SMA7 SMA8 SMA9 SMA10
1 1 1 1 1 1 6 7 8 9 10 11
2 1 1 3 4 5 6 7 9 8 10 11
3 1 1 1 1 5 6 7 8 9 10 11
4 1 1 3 4 5 6 7 8 9 10 11
5 11 10 9 8 7 3 4 2 6 5 1
6 11 1 2 3 4 5 6 9 7 8 10
7 1 2 4 3 6 7 10 8 9 5 11
8 11 1 4 5 6 8 9 2 7 10 3
9 1 1 1 1 1 1 1 1 1 1 1
10 1 1 1 1 1 1 1 1 1 1 1
11 1 1 1 1 1 1 1 1 1 1 1
12 11 10 9 7 6 8 5 4 2 3 1
13 11 10 2 1 8 4 5 3 9 7 6
Average 4.85E+00 3.15E+00 3.15E+00 3.08E+00 4.31E+00 4.77E+00 5.38E+00 4.92E+00 6.00E+00 6.23E+00 6.08E+00
Rank 5 2 2 1 3 4 7 6 8 10 9

Fig. 16. Structure of I-beam based on [26].

−
→
• The adaptive weight W enables the SMA to maintain a cer- • Vibration parameter v b allows the individual position of
tain disturbance rate while guaranteeing fast convergence, slime mould to contract in a specific way, thus ensuring the
thus avoiding optimal local trapping during fast conver- efficiency of the early exploration and the accuracy of the
gence. later exploitation.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 321

Table 20 enhance the efficacy of the algorithm. The binary version of the
Results of welded beam structure problem compared with other competitors. algorithm can also be developed for feature selection. Moreover,
Algorithms Optimum values for variables Optimum cost SMA can also be used to optimize parameters of classifiers such
h l t b as kernel extreme learning machine or support vector machine.
SMA 0.2054 3.2589 9.0384 0.2058 1.69604
MFO [26] 0.2057 3.4703 9.0364 0.2057 1.72452 Declaration of competing interest
SSA [55] 0.2057 3.4714 9.0366 0.2057 1.72491
Random [79] 0.4575 4.7313 5.0853 0.6600 4.11856
Siddall [80] 0.2444 6.2189 8.2915 0.2444 2.38154
The authors declare that they have no known competing finan-
Ragsdell [79] 0.2455 6.1960 8.2730 0.2455 2.38594 cial interests or personal relationships that could have appeared
Coello and Montes [81] 0.2060 3.4713 9.0202 0.2065 1.72822 to influence the work reported in this paper.
GWO [24] 0.2057 3.4784 9.0368 0.2058 1.72624
WOA [53] 0.2054 3.4843 9.0374 0.2063 1.73050
CRediT authorship contribution statement
GSA 0.1821 3.8570 10.0000 0.2024 1.87995
Simplex [79] 0.2792 5.6256 7.7512 0.2796 2.53073
David [79] 0.2434 6.2552 8.2915 0.2444 2.38411 Shimin Li: Writing - original draft, Writing - review & editing,
Software, Visualization, Investigation. Huiling Chen: Conceptu-
alization, Methodology, Formal analysis, Investigation, Writing
Table 21
Results of pressure vessel design problem compared with other competitors. - review & editing, Funding acquisition, Supervision. Mingjing
Algorithms Optimum values for variables Optimum
Wang: Writing - review & editing, Software. Ali Asghar Heidari:
cost Methodology, Formal analysis, Investigation, Writing - original
Ts Th R L draft, Writing - review & editing, Software, Visualization. Seyedali
Mirjalili: Supervision, Writing - review & editing, Investigation.
SMA 0.7931 0.3932 40.6711 196.2178 5994.1857
MFO [26] 0.8125 0.4375 42.0984 176.6366 6059.7143
BA [82] 0.8125 0.4375 42.0984 176.6366 6059.7143 Acknowledgements
HPSO [83] 0.8125 0.4375 42.0984 176.6366 6059.7143
CSS [7] 0.8125 0.4375 42.1036 176.5727 6059.0888 This research is supported by the Science and Technology Plan
CPSO [84] 0.8125 0.4375 42.0912 176.7465 6061.0777
ACO [85] 0.8125 0.4375 42.1036 176.5727 6059.0888
Project of Wenzhou, China (2018ZG012), and the National Natural
GWO [24] 0.8125 0.4345 42.0892 176.7587 6051.5639 Science Foundation of China (U1809209).
WOA [53] 0.8125 0.4375 42.0983 176.6390 6059.7410
MDDE [86] 0.8125 0.4375 42.0984 176.6360 6059.7017 References
Lagrangian multiplier [87] 1.1250 0.6250 58.2910 43.6900 7198.0428
Branch-bound [88] 1.1250 0.6250 47.7000 117.7010 8129.1036 [1] H. Chen, et al., An opposition-based sine cosine approach with local search
for parameter estimation of photovoltaic models, Energy Convers. Manage.
195 (2019) 927–942.
Table 22 [2] H. Chen, et al., A balanced whale optimization algorithm for constrained
Results of cantilever beam structural problem in comparison with other engineering design problems, Appl. Math. Model. 71 (2019) 45–59.
competitors. [3] M. Wang, H. Chen, Chaotic multi-swarm whale optimizer boosted support
Algorithm Optimum values for variables Optimum vector machine for medical diagnosis, Appl. Soft Comput. (2019) 105946,
cost http://dx.doi.org/10.1016/j.asoc.2019.105946.
[4] M. Wang, et al., Toward an optimal kernel extreme learning machine using
X1 X2 X3 X4 X5
a chaotic moth-flame optimization strategy with applications in medical
SMA 6.017757 5.310892 4.493758 3.501106 2.150159 1.339957 diagnoses, Neurocomputing 267 (Supplement C) (2017) 69–84.
MFO [26] 5.9830 5.3167 4.4973 3.5136 2.1616 1.33998 [5] S. Osher, et al., Laplacian Smoothing Gradient Descent, 2018.
SOS [89] 6.0188 5.3034 4.4959 3.4990 2.1556 1.33996 [6] S. Mirjalili, S.M. Mirjalili, A. Hatamlou, Multi-verse optimizer: a nature-
CS [90] 6.0089 5.3049 4.5023 3.5077 2.1504 1.33999 inspired algorithm for global optimization, Neural Comput. Appl. 27 (2)
MMA [91] 6.0100 5.3000 4.4900 3.4900 2.1500 1.3400 (2016) 495–513.
GCA [91] 6.0100 5.3000 4.4900 3.4900 2.1500 1.3400 [7] A. Kaveh, S. Talatahari, S. Talatahari, A novel heuristic optimization
method: Charged system search, Acta Mech. 213 (3-4) (2010) 267–289,
Vol. 213, 267-289.
Table 23 [8] E. Rashedi, H. Nezamabadi-pour, S. Saryazdi, GSA: A Gravitational Search
Results of I-beam structural problem in comparison with other methods. Algorithm, Vol. 179, 2009, pp. 2232–2248.
[9] S. Mirjalili, SCA: A Sine Cosine algorithm for solving optimization problems,
Algorithm Optimum values for variables Optimum
Knowl.-Based Syst. 96 (2016) 120–133.
cost [10] R. Venkata Rao, V. Savsani, D.P. Vakharia, Teaching–Learning-Based Opti-
b h Tw Tf mization: An Optimization Method for Continuous Non-Linear Large Scale
Problems, Vol. 183, 2012, pp. 1–15.
SMA 49.998845 79.994327 1.764747 4.999742 0.006627
[11] R. Formato, Central Force Optimization: A New Metaheuristic with
ARSM [92] 37.0500 80.0000 1.7100 2.3100 0.0157
Applications in Applied Electromagnetics, Vol. 77, 2007, pp. 425–491.
SOS [89] 50.0000 80.0000 0.9000 2.3218 0.0131
[12] D. Fogel, Artificial Intelligence Through Simulated Evolution, 2009, pp.
CS [90] 50.0000 80.0000 0.9000 2.3217 0.0131 227–296.
IARSM [92] 48.4200 79.9900 0.9000 2.4000 0.1310 [13] L.B. Booker, D.E. Goldberg, J.H. Holland, Classifier systems and genetic
algorithms, Artificial Intelligence 40 (1) (1989) 235–282.
[14] J.R. Koza, J.P. Rice, Automatic programming of robots using genetic
programming, in: Proceedings Tenth National Conference on Artificial
• The adequate utilization of individual fitness values allows Intelligence, 1992.
SMA to make better decisions based on historical informa- [15] N. Hansen, S.D. Müller, P. Koumoutsakos, Reducing the Time Complexity of
the Derandomized Evolution Strategy with Covariance Matrix Adaptation
tion. (CMA-ES), Vol. 11, 2003, pp. 1–18.
• The location updating decision parameter p and three dif- [16] X. Yao, Y. Liu, G. Lin, Evolutionary Programming Made Faster, Vol. 3, 1999,
ferent location updating methods ensure better adaptability pp. 82–102.
of the SMA in different search phases. [17] R. Storn, K. Price, Differential Evolution - A Simple and Efficient Heuris-
tic for Global Optimization over Continuous Spaces, Vol. 11, 1997, pp.
On the purpose of improving the extensibility of the algorithm, 341–359.
[18] Weibiao Qiao, Hossein Moayedi, Loke Kok Foong, Nature-inspired hybrid
the development of the algorithm is established on the principle techniques of IWO, DA, ES, GA, and ICA, validated through a k-fold
of being as simple as possible. In future work, various mutation validation process predicting monthly natural gas consumption, Energy
mechanisms or acceleration mechanisms can be employed to Build. (2020) 110023, http://dx.doi.org/10.1016/j.enbuild.2020.110023.
322 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323

[19] H. Moayedi, S. Hayati, Modelling and optimization of ultimate bearing [57] D. Simon, Biogeography-Based Optimization, Vol. 12, 2009, pp. 702–713.
capacity of strip footing near a slope by soft computing methods, Appl. [58] R. Storn, K. Price, Differential evolution - a simple and efficient heuristic
Soft. Comput. 66 (2018) 208–219, http://dx.doi.org/10.1016/j.asoc.2018.02. for global optimization over continuous spaces, J. Global Optim. 11 (4)
027. (1997) 341–359.
[20] H. Moayedi, A. Rezaei, An artificial neural network approach for under- [59] X. Ju, F. Liu, Wind farm layout optimization using self-informed genetic
reamed piles subjected to uplift forces in dry sand, Neural Comput. Appl. algorithm with information guided exploitation, Appl. Energy 248 (2019)
31 (2) (2019) 327–336, http://dx.doi.org/10.1007/s00521-017-2990-z. 429–445.
[21] G. Beni, J. Wang, Swarm Intelligence in Cellular Robotic Systems, 1993, pp. [60] X. Chen, et al., Biogeography-Based Learning Particle Swarm Optimization,
703–712. Vol. 21, 2017, pp. 7519–7541.
[22] J. Kennedy, R. Eberhart, Particle swarm optimization, in: IEEE International [61] Y. Cao, et al., Comprehensive Learning Particle Swarm Optimization
Conference on Neural Networks - Conference Proceedings, 1995. Algorithm with Local Search for Multimodal Functions, 2018, p. 1.
[23] X.-S. Yang, A New Metaheuristic Bat-Inspired Algorithm, Vol. 284, 2010. [62] B.R. Adarsh, et al., Economic Dispatch Using Chaotic Bat Algorithm, Vol.
[24] S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer, Adv. Eng. Softw. 96, 2016, pp. 666–675.
[63] H. Liang, et al., A Hybrid Bat Algorithm for Economic Dispatch with
69 (2014) 46–61.
Random Wind Power, 2018, p. 1.
[25] W.T. Pan, A new fruit fly optimization algorithm: Taking the financial
[64] J. Yong, et al., A Novel Bat Algorithm Based on Collaborative and Dynamic
distress model as an example, Knowl.-Based Syst. 26 (2012) 69–74.
Learning of Opposite Population, 2018, pp. 541–546.
[26] S. Mirjalili, Moth-flame optimization algorithm: A novel nature-inspired
[65] S. Gupta, K. Deep, A Hybrid Self-Adaptive Sine Cosine Algorithm with
heuristic paradigm, Knowl.-Based Syst. 89 (2015) 228–249.
Opposition Based Learning, Vol. 119, 2018.
[27] M. Dorigo, C. Blum, Ant colony optimization theory: A survey, Theoret. [66] M. Tubishat, et al., Improved Whale Optimization Algorithm for Feature
Comput. Sci. 344 (2–3) (2005) 243–278. Selection in Arabic Sentiment Analysis, 2018.
[28] A.A. Heidari, et al., Harris hawks optimization: Algorithm and applications, [67] E. Emary, H. Zawbaa, M. Sharawi, Impact of Lèvy Flight on Modern
Future Gener. Comput. Syst. 97 (2019) 849–872. Meta-Heuristic Optimizers, 2018.
[29] D. Karaboga, B. Basturk, A powerful and efficient algorithm for numerical [68] G. Ismail Sayed, G. Khoriba, M. Haggag, A Novel Chaotic Salp Swarm
function optimization: Artificial bee colony (ABC) algorithm, J. Global Algorithm for Global Optimization and Feature Selection, 2018.
Optim. 39 (3) (2007) 459–471. [69] J. Alcala-Fdez, et al., KEEL: A Software Tool to Assess Evolutionary
[30] L. Lin, M. Gen, Auto-Tuning Strategy for Evolutionary Algorithms: Balancing Algorithms for Data Mining Problems, Vol. 13, 2009, pp. 307–318.
Between Exploration and Exploitation, Vol. 13, 2009, pp. 157–168. [70] T. Eftimov, P. Korošec, A novel statistical approach for comparing meta-
[31] D. Wolpert, W. Macready, No Free Lunch Theorems for Optimization, Vol. heuristic stochastic optimization algorithms according to the distribution
1, 1997, pp. 67–82. of solutions in the search space, Inform. Sci. 489 (2019) 255–273.
[32] D. Monismith, B. Mayfield, Slime Mold as a Model for Numerical [71] D.J. Sheskin, Handbook of Parametric and Nonparametric Statistical
Optimization, 2008, pp. 1–8. Procedures, Chapman & Hall/CRC, 2007.
[33] K. Li, et al., Slime Mold Inspired Routing Protocols for Wireless Sensor [72] S. Holm, A simple sequentially rejective multiple test procedure, Scand. J.
Networks, Vol. 5, 2011, pp. 183–223. Stat. 6 (1979) 65–70.
[34] T. Qian, et al., An Ant Colony System Based on the Physarum Network, [73] W. Gao, H. Wu, M.K. Siddiqui, A.Q. Baig, Study of biological networks using
2013, pp. 297–305. graph theory, Saudi J. Bio. Sci. 25 (6) (2018) 1212–1219.
[35] T. Schmickl, K. Crailsheim, A Navigation Algorithm for Swarm Robotics [74] W. Gao, W. Wang, D. Dimitrov, Y. Wang, Nano properties analysis via
Inspired by Slime Mold Aggregation, Vol. 4433, 2006, pp. 1–13. fourth multiplicative ABC indicator calculating, Arabian J. Chem. 11 (6)
[36] M. Becker, On the Efficiency of Nature-Inspired Algorithms for Generation (2018) 793–801.
of Fault-Tolerant Graphs, 2015, pp. 1657–1663. [75] W. Gao, J.L.G. Guirao, B. Basavanagoud, J. Wu, Partial multi-dividing
[37] Anthony Brabazon, Seán McGarraghy, Int. J. Innovative Comput. Appl. (1) ontology learning algorithm, Inf. Sci. 467 (2018) 35–58.
(2020) 30–45, http://dx.doi.org/10.1504/IJICA.2020.105316. [76] W. Gao, J.L.G. Guirao, M. Abdel-Aty, W. Xi, An independent set degree con-
[38] F.L. Howard, The Life History of Physarum Polycephalum, Vol. 18, 1931. dition for fractional critical deleted graphs, Discrete Continuous Dynamical
[39] D. Kessler, Plasmodial Structure and Motility, 1982, pp. 145–208. Systems-S 12 (4&5) (2019) 877–886.
[40] W.G. Camp, A Method of Cultivating Myxomycete Plasmodia, Vol. 63, 1936, [77] Weibiao Qiao, Li Bingfan, Kang Zhangyang, Differential scanning calorime-
pp. 205–210. try and electrochemical tests for the analysis of delamination of 3PE
[41] N. Kamiya, The Control of Protoplasmic Streaming, Vol. 92, 1940, pp. coatings, Int. J. Electrochem. Sci. 14 (2019) 7389–7740, http://dx.doi.org/
462–463. 10.20964/2019.08.05.
[78] H. Moayedi, S. Hayati, Applicability of a cpt-based neural network solution
[42] T. Nakagaki, H. Yamada, T. Ueda, Interaction Between Cell Shape and
in predicting load-settlement responses of bored pile, Int. J. Geomech.
Contraction Pattern in the Physarum Plasmodium, Vol. 84, 2000, pp.
18 (6) (2018) 6018009, http://dx.doi.org/10.1061/(ASCE)GM.1943-5622.
195–204.
0001125.
[43] M. Becker, On the efficiency of nature-inspired algorithms for generation of
[79] K. Ragsdell, D.T. Phillips, Optimal Design of a Class of Welded Structures
fault-tolerant graphs, in: Proceedings - 2015 IEEE International Conference
using Geometric Programming, Vol. 98, 1976.
on Systems, Man, and Cybernetics, SMC 2015, 2016.
[80] J.N. Sidall, Analytical Decision-Making in Engineering Design, 1972.
[44] V. Šešum Čavić, E. Kühn, D. Kanev, Bio-inspired search algorithms for [81] A.C. Coello Coello, E. Mezura-Montes, Constraint-Handling in Genetic
unstructured P2P overlay networks, Swarm Evol. Comput. 29 (2016) 73–93. Algorithms Through the Use of Dominance-Based Tournament Selection,
[45] K. Daniel Yu, et al., Bicycle pathway generation through a weighted digital Vol. 16, 2002, pp. 193–203.
slime mold algorithm via topographical analysis, in: CAADRIA 2018-23rd [82] A.H. Gandomi, X. Yang, A.H. Alavi, et al., Bat algorithm for constrained
International Conference on Computer-Aided Architectural Design Research optimization tasks, Neural Comput. Appl. 22 (2013) 1239–1255, http:
in Asia: Learning, Prototyping and Adapting, 2018. //dx.doi.org/10.1007/s00521-012-1028-9.
[46] M. Beekman, T. Latty, Brainless but multi-headed: Decision making by the [83] Q. He, L. Wang, L. Wang, A hybrid particle swarm optimization with a
acellular slime mould physarum polycephalum, J. Mol. Biol. 427 (23) (2015) feasibility-based rule for constrained optimization, Appl. Math. Comput.
3734–3743. 186 (2007) 1407–1422, Vol. 186, 1407-1422.
[47] T. Latty, M. Beekman, Food Quality and the Risk of Light Exposure Affect [84] Q. He, L. Wang, An Effective Co-Evolutionary Particle Swarm Optimization
Patch-Choice Decisions in the Slime Mold Physarum Polycephalum, Vol. for Constrained Engineering Design Problems, Vol. 20, 2007, pp. 89–99.
91, 2010, pp. 22–27. [85] A. Kaveh, S. Talatahari, An Improved Ant Colony Optimization for
[48] T. Latty, M. Beekman, Speed-Accuracy Trade-Offs During Foraging Decisions Constrained Engineering Design Problems, Vol. 27, 2010, pp. 155–182.
in the Acellular Slime Mould Physarum Polycephalum, Vol. 278, 2011, pp. [86] E. Mezura-Montes, et al., Multiple Trial Vectors in Differential Evolution
539–545. for Engineering Design, Vol. 39, 2007, pp. 567–589.
[49] T. Latty, M. Beekman, Slime Moulds Use Heuristics Based on Within-Patch [87] B.K. Kannan, S.N. Kramer, An Augmented Lagrange Multiplier Based
Experience to Decide when to Leave, Vol. 218, 2015. Method for Mixed Integer Discrete Continuous Optimization and its
[50] P. Kareiva, G. Odell, Swarms of Predators Exhibit Preytaxis if Individual Applications to Mechanical Design, Vol. 116, 1994.
Predators Use Area-Restricted Search, Vol. 130, 1987. [88] E. Sandgren, Nonlinear Integer and Discrete Programming in Mechanical
[51] T. Latty, M. Beekman, Food quality affects search strategy in the acellular Design, Vol. 14, 1988.
slime mould, Physarum polycephalum. 20 (2009) 1160–1167. [89] M.-Y. Cheng, D. Prayogo, Symbiotic Organisms Search: A New Metaheuris-
[52] F. van den Bergh, A. Engelbrecht, A Study of Particle Swarm Optimization tic Optimization Algorithm, Vol. 139, 2014.
Particle Trajectories, Vol. 176, 2006, pp. 937–971. [90] A. Gandomi, X.-S. Yang, A. Alavi, Cuckoo Search Algorithm: a Metaheuristic
[53] S. Mirjalili, A. Lewis, The whale optimization algorithm, Adv. Eng. Softw. Approach to Solve Structural Optimization Problems, Vol. 29, 2013, pp.
95 (2016) 51–67. 1–19.
[54] X.S. Yang, Firefly algorithms for multimodal optimization, in: Lecture [91] H. Chickermane, H.C. Gea, Structural Optimization Using a New Local
Notes in Computer Science (including subseries Lecture Notes in Artificial Approximation Method, Vol. 39, 1996, pp. 829–846.
[92] G. Wang, Adaptive Response Surface Method Using Inherited Latin
Intelligence and Lecture Notes in Bioinformatics), 2009, pp. 169–178.
Hypercube Design Points, Vol. 125, 2003, pp. 210–220.
[55] S. Mirjalili, et al., Salp swarm algorithm: A bio-inspired optimizer for
engineering design problems, Adv. Eng. Softw. 114 (2017) 163–191.
[56] S. Mirjalili, The Ant Lion Optimizer, Vol. 83, 2015.
 S. Li, H. Chen, M. Wang et al. / Future Generation Computer Systems 111 (2020) 300–323 323

Shimin Li is now a student in the college of com- Ali Asghar Heidari is a research intern at the School of
puter science and artificial intelligence at Wenzhou Computing, National University of Singapore (NUS). He
University, China. His present research interests are is also an exceptionally talented Ph.D. at the University
meta-heuristics and machine learning. of Tehran and he is awarded and funded by Iran’s
National Elites Foundation (INEF). He has been recog-
nized by publons.com as the top 1% peer reviewers in
computer science and cross-field. His research focuses
specifically on Advanced Machine Learning, Evolution-
ary Computation, Optimization, Prediction, Information
Systems, and Spatial Modelling. He has authored more
than 40 journal articles with over 1.45K citations (H-
Huiling Chen is currently an associate professor in index of 18) and over 5.4K readers in prestigious international journals such as
the college of computer science and artificial intelli- Information Fusion, Information Sciences, Future Generation Computer Systems,
gence at Wenzhou University, China. He received his Journal of Cleaner Production, Energy, Energy Conversion and Management,
Ph.D. degree in the department of computer science Applied Soft Computing, Knowledge-Based Systems, and Expert Systems with
and technology at Jilin University, China. His present Applications.
research interests centre on machine learning and data
mining, as well as their applications to medical di-
Seyedali Mirjalili is an associate professor and the
agnosis and bankruptcy prediction. He has published
director of the Centre for Artificial Intelligence Research
more than 100 papers in international journals and
and Optimization at Torrens University Australia. He is
conference proceedings, including Pattern Recognition,
internationally recognized for his advances in Swarm
Information Sciences, Expert Systems with Applica-
Intelligence and Optimization, including the first set
tions, Knowledge-Based Systems, Soft Computing, Neurocomputing, PAKDD, and
of algorithms from a synthetic intelligence standpoint
among others.
- a radical departure from how natural systems are
typically understood - and a systematic design frame-
work to reliably benchmark, evaluate, and propose
Mingjing Wang is now a Ph.D. research intern at the
computationally cheap robust optimization algorithms.
School of Computer Science and Engineering, Southeast
Ali has published over 200 publications with over
University in China. Currently, he is also a member of
16,000 citations and an H-index of 44. As the most cited researcher in Robust
the Medical Data Mining and Computation Intelligence
Optimization, he is in the list of 1% highly-cited researchers and named as
in the college of computer science and artificial intelli-
one of the most influential researchers in the world by Web of Science.
gence at Wenzhou University, China. His main research
Ali is a senior member of IEEE and an associate editor of several journals
interests are advanced machine learning, evolutionary
including Neurocomputing, Applied Soft Computing, Advances in Engineering
computation, meta-heuristics, prediction, information
Software, Applied Intelligence, and IEEE Access. His research interests include
systems, and spatial modelling.
Robust Optimization, Engineering Optimization, Multi-objective Optimization,
Swarm Intelligence, Evolutionary Algorithms, and Artificial Neural Networks.
He is working on the application of multi-objective and robust meta-heuristic
optimization techniques as well.