Alexandria Engineering Journal 83 (2023) 257–285

Contents lists available at ScienceDirect

Alexandria Engineering Journal

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

Original Article

Snow avalanches algorithm (SAA): A new optimization algorithm for

engineering applications
Keyvan Golalipour a, Saber Arabi Nowdeh b, Ebrahim Akbari c, Seyed Saeed Hamidi d,
Danyal Ghasemi e, Almoataz Y. Abdelaziz f, Hossam Kotb g, *, Amr Yousef h, i
a
Department of Computer Engineering, Sari Branch, Islamic Azad University, Sari, Iran
b
Golestan Technical and Vocational Training Center, Gorgan, Iran
c
Department of Computer Engineering, Islamic Azad University, Sari Branch, Sari, Iran
d
Department of Computer Engineering, Qaemshahr Branch, Islamic Azad University, Qaemshahr, Iran
e
Department of Electrical Engineering, Aliabad Katoul Branch, Islamic Azad University, Aliabad Katoul, Iran
f
Faculty of Engineering and Technology, Future University in Egypt, Cairo 11835, Egypt
g
Department of Electrical Power and Machines, Faculty of Engineering, Alexandria University, Alexandria 21544, Egypt
h
Engineering Mathematics Department, Faculty of Engineering, Alexandria University, Alexandria 21544, Egypt
i
Electrical Engineering Department, University of Business and Technology, Ar Rawdah, Jeddah, 23435, Saudi Arabia

A R T I C L E I N F O A B S T R A C T

Keywords: This paper proposes a novel efficient inspired algorithm based on snow avalanches in nature which is named the
Optimization Snow Avalanches Algorithm (SAA), for solving the benchmark and engineering optimization problems and
Meta-heuristic Algorithm determining the global solution. The proposed algorithm is modeled using four phases including avalanche due
Engineering Optimization Problem
to mountain slope, human factors, weather in the region as well as normal conditions and it has only one control
Benchmark Test Functions
Snow Avalanches Algorithm
parameter. The advantages of this algorithm are low control parameters, simple structure and also easy
implementation. The effectiveness of the SAA algorithm is examined on 23 classic benchmark test functions.
Then, the effectiveness of the SAA to achieve accurate results in different aspects is examined and proved on
engineering problems including six different cases. The superiority of the SAA to solve the classic benchmark test
functions is compared with spotted hyena optimization (SHO), particle swarm optimization (PSO), Aquila
optimizer (AO), differential evolution (DE), bat algorithm (BA), dwarf mongoose optimization (DMO), genetic
algorithm (GA), artificial bee colony (ABC), and ant colony optimization (ACO). The simulation results provide
evidence for the well-organized and efficient performance of the SAA in solving a great diversity of engineering
problems. The results demonstrated that the SAA can be more effective than other algorithms to solve the test
functions in terms of optimization accuracy and convergence rate. Moreover, the results proved that the SAA
obtained more competitive results than the previous methods to solve constrained engineering optimization
problems, especially hybrid energy system design as well as economic load dispatch problems.

1. Introduction differentiation of the objective functions in solving real-world engi­

neering optimization problems [1,2]. Multifaceted problems such as
In different scientific fields, various optimization problems are real-world engineering problems include many control variables that
widely seen. Decision variables in solving an optimization problem determine their optimal value and are very effective on the output of
should be optimally determined by an algorithm. The goal of optimi­ these systems. Therefore, the optimal setting of these variables and
zation is to determine the best feasible solution for a specific problem by achieving an optimal solution requires providing an optimization model
satisfying the constraints of the problem [1,2]. Most of the conventional and a powerful optimization method. The employment of meta-heuristic
optimization methods are according to the first derivative, which does techniques to tackle optimization issues has been highly popular in
not guarantee the achievement of optimal solutions due to the lack of recent years owing to their rapid speed of finding nearly optimum

* Corresponding author.
E-mail addresses: [email protected] (E. Akbari), [email protected] (D. Ghasemi), [email protected] (A.Y. Abdelaziz),
[email protected] (H. Kotb), [email protected] (A. Yousef).

https://doi.org/10.1016/j.aej.2023.10.029
Received 15 May 2023; Received in revised form 28 September 2023; Accepted 15 October 2023
Available online 31 October 2023
1110-0168/© 2023 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Alexandria University. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 1. Avalanche formation due to human factors.

solutions, lack of participation with side computations like the deriva­ first category is the methods that move to the lowest slope of the curve
tive, and advantages of the component Random. As a result, these based on repetition and derivative and depend on the initial guess,
techniques are employed to resolve real-world optimization issues for derivation or gradient, problem type, and search space. These methods
which it is difficult or impossible to calculate the derivative analytically do not guarantee the achievement of the global optimum and are not
[3]. Given the variety of optimization challenges, offering fresh ap­ able to provide an optimal solution in solving real-world optimization
proaches or enhancing established ones may result in the best answers problems [8,9]. Another category of optimization methods is MHAs. The
[4]. With the presentation of multiple optimization algorithms, the MHAs start with an initial population creation and each member is a
question arises as to why other algorithms are still needed. This question candidate solution to the problem. By randomly but purposefully
should be answered using the theory of no free lunch (NFL) [5]. Ac­ changing the original population of the algorithm, a new or updated
cording to the logical validity of this theory, an optimization algorithm population is generated in each iteration. The fitness value for each
may show very competitive and suitable results in solving a certain population member clears the value of that member. The steps described
category of optimization problems, but the same algorithm may perform in each step of the algorithm are repeated by selecting better solutions.
poorly to solve another category of optimization problems [6,7]. Meta-heuristic methods with random behavior try to simulate nature’s
Therefore, there is still a need to provide new meta-heuristic algorithms behavior in choosing the best ones [10,11]. In the literature review,
(MHAs) to solve optimization problems, which is the main motivation of studies on the presentation of the MHAs were divided into three cate­
the authors to provide a new and efficient meta-heuristic algorithm gories: improving and modifying existing meta-heuristic algorithms,
(MHA) named the snow avalanches algorithm (SAA) for solving real- combining meta-heuristic algorithms with other algorithms, and pre­
world optimization problems. senting new MHAs. In the first category, the ability to optimize the
The structure of the paper is presented as follows: In section 2, MHAs based on mathematical or random operators is improved. In [12]
literature review is described. A mathematical model of the SAA is chaos theory, in [13] modified oppositional chaotic local search strategy
formulated in section 3. In section 4, the optimization results of the SAA based Aquila optimizer (CmOBL-AO), in [14,15] evolutionary operators,
implementation on test benchmark functions and real-world problems and in [16,17] algorithms based on local search are presented to
and also comparative study results are presented. Also, the conclusions improve the capability of the MHAs. In the second category, the capa­
are presented in section 5. bility to optimize the MHAs is improved by combining with each other
to benefit from the advantages of each or to eliminate the shortcomings
2. Literature review of each [18,19] such as PSO-GA [20], CSA-DE [21], WCO-FSO [22],
hybrid butterfly and flower pollination algorithm (MBFPA) [23], GWO-
Optimization problems are divided into two classifications including SCA [24], GWO-CSA [25], enhanced Aquila optimizer (enAO) [26], and
traditional algorithms [8,9] and meta-heuristic methods [10,11]. The GWO-TLBO [27]. In the third category, new MHAs are presented. New

258
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

objective marine predator algorithm (MOMPA) [36], golden jackal

optimization (GJO) [37], artificial bee colony (ABC) [38] and bat al­
gorithm (BA) [39]. Some algorithms based on physics and chemistry
include gravitational search algorithm (GSA) [40], multi-objective
evolutionary algorithm (MOEA) [41], turbulent flow of water-based
optimization (TFWO) [42], artificial chemical reaction optimization
algorithm (ACROA) [43], water evaporation optimization (WEO) [44],
supernova optimizer (SO) [45] and some of the human-related algo­
rithms include the league championship algorithm (LCA) [46], mine
blast algorithm (MBA) [47], gaining–sharing knowledge-based optimi­
zation algorithm (GSK) [48], teaching–learning-based pathfinder algo­
rithm (TLPFA) [49], imperialist competitive algorithm (ICA) [50],
Driving Training-Based Optimization (DTBO) [51] and volleyball pre­
mier league algorithm (VPLA) [52].
Literature analysis has demonstrated that when tackling real-world
problems, we encounter a wide variety of intricate issues with wildly
varying performance. Why are new optimization algorithms still being
developed given the enormous number of existing algorithms? The No
Free Lunch Theorem of Optimization [5], which asserts that no opti­
mizer performs optimally for all optimization applications, provides the
solution. Therefore, metaheuristic algorithms with different formula­
tions are needed to achieve acceptable optimal solutions for different
test functions. Various MHAs with different capabilities have been pre­
sented in recent years, but in solving real-world problems, the use of
simpler, more comprehensive MHAs, with fewer control parameters and
with wide functional capabilities, has been taken into consideration,
such that it is feasible to guarantee the achievement of the solution in
most optimization problems, optimally.
The research gaps that metaheuristic algorithms faced in the litera­
ture review are listed as follows:

• A lot of the optimization techniques that are given include numerous

modification parameters. One of the main issues of applying these
algorithms to tackle various optimization problems is that the
adjustment parameters value has a significant impact on the algo­
rithm capabilities. A meta-heuristic algorithm’s capacity to solve
different optimization problems can therefore be improved by giving
it a limited set of control parameters.
• The fact that the aforementioned algorithms only address a portion
of the difficulties and cannot offer a workable solution to the other
problems is another difficulty.
• Many of the meta-heuristic algorithms that have been proposed have
complicated formulations, making it challenging to use them to solve
different optimization issues.
• Some algorithms call for intricate computations, which makes it
difficult for users to apply them to actual problems.

In this paper, a snow avalanches algorithm is presented based on the

avalanche phenomenon and its factors such as mountain slopes, regional
weather, and human factors. The contributions of this paper to address
the mentioned research gaps are described as follows:
The SAA is motivated by the need for efficient meta-heuristic nature-
human inspired optimizer to address particular real-world situations
and is based on the most crucial avalanche elements, including moun­
tain slopes, local weather, and human aspects, which have been taken
Fig. 2. The flowchart of the SAA. into account and equated in the presentation of this optimization
approach.
MHAs are inspired by the evolutionary approach, swarm intelligence Unlike many metaheuristic algorithms that have several control
behavior, the laws of physics and chemistry, and human-related be­ parameters and therefore need to spend a lot of time to get the best set,
haviors. Some of the evolutionary algorithms include genetic algorithms the SAA has only one control parameter which makes it very easy to use.
(GA) [28], differential evolution (DE/rand/1) [29], and biogeography- One of the challenges of most recent optimization algorithms is that
based optimization algorithm (BBO) [30]. Some swarm intelligence al­ they cover a certain range of problems and do not have an acceptable
gorithms include ant colony optimization (ACO) [31], spotted hyena solution for many problems. But the SAA has a suitable and acceptable
optimizer (SHO) [32], particle swarm optimization (PSO) [33], Aquila optimal solution for a wide range and variety of problems in different
Optimizer (AO) [34], dwarf mongoose optimization (DMO) [35], multi- fields such as benchmark test functions, engineering applications, en­
ergy systems, and electrical optimization applications.

259
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 1
Benchmark test functions.
Test function Dimensions Range Fmin

The unimodal test functions

∑
f1 (x) = Di=1 x2i 30 [-100, 0
100]
∑D ∏D
f2 (x) = i=1 |xi | + i=1 |xi |
30 [-10, 10] 0
∑D (∑i )2
30 [-100, 0
f3 (x) = i=1 j=1 xj
100]
f4 (x) = maxi {|xi |, 1 ≤ i ≤ D } 30 [-100, 0
100]
∑ 1[ ( 2 )2 ]
30 [-30, 30] 0
f5 (x) = D−
i=1 100 xi − xi+1 +(xi − 1)
∑
f6 (x) = Di=1 ([xi + 0.5] )2 30 [-100, 0
100]
∑D
f7 (x) = i=1 ixi
4
+ random(0, 1) 30 [-1.28, 0
1.28]
∑D ( √̅̅̅̅̅̅̅ )
f8 (x) = i=1 − xi sin |xi | 30 [-500, − 418.9829 ×
500] D
∑n [ ]
f9 (x) = 2
i=1 xi − 10cos(2πxi ) +10 30 [-5.12, 0
5.12]
( √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ )
∑ ( ∑ ) 30 [–32, 32] 0
f10 (x) = − 20exp − 0.2 D− 1 Di=1 x2i − exp D− 1 Di− 1 cos(2πxi ) + 20 + e
1 ∑D 2 ∏D xi 30 [-600, 0
f11 (x) = x − i=1 cos √̅ + 1
4000 i=1 i i 600]
⎧ ⎫
⎪ ⎪ 30 [-50, 50] 0
2
⎪
⎪ 10sin ( π y1 ) ⎪
⎪ ⎧
k(xi − α)m , xi > α,
⎪
⎪ ⎪
⎪
π ⎨ ∑D− 1 ( )2 [ ( )] ⎬ ∑D ⎨
f12 (x) = + i=1 yi − 1 1 + 10sin2 πyi + 1 + i=1 u(xi , α, k, m)u(xi , α, k, m) = 0, − α ≤ xi ≤ α, Where yi =
30 ⎪
⎪ ⎪ ⎩
⎪ ⎪
⎪ k( − xi − α)m , xi < − α.
+(yD − 1)2
⎪
⎪ ⎪
⎪
⎩ ⎭

1 + (xi +1)/4, ɑ=5, k = 100 and m = 4.

⎧ ⎫
⎪
⎪ ⎪
⎪ 30 [-50, 50] 0
⎪
⎪
⎪ sin2 (3πx1 ) ⎪
⎪
⎪
⎪
⎨ ∑D ⎪
⎬
[ ] ∑D
f13 (x) = 0.1 + i=1 (xi − 1)2 1 + sin2 (3πxi + 1) + i=1 u(xi , α, k, m).
⎪
⎪ ⎪
⎪
⎪ [ ] ⎪
+(xD − 1)2 1 + sin2 (2πxD )
⎪
⎪ ⎪
⎪
⎪
⎩ ⎪
⎭
The multimodal fixed-dimension test functions
[ ]− 1
2 [-65, 65] 0.998
1 ∑25 1
f14 (x) = + ∑2
500 j=1
j + i=1 (xi − aij )6

( ) ⎤2
4 [-5, 5] 0.0003
∑11
[ x 1 b2i + bi x2
f15 (x) = i=1 ai − 2 ⎦
bi + bi x3 + x4
1 2 [-5, 5] − 1.0316
f16 (x) = 4x21 − 2.1x41 + x91 + x1 x2 − 4x22 + 4x42
3
)2 )
( 5.1 2 5 ( 1 2 [-5, 5] 0.398
f17 (x) = x2 − x + x1 − 6 + 10 1 − cosx1 + 10
4π2 1 π 8π
[ )]
2(
f18 (x) = 1 +(x1 + x2 + 1) 19 − 14x1 + 3x21 − 14x2 + 6x1 x2 + 3x22 × 2 [-2, 2] 3
[ ( )]
30 +(2x1 − 3x2 )2 × 18 − 32x1 + 12x21 + 48x2 − 36x1 x2 + 27x22
( )
∑4 ∑3 ( )2 3 − 3.86
f19 (x) = − i=1 ci exp − j=1 aij xj − pij [1,3]
( )2 )
∑4 ∑6 ( 6 − 3.32
f20 (x) = − i=1 ci exp − j=1 aij xj − pij [1,3]
[ ]−
∑5 1 4 [0, 10] − 10.1532
f21 (x) = − i=1 (x − ai )(x − ai )T + ci
[ ]−
∑7 1 4 [0, 10] − 10.4028
f22 (x) = − i=1 (x − ai )(x − ai )T + ci
∑10 [ ]− 1 4 [0, 10] − 10.5363
f23 (x) = − i=1 (x − ai )(x − ai )T + ci

260
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 2 simple calculations and fast convergence.

Initial values for tuning and controlling algorithms. The traditional benchmark test functions, which consist of 23 various
Spotted hyena optimizer (SHO) Control parameter h [0 4] functions including unimodal, multimodal, and multimodal functions
[32] Constant M [0.25 0.89] with fixed dimensions, are utilized to show the superiority and optimi­
Co-efficient vector B [0.01 1.85] zation power of the SAA to optimize the functions in the actual world.
Co-efficient vector E [0.01 1.85]
Eight recently published modern standard algorithms, including the
Particle swarm optimization Constriction factor χ 0.80 spotted hyena optimizer (SHO) [32], the particle swarm optimizer (PSO)
(PSO) [33] Acceleration control 2.00 [33], the Aquila Optimizer (AO) [34], the differential evolution (DE/
coefficient c1
Acceleration control 2.00
rand/1) [29], the bat algorithm (BA) [39], the dwarf mongoose opti­
coefficient c2 mization (DMO) [35], the genetic algorithm (GA) [28], artificial bee
Aquila Optimizer (AO) [34] U 0.00566 colony (ABC) [38], and ant colony optimization (ACO) [31] are applied
r 9 to compare the proposed algorithm capability.
ω 0.0025
α 0.05
δ 0.05 3. Snow avalanches algorithm (SAA)
G1 [ − 1 1]
G2 [2 0] 3.1. Inspiration
Differential evolution (DE/rand/ Crossover probability 0.53
1) [29] Scaling factor 0.29
One of nature’s most remarkable phenomena, the avalanche, is the
Bat algorithm [39] Frequency maximum 1.999
Frequency minimum 0.0005 mass of snow that slides down a mountainside [53]. Living things are
A Loudness 0.4991 killed and stuff is buried by large avalanche masses. This is a sudden and
r Pulse rate 0.4990 quick phenomenon when a mass of snow is streaming down the moun­
Dwarf mongoose optimization K-fold cross-validation 10
tainside [53,54]. In altitudes where the duration of cold weather is
(DMO) [35] number
Agent number [1 0 0] longer throughout the year, this flow in the starting area is typically
Genetic algorithm (GA) [28] Crossover factor 0.81 initiated by the mechanical breaking of a slab avalanche when the
Mutation factor 0.62 pressure on the snow exceeds its resistance power [53–56]. Sometimes,
Artificial bee colony (ABC) [38] no 50 % of the this pressure is applied gradually while snow that has partially frozen
colony
expands. Following the first acceleration, the avalanche typically ac­
ne 50 % of the
colony celerates more swiftly as additional snow is immediately incorporated,
ns 1 increasing the avalanche’s size, volume, and mass. If the avalanche
limit ne × D moves swiftly enough to mix the snow with the air, a powder snow
No. of ants used in an 2
avalanche, which is a kind of gravity flow, may form. As it descends the
iteration
Ant colony optimization (ACO) Speed of convergence 0.85
valley, this gravity flow may travel swiftly and even briefly upward
[31] Locality of the search 0.0001 [53–56]. A big quantity of snow migrating down a mountain slope is
process known as an avalanche or snow, one of nature’s most stunning events.
Archive size 50 The SAA is based on the most crucial avalanche elements, including
mountain slopes, local weather, and human aspects, which have been
Unlike many meta-heuristic algorithms, the SAA has a simple taken into account and equated in the presentation of this optimization
formulation and easy implementation to solve classic functions and real- approach. The authors have thus introduced an optimization algorithm
world engineering optimization problems. In addition, unlike some based on this natural phenomenon for the first time since no meta-
metaheuristic algorithms that have complex calculations and slow heuristic approach has before been given based on it.
convergence in solving optimization problems, the proposed SAA has

Table 3
Mean results of the test functions for the different values of si in SAA.
si 0.1 + 0.9*rand 0.2 + 0.8*rand 0.3 + 0.7*rand 0.4 + 0.6*rand 0.5 + 0.5*rand 0.6 + 0.4*rand 0.7 + 0.3*rand 0.8 + 0.2*rand 0.9 + 0.1*rand

F1 4.0241583e-63 2.8237633e-69 3.2840809e-73 4.4338098e-81 1.7166437e-84 1.0101034e-85 9.9380578e-91 1.3158601e-89 6.0019867e-95

F2 2.0781329e-25 5.3001536e-27 8.3080672e-25 1.0643578e-26 6.5218663e-28 9.6427239e-28 7.4009e-26 7.7111293e-30 1.5359254e-24
F3 7.1912586e-07 7.2311e-07 1.0494615e-13 3.2601398e-16 7.7102839e-15 1000 1.9115854e-16 1.336806e-17 500
F4 9.0888367 11.211643 9.8726596 9.7054813 7.9759819 11.238239 11.166152 12.521723 10.420984
F5 1.2030616 0.89394309 1.5990647 2.8077603 1.6588922 1.9936556 0.39882169 1.5947432 1.594712
F6 8.3816471e-32 3.8518599e-32 5.7623824e-32 5.4850485e-32 7.2106817e-32 3.0198582e-32 4.3079201e-31 8.1511826e-29 8.6472714e-30
F7 0.042842807 0.028176643 0.030224987 0.03987721 0.042933824 0.037525555 0.05479652 0.067532529 0.057606275
F8 − 8772.4014 − 8554.8165 − 8376.2003 − 8603.0029 − 8077.8246 − 8930.7502 − 8628.2952 − 8769.7781 − 8262.1573
F9 81.540198 88.750043 73.029774 88.650553 84.272726 87.854579 89.545975 91.038413 96.432465
F10 1.2897057 0.16462236 0.71315362 0.99146788 1.2249757 0.90747416 1.1616812 0.59763877 1.460587
F11 0.021415463 0.026218313 0.034221218 0.10483877 0.054878534 0.028180162 0.26798198 0.12424573 0.059365909
F12 3.4968389 4.3163666 3.6305504 3.6970316 5.0493191 3.8965288 2.2177775 3.6833249 0.76037602
F13 1.1724377 1.8773425 0.91314664 2.3085798 1.9345036 1.9119695 3.0981967 3.2779084 3.4379491
F14 2.6954338 2.4776718 1.7924349 2.3825597 1.0974065 5.3578095 4.3112713 3.6584061 1.3944312
F15 0.00049551812 0.0012016764 0.0013398636 0.0016320125 0.0033901815 0.0024642663 0.004013698 0.0023605883 0.0045505626
F16 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285
F17 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736
F18 3 3 3 3 5.7 3 3 3 3
F19 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821
F20 − 3.2684933 − 3.2506593 − 3.2506593 − 3.2506593 − 3.23877 − 3.2744379 − 3.2863272 − 3.2744379 − 3.2441133
F21 − 7.0198885 − 5.144093 − 4.8977678 − 5.3760878 − 4.6402365 − 7.1381647 − 4.1297549 − 6.3963696 − 5.1245003
F22 − 5.4847937 − 7.2004662 − 6.3866384 − 6.7791094 − 3.9982888 − 6.0616305 − 4.8575422 − 5.6240393 − 7.9165487
F23 − 6.6064046 − 6.5862906 − 4.8743309 − 5.5946857 − 4.6879099 − 4.2859529 − 6.264784 − 5.8444551 − 7.312593

261
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

3.1.1. How to form an avalanche

1.8110625e-106
Avalanches often form during precipitation or after a snowfall, and

0.00030748599
3.1273674e-33
1.1152061e-23

1.6588636e-10
3.0814879e-33

4.4408921e-15

1.7238238e-32
0.9 + 0.1*rand

0.0098572846
0.019085495
all avalanches depend on the three factors of snow, stimulus, and a ramp

− 9461.1179

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
0.40843238
0.99800384

0.39788736

− 10.53641
5.3264695

65.667126

− 10.1532
to form. An avalanche generally occurs when a fresh layer of snow
settles on a steep or moderately steep ramp. The melting of the outer
layer of snow causes the formation of an unstable ice sheet. This ice

3
sheet is buried in a pile of snow as a result of more precipitation. When
this happens, only one stimulus is needed to move the snowpack. This
stimulus can be the movement of a skier on a sloped surface. With this
1.8877563e-102

0.00030748599
0.8 + 0.2*rand

5.7600386e-37
1.3890941e-21

8.5259939e-08

4.4408921e-15
4.5519144e-15
1.7076892e-32
1.3675735e-31
0.0063547742
1.540744e-32

movement, falling snow and ice piles reach a speed of 320 km or more
− 9682.8566

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
0.99800384

0.39788736

− 10.53641
7.3308869

64.672273

− 10.1532
within a few seconds, and drag everything that is in their way down with
them. Avalanche types include slab, powder, shell, and spring snow
avalanches, and avalanches often occur on a slope with an angle of 25 to

3
60 degrees relative to the surface of the earth. In this way, when it snows
in the beds with a slope of more than 45 degrees, a shell avalanche is
formed, and at a slope of fewer than 45 degrees, there is a possibility of a
0.00030748599
7.4397664e-98

2.4229637e-21

1.1307143e-07
1.2325952e-32

4.4408921e-15
2.1094237e-15
1.7560931e-32
0.7 + 0.3*rand

0.020364253

0.010987366
− 9283.5631

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
slab avalanche, which is due to the freezing of the layer and the lack of
0.99800384

0.39788736
3.6911e-37

− 10.53641
6.1082205

57.707535

− 10.1532
adhesion between the new snow and the old layer. A slope of fewer than
25 degrees usually does not cause an avalanche, and there is no possi­
bility of snow accumulation at a slope steeper than 60 degrees, therefore
3

the risk of an avalanche in these cases is zero. Interestingly, in most

deadly avalanches, humans are the driving force and the main cause of a
0.00030748599

disaster. When there is a loose layer in a deep part of the snow mass, it
2.8266164e-92
2.1864133e-37

4.4908439e-08
3.0814879e-33

4.4408921e-15
6.6613381e-16
1.6108814e-32
0.6 + 0.4*rand

0.0086640594
1.050353e-19

0.021023766
− 9567.3526

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
0.99800384

0.39788736

can create a very dangerous slab avalanche. In this type of avalanche,

− 10.53641
5.5452309

75.616707

− 10.1532

the cohesive layer of the snow mass often moves on the snow bed; just
like the snow that slides down the car window. The strength of the slab
3

avalanche depends on factors such as the depth of the loose layer and the
characteristics of the sheets. Each avalanche consists of a starting point,
an avalanche path, and an end part. The starting point is where the snow
0.00030748599

layer separates from the snow below. The path of the avalanche is the
1.4905661e-92
5.8732976e-36

6.6420241e-05
9.2444637e-33

4.4408921e-15
3.1086245e-15
0.5 + 0.5*rand

1.739026e-19

0.010337352

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
0.41467193
0.04394886
0.99800384

0.39788736

distance that the mass of snow moves toward the bottom of the moun­
− 9265.313

− 10.53641
3.7971673

47.757959

− 10.1532

tain, and the end part is called the place where the avalanche stops
[53–56].
3

3.1.2. Causes of avalanches

The most important avalanche factors, which have been considered
0.000307485988
7.81200628e-85
9.73412078e-35

5.66273387e-07
9.24446373e-33

1.71575647e-32

and equated in the presentation of this optimization method, include

4.4408921e-15
8.8817842e-16
0.4 + 0.6*rand

0.0141462642

0.0109873658
3.804693e-20

− 10036.6986

− 1.03162845

− 3.86278215
− 3.32199517
− 10.1531997
− 10.4029406
− 10.5364098
0.998003838

0.397887358

mountain slopes, regional weather, and human factors [53–56].

6.33345861

55.7176114

The slope of the mountain: The slope of the mountain is one of the
factors influencing the occurrence of avalanches, and according to sta­
tistics, most avalanches occur on convex slopes between 30 and 45 de­
3

grees. The snow pile is more unstable on convex slopes, rocky strips,
windy slopes, etc. In some places, such as steep and narrow corridors,
0.000307485988

the snow accumulates like a bowl and becomes a trap for climbers.
Best results of the classic test functions for the different values of si in SAA.

1.88135904e-82
2.16000556e-33
5.92723967e-17

9.29741134e-07

1.77222768e-32
4.4408921e-15
4.4408921e-16
0.3 + 0.7*rand

0.0173992846

0.0109873658
− 10080.7455

− 1.03162845

− 3.86278215
− 3.32199517
− 10.1531997
− 10.4029406
− 10.5364098

The weather of the area: fresh snow increases the possibility of

0.998003838

0.397887358
4.77418254

43.7781229

avalanches, and subsequently, the wind increases this risk by accumu­

lating a large amount of snow on a slope. The most dangerous time for an
avalanche is often the first sunny day after snowfall; because the high
0

temperature of the air prevents the consolidation of fresh snow and

therefore the occurrence of avalanches becomes more likely. If the
amount of fresh snow is two centimeters or more, or if the snow is falling
0.00030748599
0.2 + 0.8*rand

1.4183921e-75
1.3445847e-31
4.9431794e-16

5.0292926e-06

4.4408921e-15

1.7157565e-32
1.8428219e-32
0.0054846271

1.110223e-15

more than two centimeters per hour, the probability of an avalanche

− 9265.3432

− 1.0316285

− 3.8627821
− 3.3219952

− 10.402941
0.99800384

0.39788736

− 10.53641
6.0972214

− 10.1532
51.73777

increases. Rain may also act as a lubricant or thickener for snow.

Human factors: Many avalanches are caused by the mistakes of
mountaineers and skiers, which depend on several reasons, including
0

the way of decision-making, wrong attitude, improper training, lack of

awareness, choosing the wrong path, ignoring the warning signs of av­
alanches, etc. (see Fig. 1).
0.000326143234

0.000307485988
9.47725492e-68

7.05265342e-16

2.22044605e-16

2.58237897e-32
9.3805805e-33

4.4408921e-15

1.6350833e-32
0.1 + 0.9*rand

0.0187011925
− 10357.0934

− 1.03162845

− 3.86278215
− 3.32199517
− 10.1531997
− 10.4029406
− 10.5364098
0.998003838

0.397887358
58.7025037
3.0682627

3.2. SAA mathematical model

In this research, based on the phenomenon of avalanches, a new and

0

effective MHA called the snow avalanches algorithm (SAA) has been
designed. In this optimization algorithm, the downward movement of
Table 4

the avalanche is actually assumed and equated to the movement of the

F10
F11
F12
F13
F14
F15
F16
F17
F18
F19
F20
F21
F22
F23
F1
F2
F3
F4
F5
F6
F7
F8
F9
si

population towards better and more effective solutions. The proposed

262
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 5
Standard deviation results of the classic test functions for the different values of si in SAA.
si 0.1 + 0.9*rand 0.2 + 0.8*rand 0.3 + 0.7*rand 0.4 + 0.6*rand 0.5 + 0.5*rand 0.6 + 0.4*rand 0.7 + 0.3*rand 0.8 + 0.2*rand 0.9 + 0.1*rand

F1 8.412462e-63 8.7780263e-69 1.025992e-72 9.377087e-81 2.834257e-84 3.1893409e-85 2.545495e-90 4.160676e-89 1.644407e-94

F2 7.8276e-25 1.0797001e-26 2.603933e-24 3.197882e-26 1.980768e-27 3.045579e-27 2.5505e-25 1.55628e-29 4.8542301e-24
F3 3.215163e-06 2.8006e-06 2.40749e-13 7.183013e-16 2.139635e-14 2108.1851 5.854142e-16 3.310333e-17 1581.1388
F4 3.73262 4.1722157 3.393416 3.185273 2.444025 3.8256168 3.983356 3.448309 4.6897681
F5 1.779485 1.6884601 2.062256 1.900718 2.008492 2.1014329 1.260645 2.058617 2.0587184
F6 1.825789e-31 3.7523082e-32 7.228263e-32 6.203586e-32 4.833562e-32 2.0839001e-32 7.489612e-31 2.574166e-28 2.6812346e-29
F7 0.02700681 0.015130051 0.009275944 0.02950269 0.02872885 0.018209074 0.03344369 0.04354878 0.060477006
F8 860.6193 481.71192 835.005 732.725 756.6707 360.04399 468.245 767.3074 810.06345
F9 20.08439 24.284493 18.04496 28.44267 29.51521 7.4462674 20.12982 21.94035 27.948984
F10 1.749757 0.52058162 0.9649275 1.3996 0.9352481 0.85348329 1.144048 0.9752241 1.0954511
F11 0.0317635 0.027493352 0.0464512 0.2132045 0.03432826 0.027582024 0.461566 0.1946575 0.061091749
F12 4.844591 7.4731264 3.391325 4.135145 5.825254 4.4433878 2.302488 3.160396 0.89640961
F13 0.9382149 2.336711 1.319794 3.287721 1.089317 2.8602251 7.261957 4.848601 5.0579766
F14 4.481863 2.330487 0.7827635 1.81435 0.314339 6.685823 5.224365 2.810576 0.95678317
F15 0.0004969337 0.002522408 0.002486774 0.003446386 0.005018872 0.0034930703 0.005870641 0.006327284 0.0083462892
F16 2.27813e-16 0 0 0 0 0 0 0 0
F17 0 0 0 0 0 0 0 0 0
F18 5.940635e-16 2.9605947e-16 0 2.960595e-16 8.53815 7.4014868e-16 4.440892e-16 7.401487e-16 4.4408921e-16
F19 2.27813e-15 9.3622226e-16 9.362223e-16 9.362223e-16 9.362223e-16 9.3622226e-16 9.362223e-16 9.362223e-16 9.3622226e-16
F20 0.06068516 0.061396144 0.06139614 0.06139614 0.05743083 0.061396144 0.05743083 0.06139614 0.069927067
F21 3.333227 2.8755767 3.626653 2.750618 3.069224 3.3010792 2.408376 3.380685 2.8857877
F22 3.470591 3.536793 3.588146 3.861214 2.492016 3.8560678 3.072911 3.421648 3.2785473
F23 3.692028 3.4843785 3.951659 3.505479 3.201939 3.3375097 3.762842 3.392614 3.4832922

Table 6
Friedman’s test rank of statistical assessment results for the different values of si in SAA.
si 0.2 + 0.8*rand 0.3 + 0.7*rand 0.4 + 0.6*rand 0.5 + 0.5*rand 0.6 + 0.4*rand 0.7 + 0.3*rand 0.8 + 0.2*rand

F1 8 7 6 5 4 2 3
F2 4 8 5 2 3 6 1
F3 7 5 3 4 9 2 1
F4 7 4 3 1 8 6 9
F5 2 6 9 7 8 1 4.5
F6 2 4 3 5 1 7 9
F7 1 2 4 6 3 7 9
F8 6 7 5 9 1 4 2
F9 6 1 5 3 4 7 8
F10 1 3 5 7 4 6 2
F11 2 4 7 5 3 9 8
F12 8 4 6 9 7 2 5
F13 3 1 6 5 4 7 8
F14 5 3 4 1 9 8 7
F15 2 3 4 7 6 8 5
F16 5 5 5 5 5 5 5
F17 5 5 5 5 5 5 5
F18 4.5 4.5 4.5 9 4.5 4.5 4.5
F19 5 5 5 5 5 5 5
F20 6 6 6 9 2.5 1 2.5
F21 5 7 4 8 1 9 3
F22 2 4 3 9 5 8 6
F23 3 7 6 8 9 4 5
MFr 4.3261 4.5870 4.9348 5.8261 4.8261 5.3696 5.1087
Total rank 1 2 4 7 3 6 5

algorithm steps are formulated in the following. In this study, the equal to rand(1, D) × (Xr1 − Xr2 ) in its new position. Therefore, the
maximum chance of avalanche occurrence is assumed and equated. The equation of this phase for the ith member is modeled in the following
SAA, like any other algorithm, consists of the number of population NP form:
(snow piles), whose initial position is produced between the lower limit
Xinew = XBest + rand(1, D) × (Xr1 − Xr2 ) (1)
of Xmin and the higher limit of Xmax, randomly. The number of human NP
in the problem areas is climbing, and then the following phases are Where, D represents dimensions number of the problem, rand de­
applied to them. notes a number between 0 and 1, randomly, Xr1 and Xr2 are two mem­
bers are considered from the population, randomly.
3.2.1. First phase, avalanche due to mountain slope
In this phase, it is assumed that the best and strongest group member 3.2.2. Second phase, the avalanche caused by human factors
played the role of the slope of the mountain that caused the avalanche At this stage, the rrd
3 member, who is chosen randomly, plays the role
event and as a result of the movement of the snow masses by the of a human that causes the avalanche phenomenon and as a result the
avalanche in this direction and to the serious position of the pointer (or optimal movement of snow masses. As in the previous phase, each
in the science of optimization, it means a better and optimal position) member will have a deviation equal to rand(1, D) × (Xr1 − Xr2 ) in its new
will be achieved. Of course, each member will have a deviation value position. So, the equation of this phase for the ith member is modeled by:

263
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 3. 3-D maps of 2-D unimodal test functions.

264
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 4. 3-D maps of 2-D multimodal test functions.

265
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 5. 3-D maps of 2-D multimodal fixed-dimension test functions.

266
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 5. (continued).

Xinew = Xr3 + rand(1, D) × (Xr1 − Xr2 ) (2)

3.2.4. Fourth phase, normal conditions
In the SAA, there is a very small probability that an avalanche will
3.2.3. Third phase, the avalanche due to the weather in the region
not occur and snow masses for other reasons such as rain or wind have
It has been observed that avalanches have occurred many times due
an effect in changing positions as much as rand(1, D) × (Xmax − Xmin ).
to heavy snowfall or wind. In this phase, the ith member plays the role of
the weather in the region and causes the snow masses to fall in the di­ Xinew = Xi + rand(1, D) × (Xmax − Xmin ) (4)
rection with the amount of position deviation equal to rand(1, D) ×
(Xr1 − Xr2 ). Therefore, we have: 3.3. Probability of avalanche occurrence for the ith mass of snow (Xi)
Xinew = Xi + rand(1, D) × (Xr1 − Xr2 ) (3)
In the proposed SAA, the probability of the avalanche phenomenon

267
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 6. Comparison of convergence process of algorithms in solving classical test functions.

occurrence (si) is considered higher than normal conditions, in which (continued )

case one of the phases 1 to 3 is executed randomly, and if the avalanche 16:Xnew
i = Xi +rand(1, D) × (Xmax − Xmin )
phenomenon does not occur, phase 4, i.e. Eq. (4), is applied under 17: end if
normal conditions. Algorithm 1 gives general pseudo-code of the SAA 18: if f(Xnew
i ) < f(Xi )
19: Xi = Xnew and f(Xi ) = f(Xnew
and also the SAA flowchart is shown in Fig. 2. i i );
20: end if
Algorithm 1: The proposed SAA 21: if f(Xi ) < f(XBest )
1: adjusting the SAA parameters: maximum iteration number Itermax, and size of 22: XBest = Xi and f(XBest ) = f(Xi );
population NP, si and to set the number of iterations Iter = 0; 19: end if
2: generation of the random initial population NP (i = 1, 2, …, NP); 20: end for
3:Xi = Xmin +rand × (Xmax − Xmin ) 21: end while
4: to calculate of each individual fitness; Return the best solution obtained by the SAA: XBest
5: while the i till upper limit of the iterations Itermax do
6: Setting the number of iterations Iter = Iter + 1;
7: for i = 1 to NP do 3.4. Computational complexity of SAA
8: selecting the members XBest , Xr1 ,Xr2 ,Xr3 ;
9: if rand < si
10: Xnew
i = XBest + rand(1, D) × (Xr1 − Xr2 ); Keep in mind that the algorithm’s computing complexity (CC) is
11: elseif rand < si primarily determined by three steps: startup, fitness assessment, and
12: Xi new
= Xr3 + rand(1, D) × (Xr1 − Xr2 ); population update. Be aware that the initialization method’s computa­
13: elseif rand < si tional complexity is O for NP persons O(NP). The CC of the updating
14: Xnew = Xi + rand(1, D) × (Xr1 − Xr2 );
i
15: else
structure is O(Itermax × NP) + O(Itermax × NP × D), which is created to
search the optimal position and the position vector update related to all
(continued on next column)
population. Where, Itermax refers to the maximum iterations number and

268
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

D clears the problem dimension. So, the CC of SAA is as O(NP ×(Itermax

2.8237633e-69
5.3001536e-27

3.8518599e-32
+ Itermax × D + 1).

0.0012016764
0.028176643

0.026218313
− 8554.8165

− 1.0316285

− 3.8627821
− 3.2506593

− 7.2004662
− 6.5862906
0.89394309

0.16462236

0.39788736
7.2311e-07

− 5.144093
11.211643

88.750043

4.3163666
1.8773425
2.4776718
4. Simulation results
SAA

3
The efficacy of the suggested algorithm is evaluated in addressing six
technical issues as well as six traditional test functions listed in Table 1
[57]. Nine different optimization methods including SHO, PSO, AO, DE,
0.0014809686

0.0027989789
5.082321e-06
0.72023577

0.39788736
BA, DMO, GA, ABC, and ACO are applied to compare the SAA. Different
2.1669609

8.8093984
4.5413034
5.8088019

8.0801248

− 1.0316285

− 3.8627821
− 3.2625486
− 5.4023057
− 5.8157199
− 5.8550347
41.596817
94.123452

94.819302

14.371694
indices such as “Best,” “Mean,” “Worst,” and “Std.” related to the
1309.6291

− 7894.6683

objective function are shown for each evaluation. Table 2 displays the

11.1
settings for these algorithms’ control parameters. The number of pop­
ACO

ulation, the maximum number of repetitions, and the number of inde­

pendent runs for each algorithm are considered to be 1500, 60, and 30,
respectively. The value of these general parameters is selected based on
6.9046069e-17
8.2288606e-06

1.2908634e-13

0.0023274009
0.076714454

trial and error method and user experience for multiple runs. The value
− 8468.3728

− 1.0316285

− 3.8627821
− 3.2506593
− 5.6500406
− 7.1500516
− 5.7730884
0.28148318

0.39788736
34.214637
33.975316

5.6481482

3.5844921
4.4816744
81.68594

4.222927

of the control parameters of each algorithm is also selected based on the

1002.24

reference paper of that algorithm.

ABC

5.7

4.1. Results considering avalanche phenomenon occurrence probability

0.00036279729
1.2946248e-12

− 0.95001199

Tables 3-5 display the simulation results for the algorithm’s evalu­
0.010338602
− 8141.7555

− 3.2478919
− 3.6441046
− 5.0588409
0.95945237

0.39788736

− 3.971655
1.0060547
1003.3471
37.020964
48.056037

1.3167364

106.02037
5.5612853

5.3975778
4.7945703
5.1614142

ation to solve the unimodal, multimodal, and multimodal functions with

− 3.82797

fixed dimensions while accounting for the probability of avalanche

11.1
GA

occurrence (si). As can be observed, when the probability of the

avalanche phenomenon occurring is considered, the test functions have
better values. The optimal solution obtained from the algorithm has the
7.9453865e-20

4.3972616e-18

0.0037099917

highest avalanche occurrence probability for all test functions, accord­

0.034155784

0.035095782
− 5506.3407

− 1.0316284

− 3.8627821
− 3.2338132
− 6.4405312
− 7.1897755
0.01713567

12.1197906

0.99800709

0.39791449

− 4.516861
20.636399

28.740923

116.05761
3.3052636

5.9997228
7.2497319

ing to the results of the Friedman test in Table 6, which evaluates the
algorithm’s average performance for subsequent statistical comparison.
DMO

The results show that, with a probability of 80 % (see Table 5), the al­
3

gorithm performed better when solving test functions and obtaining

better values. The 3-D maps for the 2-D unimodal, multimodal, and
8.6818092e-17

1.1771424e-14

0.0030776835

multimodal fixed-dimension benchmark functions are shown in Figs. 3-

0.043318071

0.058706219

0.018063918
− 5491.3762

− 1.0316283

− 3.8627821
− 3.1929895
− 4.9012882
− 6.0691774
− 5.6315077
17.7002593

4.75424132

0.99800904

0.39788936
108.55563

58.098861

130.51756
4.2990418

38.783869

5, respectively. Also, in order to achieve a balance between the explo­

ration and exploitation phases, randomization is utilized in this paper to
explore various options with several probabilities and exploit the best
BA

option with the residual probabilities.

1.0242218e-16

4.7455445e-17

0.0098930146

4.2. Results in solving classical test functions

0.075715507

0.041272316
− 5372.3933

− 1.0316285

− 3.8627821
− 3.2413341
− 4.7186159
− 5.7794165
12.1307513

0.03788369

0.39790643

− 5.311493
77.793502

52.309303

96.464196
3.1848214

5.8420939
7.3495953
1.163675

This section compares the capability of the SAA method with 10

optimization methods, including SHO, PSO, AO, DE, BA, DMO, GA, ABC,
DE

and ACO in solving 23 classical test functions. The values for “Best,”
“Mean,” “Worst,” and “Standard” are shown for each assessment.
Mean results of different algorithms to solve the classic test functions.

5.4100295e-20

8.4329049e-20

0.0093004408

Table 2 displays the settings for these algorithms’ major parameters. The
0.041373554

0.051227924
− 5266.9035

− 1.0316284

− 3.8627821
− 3.2467352
− 4.5380137
− 6.2593125
11.6614878

4.72522565

0.99805141

0.39788745

− 5.406272
0.2346933
25.925291

14.076205

117.19386
3.8746074

24.785325

SAA is thought to have a population of 60 and 1500 iterations, respec­

tively, with 30 independent runs being taken into consideration for the
population size of all methods. Wilcoxon’s test [36] is also applied for
AO

comparison the behavior of methods while solving test functions,

moreover to Friedman’s test, which is used to assess the average per­
2.8875324e-14

3.9030829e-10

0.0021092882

formance of various algorithms and ranking. Fig. 6 compares the

0.043895494

0.033946413
− 5491.6283

− 1.0316284

− 3.8627821
− 3.2031021
− 5.0631911
− 5.2225618
− 5.8099071
0.33171097

0.99800384

convergence process of solving certain traditional test functions gener­

71.938327
11.236808
22.443515

113.84734

5.0328831

0.3979944
50.33473
4.12442

ated using the technique and other algorithms. The suggested method,
PSO

as is evident from these data, has a competitive and superior conver­

3

gence feature to compare the other methods in reaching the optimal

solution with a quicker convergence rate. Additionally, the indices of
1.7402073e-22

9.6113873e-22

“Mean,” “Worst,” and “Standard” from the SAA are compared with SHO,
0.0039580566

0.0043532998
0.032773822
− 5476.9598

− 1.0316285

− 3.8627821
− 3.1913759
− 5.7037758
− 6.7002367
12.5330143

0.15502593

0.99800384

− 5.883441

PSO, AO, DE, BA, DMO, and GA in Tables 7, 10-11, respectively. Tables 8
11.519564

38.057048

123.19367
3.0409572

4.8260723
18.191523

0.3979058

and 9 respectively include the Friedman test results, which evaluates the
SHO

average capability of several algorithms, and the Wilcoxon test. Based

3

on the findings, it is evident that the algorithm successfully solved the

test functions, earning a score of 1 and superior Mean and Best values. In
Function

solving all of the test functions, the SAA has produced the objective
Table 7

function’s best value. The outcomes of numerous tests demonstrate that

F10
F11
F12
F13
F14
F15
F16
F17
F18
F19
F20
F21
F22
F23
F1
F2
F3
F4
F5
F6
F7
F8
F9

the suggested algorithm outperforms the other approaches and has done

269
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 8
Friedman’s test rank summary of statistical analysis results for different optimization algorithms.
Function SHO PSO AO DE BA DMO GA ABC ACO SAA

F1 2 7 3 6 5 4 8 5 10 1
F2 2 7 6 5 4 3 8 2 10 1
F3 2 5 4 6 7 3 8 2 4 1
F4 6 2 3 5 7 4 8 8 10 1
F5 5 3 2 7 8 4 6 5 10 1
F6 2 7 3 5 6 4 8 7 9 1
F7 2 6 5 4 7 3 8 8 9 1
F8 6 4 8 7 5 3 2 2 4 1
F9 7 4 6 2 8 5 3 1 3 1
F10 2 6 5 3 7 4 8 9 10 1
F11 7 3 6 5 1 4 8 7 10 2
F12 4 5 2 7 3 8 6 1 8 1
F13 5 8 6 4 7 3 2 2 6 1
F14 1.5 1.5 5 6 4 3 8 8 10 7
F15 5 2 6 7 3 4 8 3 4 1
F16 2 5 5 2 7 5 8 3 3 2
F17 5 8 3 6 4 7 1.5 2.5 2.5 1.5
F18 4 4 4 4 4 4 8 8 9.5 4
F19 4 4 4 4 4 4 8 5 5 4
F20 8 6 3 4 7 5 2 2.5 1 1
F21 2 4 8 6 5 1 7 3 4 3
F22 3 7 4 6 5 2 8 3 7 1
F23 2 3 5 6 4 7 8 5 3 1
MFr 4.7174 5.8913 5.6087 6.2174 6.4348 5.0000 8.0000 4.4348 6.6087 2.0870
Total rank 3 6 5 7 8 4 10 2 9 1

Table 9
Wilcoxon’s test results between the SAA and other algorithms.
i j MoNR MoPR SNR SPR F(i) < F(j) F(j) < F(i) p-value 0.95 Confidence interval
SAA SHO 10.000000 15.000000 180 30 18 2 0.0054144367 − 17.221819497 − 0.0296326110

PSO 10.850000 14.000000 217 14 20 1 0.0004472267 − 24.228659169 − 0.0404294681

AO 10.800000 15.000000 216 15 20 1 0.0005093984 − 12.964550894 − 0.1298641635
DE 10.421053 12.000000 198 12 19 1 0.0005538304 − 25.712105582 − 0.2127927083
BA 11.105263 10.000000 211 20 19 2 0.0009600603 − 28.602543204 − 0.0577520669
DMO 10.684211 14.000000 203 28 19 2 0.0024952608 − 13.004344006 − 0.0066023306
GA 11.523810 11.000000 242 11 21 1 0.0001887999 − 17.270275892 − 1.0708473620
ABC 10.732518 14.000000 214 16 20 1 0.0006729843 –22.752101384 − 0.06821400325
ACO 11.394952 11.000000 240 21 19 1 0.0020882465 − 16.821354732 − 1.18392136282

so. Among the reasons for the superiority of the SAA in solving classical This approach is easy to implement with only one adjustable
functions, we can mention the establishment of a favorable balance parameter, which in turn makes it very potential for many applications
between the exploration and exploitation phases, which has made the in optimization fields. As shown in Fig. 7, the accuracy and high speed of
SAA, has a favorable convergence speed. Also, the adjustment value of convergence has increased the capability of this proposed algorithm due
avalanche probability is effective in the performance of the algorithm, to the appropriate balance between exploration and exploitation phases.
which is carefully determined in section 4.1. Additionally, the convergence of several algorithms considering the
i,j: algorithm index corresponding to rows of Table 2, MoNR: mean of best fitness value for the functions F3, and F10 according to Fig. 7,
negative rank; MoPR: mean of positive rank; SNR: sum of negative rank; produced results that are significantly better and more effective than
SPR: sum of positive rank; F(.): set of objective functions. those of the other algorithms, demonstrating the SAA algorithm’s effi­
cacy as a potent new emerging algorithm. Additionally, the hybrid child
4.3. Results in solving CEC-2019 test functions drawing development optimization - harmony search (CDDO-HS) [58]
and the SAA results are compared, with the results showing that the SAA
The effectiveness of the CVSO in solving the CEC-2019 test functions performs better overall in terms of obtaining better mean and Std values.
[58] is contrasted with the SHO, AO, DE, BA, DMO, GA, ACO, and ABC
algorithms, in order to more thoroughly evaluate the SAA’s abilities. The 4.4. Results to solve the engineering optimization problems
test features for CEC-2019 are listed in Table 12. Every algorithm’s test
function is tested in a system with 30 unique runs, and all simulations’ This section contrasts the capability of the suggested approach with a
evolutions and dimensions are multiplied by 30,000 and 30, respec­ number of challenging, real-world engineering problems. The SAA is
tively, for all tests. Table 13 displays the optimization results for several thought to have a population of 60 and 2000 iterations, respectively,
approaches based on the mean and standard deviation (Std) values for with 30 independent runs being taken into consideration for the popu­
each algorithm and each function. It is clear that compared to all other lation size of all methods. Each algorithm’s parameter setting is deter­
algorithms, the suggested SAA approach has been able to produce better mined by the reference research for that particular approach.
and optimal results for the mean criteria as well as fewer Std values. The
SHO, AO, DE, BA, DMO, GA, ACO, and ABC algorithms are utilized in 4.4.1. Pressure vessel design (PVD)
practice in the comparative study under equal and fair settings to In this instance, shaping and welding the tank in accordance with
confirm the SAA’s capacity to optimize the real test functions of CEC- Fig. 8 should reduce the tank’s overall cost under pressured circum­
2019. stances. Four variables and four restrictions are taken into account while

270
 Table 10
Best results of different algorithms to solve the classic test functions.
Function SHO PSO AO DE BA DMO GA ABC ACO SAA
K. Golalipour et al.

F1 9.2789564e-25 6.7230304e-16 3.2406472e-23 1.7763908e-18 6.6457868e-18 4.0133237e-21 1.4223538e-20 1.3387747e-20 2.2142207e-09 1.4183921e-75
F2 0.00015452127 0.0030190198 0.0001558359 0.00017766765 0.00077010509 0.00029214464 1.7981949e-08 2.6976528e-11 4.2143367e-05 1.3445847e-31
F3 1.0562671 2.9934449 2.4455103 6.7225332 5.0578735 0.096944921 0.022824405 0.0026139071 5.9290214 4.9431794e-16
F4 6.4884198 7.3673913 7.40033041 6.2312406 7.6691813 8.57056545 28.413439 8.372049 14.819092 6.0972214
F5 17.211125 20.388094 2.2061381 16.62602 12.19319 8.7446656 0.48481083 1.9409694 2.365643 5.0292926e-06
F6 7.9547176e-24 1.2302729e-14 9.3751885e-24 3.4992849e-19 1.4306017e-17 2.4943187e-20 7.7123259e-23 8.2659831e-22 1.662719e-09 0.0
F7 0.015386457 0.027737435 0.0078750151 0.016321635 0.015271406 0.0099022533 0.25542871 0.075485994 0.2153694 0.0054846271
F8 − 6694.1135 − 6243.2892 − 6038.3967 − 6052.9424 − 6137.4876 − 5692.8214 − 9173.6827 − 9169.9689 − 9092.1039 − 9265.3432
F9 78.157727 53.068493 101.80824 41.788225 62.78323 98.725684 53.727678 60.692422 66.66207 51.73777
F10 7.9936058e-15 4.4408921e-15 4.4408921e-15 4.4408921e-15 1.5099033e-14 4.4408921e-15 4.4408921e-15 4.4408921e-15 8.274128e-14 4.4408921e-15
F11 0.022126733 5.2180482e-15 0 0.017226294 2.4424907e-15 0.012316073 0.024644477 0.014772408 0.029477033 1.110223e-15
F12 2.5700982e-07 4.4895779e-06 5.0132664e-08 1.8708246 7.2358995e-11 2.4410252e-11 3.4664471e-15 3.1082561e-14 5.398864e-05 1.7157565e-32
F13 0.93455562 39.58436 11.048818 1.5360417 8.2802542 1.3400603 0.044852608 4.678859e-09 0.97627538 1.8428219e-32
F14 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384 0.99800384
F15 0.00030748599 0.00052818331 0.00030748599 0.00057138126 0.00070582442 0.00046339943 0.00030748599 0.00030748599 0.00030748599 0.00030748599
F16 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285 − 1.0316285
F17 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736 0.39788736
F18 3 3 3 3 3 3 3 3 3 3
F19 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821 − 3.8627821
F20 − 3.2031021 − 3.2031021 − 3.3219952 − 3.3219952 − 3.3219952 − 3.3219952 − 3.3219952 − 3.3219952 − 3.3219952 − 3.3219952
F21 − 10.152826 − 10.1532 − 10.153147 − 10.043175 − 10.144305 − 10.152318 − 10.1532 − 10.1532 − 10.1532 − 10.1532
F22 − 10.402933 − 10.402941 − 10.402941 − 10.401367 − 10.402941 − 10.402593 − 10.402941 − 10.402941 − 10.402941 − 10.402941
F23 − 10.527645 − 10.53641 − 10.53641 − 10.53641 − 10.53641 − 10.536407 − 10.53641 − 10.53641 − 9.739025 − 10.53641

271
Table 11
Std. results of different algorithms to solve the classic test functions.
Function SHO PSO AO DE BA DMO GA ABC ACO SAA

F1 3.403754e-22 4.086452e-14 9.916773e-20 1.197672e-16 1.260657e-16 9.680817e-20 4.0743456e-12 2.1556712e-16 0.0046024132 8.7780263e-69
F2 0.005877851 0.5077632 0.3913052 0.178962 0.03332244 0.03196845 3.1601895 2.1530261e-05 4.1416652 1.0797001e-26
F3 11.60801 89.74446 27.08024 65.65971 206.7506 28.00488 2107.694 2107.3287 2103.9485 2.8006e-06
F4 0.8855786 3.915934 1.545157 0.7010726 1.859723 0.6467166 6.1713221 7.9431895 7.3612902 4.1722157
F5 29.33797 2.120632 8.046999 53.88624 42.9426 26.53486 37.72791 36.001708 54.105401 1.6884601
F6 2.115836e-21 9.490113e-10 1.922279e-19 8.529928e-17 2.639897e-14 7.472021e-18 0.0011472656 4.0818963e-13 8.8196957e-06 3.7523082e-32
F7 0.005615926 0.02159741 0.008860818 0.009826857 0.03621759 0.01021102 0.95874437 0.38118993 0.41846262 0.015130051
F8 551.411 599.0415 431.1943 491.4598 495.0792 201.7143 801.71877 527.82314 661.0614 481.71192
F9 25.38873 36.14927 14.01416 31.20735 47.83057 18.93002 44.850078 18.180128 25.758205 24.284493
F10 1.739309 2.252669 2.026436 1.043552 3.738179 0.8644876 2.2947361 3.7669118 5.0631008 0.52058162
F11 0.1730395 0.04371375 0.09315479 0.02188505 0.01887274 0.02165022 1.2667801 0.078018317 13.283523 0.027493352
F12 2.353687 1.424068 0.752669 1.89649 0.9073022 1.597708 6.1681947 3.2935684 5.3685491 7.4731264
F13 17.58387 7.552207 15.69328 9.604577 26.43887 8.307879 5.3168551 5.1818682 8.8235336 2.336711
F14 5.008043e-12 2.125248e-10 0.000116523 0.4058098 1.273735e-05 7.977879e-06 6.6626052 4.6853027 6.7492874 2.330487
F15 0.001136228 0.002651459 0.0005193407 0.0004449792 0.001885686 0.001641666 0.018150739 0.0032371872 0.0047618991 0.002522408
F16 1.523922e-09 2.389538e-08 2.761721e-08 3.783841e-13 2.72301e-07 5.416477e-08 0.25809392 0 0 0
F17 4.516555e-05 0.0002162162 2.184934e-07 4.01636e-05 4.916031e-06 6.637402e-05 0 0 0 0
F18 0 4.440892e-16 5.958082e-16 0 0 3.972055e-16 25.614449 8.5381497 13.042239 2.9605947e-16
F19 0 1.986027e-16 0 0 0 1.986027e-16 0.11008564 9.0042974e-16 9.3622226e-16 9.3622226e-16
F20 0.01302924 4.864754e-16 0.07187764 0.07735831 0.06019507 0.06895433 0.072358111 0.061396144 0.062662177 0.061396144
F21 3.653769 3.471244 1.046874 2.881157 3.227247 2.617945 3.7707915 3.8757547 3.4182423 2.8755767
F22 3.985787 2.697263 3.412068 3.677529 2.640578 3.351733 3.7410418 3.5197825 3.2639742 3.536793
F23 2.755275 2.673827 0.1696755 2.689681 3.734032 0.03127776 2.4543478 3.4114736 4.0512944 3.4843785
Alexandria Engineering Journal 83 (2023) 257–285
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 12
CEC-2019 Test Functions [58]

5.317e + 04
8.142e + 03
1.835e + 01

1.370e + 01

5.746e + 01
3.066e + 01
2.170e + 00

1.130e + 01
1.591e + 00
5.440e + 01
1.300e + 02
3.150e + 00

3.484e + 01

1.554e + 01
7.609e + 00
6.390e-03

5.785e-13

2.109e-01

7.640e-01

1.912e-01
CDDO-HS
Function Function Name Dimension Range

[58]
F1 STORN’S CHEBYSHEV 9 [-8192,8192]
POLYNOMIAL FITTING
F2 INVERSE HILBERT MATRIX 16 [-16384,16384]
F3 LENNARD-JONES MINIMUM 18 [-4,4]
ENERGY CLUSTER

45845.2152
8049.4183

2.266e-10
0.005038
F4 RASTRIGIN 10 [-100,100]

16.7429

13.1372

53.1832
25.1755
2.08933

55.2409
42.0437

15.2786
3.79223
0.1606
9.1014
0.4792

2.7268
0.5104
3.1567
0.1692
F5 GRIENWANK 10 [-100,100]

SAA
F6 WEIERSRASS 10 [-100,100]
F7 MODIFIED SCHWEFEL 10 [-100,100]
F8 EXPANDED SCHAFFER’S F6 10 [-100,100]
F9 HAPPY CAT 10 [-100,100]
F10 ACKLEY 10 [-100,100]

973655.6006
237754.0304

143.01762
177.2966
25.8874

30.5674

71.9835
27.2375

18.4573

40.3793
13.5510
0.2398

0.8873

3.4059
2.3358

2.8051

9.4012
1.1579
9.6533
1.7044
designing in this situation. The vessel is designed to have an ideal ca­

ABC
pacity of 750 ft3 and an acting pressure of 3000 psi [59]. The vessel
features hemispherical heads that cap each end, as seen in Fig. 8. Control
parameters {x1(Ts), x2(Th), x3(R), x4(L)} denote thickness of the shell

892371.0932
572880.2311
(Ts), the spherical head thickness (Th), inner radius (R) and the length of

155.6082
128.2047
the cylindrical section without a head (L), respectively [60]. The issue is

22.2833

30.8620

76.1298
44.0087

15.9377
2.39484

35.5335
13.2351
0.2492

1.3877

5.0564
1.9826

8.0395
1.5483
9.2077
1.6836
described numerically by [61].

ACO
Minimize:

F(X) = 0.6224x1 x3 x4 + 1.7781x2 x23 + 3.1661x12 x4 + 19.84x12 x3

8809226.6489
4753682.8353
Subject to:

168.6018
123.1209
21.5639

38.7497

73.3342
48.1348

15.6803

15.5472

10.8149

37.0932
19.2755
g1 (X) = − x1 + 0.0193x3 ≤ 0

0.4104

1.9664

6.7735
2.0935

2.2734

1.7356

2.5246
GA
g2 (X) = − x2 + 0.00954x3 ≤ 0 (5)

4 3
g3 (X) = − πx23 x4 − πx + 1, 296, 000 ≤ 0
87842.5433
32755.0912
3 3

101.7683
20.3783

31.1195

72.8646
53.4588

17.3560
1.26581

66.3566

32.0822
11.3921
0.2204

1.3255

3.4662
1.2547

8.9502
1.5541
9.1472
0.5680
DMO

g4 (X) = x4 − 240 ≤ 0
0 ≤ xi ≤ 100, i = 1, 2
10 ≤ xi ≤ 200, i = 3, 4
Numerous heuristic methods, such as GA3 [62], HAIS-GA [63], GA4
98311.2881
54537.8364

127.1830
[61], G-QPSO [64], CDE [65], ES [66], UPSO [67], BFOA [68], DHOA
20.9927

25.6644

86.0285
62.3762

16.6484
1.42937

73.8243
11.6652

11.3702

35.4274
13.7066
0.3485

2.0647

3.8643
2.0766

1.3364

0.7667
[69], QPSO [64], T-Cell [70], CPSO [71], and NHAIS-GA [72], have
BA

already been used to execute this case study. Table 14 presents the top
solutions for the pressure vessel optimum design issue utilizing the SAA.
63717.0042
40507.5899

Fig. 9 illustrates convergence characteristics to resolve the pressure

25.3346

18.6715

79.6296
42.1002

13.7408

82.8272
45.4855

19.6837
16.4371
vessel design issue for SAA. Table 15 lists the statistical findings from the
0.1988

4.7491

4.0239
3.5151

4.1094
1.0503
4.6815
0.8118
The statistical results of the different optimization algorithms for CEC-2019.

2.000

SAA and several earlier methods. Based on the “Worst,” “Mean,” and
DE

“Best” categories, the SAA offers results that are competitive with those
2762113.3142
1812667.8015

of the other methods. The acquired result presented the usefulness of the
SAA while requiring little in the way of manufacturing costs and
219.7034
131.6283
27.2416

24.6864

94.0281
63.1107

20.6303

42.8479
24.5412

computing labor.
0.3255

0.6938

3.8115
2.1492

3.2536

7.7018
3.4069
6.3601
2.1899
AO

4.4.2. Welded beam design (WBD)

17333354.4482

The objective of the WBD problem is to reduce the cost of welding

5895439.1901

beam while meeting two restrictions and taking four control variables
147.2793

235.8597
159.7590

122.2606

into account. In Fig. 10, the WBD is depicted. In an ideal design, certain
29.0861

93.7887
58.1151
15.5767

20.0695

32.7450

24.3267
0.2864

0.7365

3.6719

3.4375

3.4202

2.6601

3.2825

restrictions such deflection (δ), buckling load (Pc), stress of bending (σ ),

PSO

and end shear stress (τ ) are considered [60]. The control parameters,
{x1(h), x2(l), x3(t), x4(b)} are symbolic representations of the bar’s
737261832.0317
435718639.5550

thickness (b), clamping bar’s length (l), bar’s height (t), and weld
thickness (h) [60]. The issue is described in the following way [68]:
190.6001
141.2314

100.4542
25.1691

92.5525

99.5895
51.5666
11.6577

25.0418

23.2351

17.9220

11.4704

Minimize:
0.2406

2.5173

4.2024

3.2655

3.6665

1.3701
SHO

F(X) = 1.10471x2 x21 + 0.04811x3 x4 (14 +x2 )

Subject to the following constraints:
g1 (X) = τ(x) − τmax ≤ 0
Algorithm
Function/

F10: Ave
Table 13

F1: Ave

F2: Ave

F3: Ave

F4: Ave

F5: Ave

F6: Ave

F7: Ave

F8: Ave

F9: Ave
Std

Std

Std

Std

Std

Std

Std

Std

Std

Std

g2 (X) = σ (x) − σ max

272
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 7. Convergence process for different algorithms to solve some of CEC-2019 benchmark functions (F3, and F10).

273
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 8. Schematic of the PVD [76].

Table 14 g3 (X) = x1 − x4 ≤ 0 (6)

The best solution for the PVD.
Design variables SAA g4 (X) = 0.10471x21 + 0.04811x3 x4 (14 + x2 ) − 5 ≤ 0,
x1 0.8125
x2 0.4375 g5 (X) = 0.125 − x1 ≤ 0
x3 42.09845
x4 176.637 g6 (X) = δ(x) − δmax ≤ 0
g1(X) − 8.90E-10
g2(X) − 0.034
g7 (X) = P − Pc (x) ≤ 0,
g3(X) − 5.20E-05
g4(X) − 63.40 Where
Best 6059.7143
P = 6,000 lb; L = 14 in; E = 30e6 psi G = 12e6 psi, τmax =13,000 psi,
σ max =30,000 psi
δmax = 0.25 in,0.1 ≤ x1 ≤ 2, 0.1 ≤ x2 ≤ 10, 0.1 ≤ x3 ≤ 10,
0.1 ≤ x4 ≤ 2

Fig. 9. Convergence characteristic for PVD.

274
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 15 Table 16
The statistical results of different algorithms for PVD. The best solutions for the WBD.
Optimizers Best Mean Worst Std. Design variables SAA

GA3 [62] 6288.7445 6293.8432 6308.4970 7.4133 x1 0.20572964

HAIS-GA [63] 6832.584 7187.314 8012.615 276 x2 3.47048867
GA4 [61] 6059.9463 6177.2533 6469.3220 130.9297 x3 9.036624
G-QPSO [64] 6059.7208 6440.3786 7544.4925 448.4711 x4 0.20572964
CDE [65] 6059.7340 6085.2303 6371.0455 43.013 g1(X) − 2.2654E-07
ES [66] 6059.746 6850.00 7332.87 426 g2(X) − 3.19327E-07
UPSO [67] 6154.70 8016.37 9387.77 745.869 g3(X) 0.0
BFOA [68] 6060.460 6074.625 N.A. 156 g4(X) − 3.43299
DHOA [69] 6103.842 N.A. N.A. N.A. g5(X) − 0.08073
QPSO [64] 6059.7209 6440.3786 8017.2816 479.2671 g6(X) − 0.235541
T-Cell [70] 6390.554 6737.065 7694.066 357 g7(X) − 1.10549E-06
CPSO [71] 6061.0777 6147.1332 6363.8041 86.4545 Best 1.726007
NHAIS-GA [72] 6061.1229 6743.0848 7368.0602 457.99
SAA 6059.7143 6258.5078 6396.5624 125.37

4x22 − x1 x2 1
g2 (X) = + − 1≤0
SFO [73], GWO [74], GA4 [61], FSA [75], UPSO [71], SBM [76], 12, 566(x13 x2 − x41 ) 5, 108x21
SBO [77], IPSO [78], TEO [79], BA [80], CDE [65], CPSO [61], BFOA
[68], HSA-GA [81], and T-Cell [70] are just a few of the heuristic stra­ g3 (X) = 1 −
140.45x1
≤0
tegies that have previously been used to solve this case study. Table 16 x22 x3
offers the top SAA options for resolving this issue. Fig. 11 illustrates the
convergence characteristic used to resolve this issue for SAA. The image x1 + x2
g4 (X) = − 1≤0
makes it evident that the SAA produces its best results early in the course 1.5
of development. Table 17 presents the statistical outcomes utilizing Fig. 12 depicts the schematic for the TCSD issue. Numerous heuristic
various techniques. In terms of the “Worst,” “Mean,” and “Best” criteria, strategies, such as SI [82], CPSO [71], CDE [65], RL-BA [83], BFOA
the SAA outperformed the other methods with comparable and superior [68], CA [84], DDAO [85], GA4 [61], SBO [77], HGA [86], GA3 [62],
outcomes. The results produced demonstrated the SAA’s higher per­ GWO [74], (l + )-ES [67], and UPSO [88], have already been used to
formance in comparison to other algorithms. solve this case study. Table 18 and Fig. 13 exhibit, respectively, the best
solution based on the SAA as well as the convergence curve to arrive at
4.4.3. Tension/compression spring design (TCSD) the optimum solution to this issue. The statistical findings are evaluated
This TCSD objective is to reduce the stretched or compressed spring with those from other methods in Table 19 to investigate the SAA
weight as much as possible. The stress, wave frequency, and deflection capability. The findings demonstrated that, when compared to the al­
restrictions must be met in the ideal design. The three variables in this gorithms provided in earlier research, the SAA algorithm had competi­
design are d or x1 which stands for wire diameter, D or x2 which stands tive and superior outcomes in terms of meeting the “Worst,” “Mean,”
for mean coil diameter, and P or x3 which stands for the active coils and “Best” criteria. The findings demonstrated SAA superior capability
number. Mathematically, the objective function is represented as [60]: compared to the other algorithms.

F(X) = (x3 + 2)x2 x12

4.4.4. Three-bar truss design (3BTD)
This design aims for reduction the weight of a three-bar truss that is
0.05⩽x1 ⩽20.25⩽x2 ⩽1.32⩽x3 ⩽15
static. Three restrictions apply to this design. Fig. 14 depicts the sche­
by considering: matic of this design, and two parameters should be identified to modify
the sectional areas. Two cross-sectional area parameters (x1 and x2)
x32 x3
g1 (X) = 1 − ≤0 (7) must be optimized in this issue, as shown in Fig. 14. Three inequality
71, 785x41
design restrictions are imposed on the issue; their boundary conditions
are 0 < x1 and x2 < 1 [73]:

Fig. 10. Schematic of the WBD [76].

275
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 11. Convergence curve for WBD.

Table 18
Table 17
The best solutions for the TCSD.
Best statistical results of different algorithms for WBD.
Design variables SAA
Methods Best Mean Worst Std.
x1 0.051704
SFO [73] 1.73231 N.A. N.A. N.A. x2 0.357085
GWO [74] 1.72624 N.A. N.A. N.A. x3 11.267492
GA4 [61] 1.728226 1.792654 1.993408 7.47E-02 g1(X) − 5.669E-10
FSA [75] 2.3811 2.4041 2.4889 N.A. g2(X) − 3.944E-07
UPSO [71] 1.92199 2.83721 N.A. 6.83E-01 g3(X) − 4.055
SBM [76] 2.4426 2.5215 2.6315 N.A. g4(X) − 0.7278
SBO [77] 2.3854347 3.0025883 6.3996785 9.59E-01 Best 0.012666
IPSO [78] 2.3810 2.3819 N.A. 5.23E-03
TEO [79] 1.725284 1.768040 1.931161 5.81661E-02
BA [80] 1.7312065 1.8786560 2.3455793 2.677989E-01
Minimize:
CDE [65] 1.73346 1.768158 1.824105 2.2194E-02
CPSO [61] 1.728024 1.748831 1.782143 1.2926E-02 ( √̅̅̅ )
BFOA [68] 2.3868 2.4040 N.A. 1.6E-02 F(X) = l 2 2 x1 + x2
HSA-GA [81] 2.2500 2.26 2.28 7.8E-03
T-Cell [70] 2.3811 2.4398 2.7104 9.314E-02 Subject to:
SAA 1.726007 1.747929 1.781144 2.51E-03 √̅̅̅
2x1 + x2
g1 (X) = P √̅̅̅ 2 − σ≤0
2x1 + 2x1 x2

Fig. 12. Schematic of the TCSD.

276
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 13. Convergence curve for TCSD.

Table 19
Best statistical results of different algorithms for TCSD.
Methods Best Mean Worst Std.

SI [82] 0.013060 0.015526 0.018992 N.A.

CPSO [71] 0.0126747 0.012730 0.012924 5.19E-05
CDE [65] 0.012670 0.012703 0.012790 2.07E-05
RL-BA [83] 0.0126764 0.012745 0.0129281 7.19E-04
BFOA [68] 0.012671 0.012759 N.A. 1.36E-04
CA [84] 0.012721 0.013568 0.0151156 8.4E-04
DDAO [85] 0.0129065 0.0151829 0.0173199 1.26E-03
GA4 [61] 0.012681 0.012742 0.012973 9.5E-05
SBO [77] 0.012669249 0.012922669 0.016717272 5.92E-04
HGA [86] 0.012668 0.013481 0.016155 N.A.
GA3 [62] 0.0127048 0.012769 0.012822 3.93E-05
GWO [74] 0.0126660 N.A. N.A. N.A.
BIANCA [87] 0.012671 0.012681 0.012913 5.1232E-05
(l + λ)-ES [67] 0.012689 0.013165 N.A. 3.9E-04
UPSO [88] 0.01312 0.02294 N.A. 7.2E-03
SAA 0.012666 0.012695 0.012768 1.27E-05

x2 Fig. 14. Schematic of 3BTD [76].

g2 (X) = P √̅̅̅ 2 − σ≤0 (8)
2x1 + 2x1 x2
Table 20
1
g3 (X) = P √̅̅̅ − σ≤0 The best solutions for the 3BTD problem.
2x2 + x1
Design variables SAA
0 ≤ xi ≤ 1, i = 1, 2.
x1 0.7887
L = 100 cm, P = 2kN/cm2, σ = 2kN/cm2. x2 0.4083
Numerous heuristic methods, such as CS [67], SFO [73], and mGWO g1(X) − 2.781E-09
[89], have already been used to solve this situation. The most effective g2(X) − 1.460
g3(X) − 0.53601
solutions for this issue employing SAA and the convergence process are
Best 263.89586
depicted in Table 20 and Fig. 15, respectively. In Table 21, the statistical
findings are shown using various techniques. In terms of “Worst,”

277
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 15. Convergence curve for 3BTD.

(x3) and nF (x4) are discrete. According to [67], the goal purpose of this
Table 21 issue is as follows:
Best statistical results of different algorithms for 3BTD. Minimize:
Methods Best Mean Worst Std. (( ) ( ) )2
1 x2 x3
CS [67] 263.97156 264.0669 N.A. 9.0E-05 F(X) = − (9)
6.931 x1 x4
SFO [73] 263.89592128 N.A. N.A. N.A.
mGWO [89] 263.8961 N.A. N.A. N.A. 12 ≤ xi ≤ 60, i = 1, 2, 3, 4.
SAA 263.89586 263.99632 264.58427 8.53E-05
Many heuristic strategies, including mGWO [89], UPSO [67], and
MBA [90], have already been used to solve this situation. Table 22 and
“Mean,” and “Best,” the SAA outperformed the other algorithms and Fig. 17 show, respectively, the best solutions for this issue using SAA and
produced outcomes that were comparable and superior. The outcome the convergence process to solve it using SAA. In Table 23, the statistical
cleared that the SAA is superior when compared to other algorithms. findings are shown using various techniques. In terms of the “Worst,”
“Mean,” and “Best” criteria, the SAA outperformed the other algorithms
4.4.5. Gear train design (GTD) with comparable and superior outcomes. The results produced demon­
The objective of this task is to build a gear train gear ratio that is cost- strated the SAA’s higher performance in compared to other methods.
effective given the circumstances of parameter boundary limitations.
Fig. 16 shows the problem’s schematic in action. Because the teeth size 4.4.6. Cantilever beam design (CBD)
in a gear should be an integer, the choice variables nA (x1), nB (x2), nD A cantilever beam’s primary purpose is to reduce the weight of its 5

Fig. 16. Structure of the GTD [28].

278
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Table 22 Table 23
The best solutions for the GTD. Best statistical results of different algorithms for GTD.
Design variables SAA Methods Best Mean Worst Std.

x1 43 mGWO [89] 2.7009E-12 N.A. N.A. N.A.

x2 19 UPSO [67] 2.700857E-12 3.80562E-8 N.A. 1.09E-07
x3 16 MBA [90] 2.700857E-12 2.471635E-9 2.06290E-8 3.94E-09
x4 49 SAA 2.700857E-12 2.121561E-9 9.734686E-9 3.64E-9
Best 2.700857E-12

designs shown in Tables 24–25 demonstrate that SAA finds a better fit
hollow square pieces [60]. Adjustable parameters {x1, x2, x3, x4, x5} than other approaches. Fig. 19 depicts the convergence features to SAA’s
each describe the cross-sectional dimensions of a cube [60]. In Fig. 18, best solution to this issue; the ideal outcome is attained in the first stages
the fifth block is subjected to a vertical force, while the first block has of evolutions.
received heavy power. The issue has been mathematically formulated
using conventional beam theory [67]: 4.4.7. Hybrid system design
Minimize: The SAA’s ability to design a hybrid energy system according to
F(x) = 0.0624(x1 + x2 + x3 + x4 + x5 ), (10) Fig. 20, with photovoltaic panels, wind turbines, and fuel cells according
to the electrolyzer, hydrogen tank, and fuel cell resources [77,86] is
Subject to: assessed in this section. The project cost minimization includes capital,
61 37 19 7 1 maintenance, and operating expenses as well as cost of replacement
g(x) = + + + + − 1 ≤ 0,
x31 x23 x33 x34 x35 regardless of restrictions on reliability and device capacity [87]. The
flower pollination algorithm (FPA) had previously been utilized to solve
The design variables (x1, x2, x3, x4, x5) change between 0.01 and this design issue in Reference [67].
100. The components of the hybrid PV/WT/FC system, including renew­
Although this instance is not well-liked by academics, it could be able energy sources and storage devices, are provided in the sections
useful to use it as a standard in this field of research [67]. Numerous that follow. This PV panel’s power produced (P_PV) is calculated
heuristic methods, including WOA [47], CS [67], m-SCA [91], and SCA [67,88] taking into account both array angle and solar radiation:
[91], have already been used to solve this situation. Table 24 offers the
best statistical outcomes of the different algorithms for the cantilever IrV (t) × cos(θPV ) + IrH (t) × sin(θPV )
PPV = ×Prated MPPT
PV × ηPV × (11)
beam optimum design issue, and Table 25 presents the solutions of the Irref
studied optimizers used to the CBD. When numerous factors are taken Where, Irref stands for reference solar radiation (1000 W/m2), Prated
PV is
into account, the SAA outperforms other algorithms. The optimum the PV nominal power, IrV (t) and IrV (t) are the horizontal as well as

Fig. 17. Convergence curve for GTD.

279
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 18. The schematic of CBD.

to the fuel cell, MHST refers to the hydrogen mass of the tank, Δt is time
Table 24 step (1 h), ηHST is efficiency of the tank, HHV H2 is the value of hydrogen
The best solutions CBD. heat (39.7 kWh/kg).
Design variables SAA The fuel cells use hydrogen to generate electricity. The fuel cell’s
x1 6.0167
generated power is calculated by
x2 5.309012
PFC− = PHST− × ηFC (15)
x3 4.4940751
Inv FC

x4 3.50126 Where, ηFC is efficiency of the fuel cell.

x5 2.1527
The objective function (OF) for the PV/WT/FC system design is
g(X) − 1.13621E-08
Best 1.339966 described as the life span cost minimization by [67,67,88]
minNPC = NPCcc + NPCO&MC νNPCRC (16)
Where, NPCcc , NPCO&MC , and NPCRC refer to the costs of capital,
Table 25 maintenance and operating, and components replace
Best statistical results of various algorithms for CBD. { } { }
max PGi,min , P0Gi ⩽PGi ⩽min PGi,max , P0Gi + URGi ment, respectively.
Methods Best Mean Worst Std. Improving load reliability as a technical indicator is crucial when
WOA [47] 1.347944 1.436524 1.664954 0.076076 constructing the PV/WT/FC system. The scientific index in this scale is
CS [67] 1.33999 N.A. N.A. N.A. characterized as the probability of deficit power (PDP) [67].
m-SCA [91] 1.33999 N.A. N.A. N.A.
∑T
SCA [91] 1.356647 1.382846 1.426982 0.017955 [PLoad (t) − PInv− Load (t)]
SAA 1.339966 1.423006 1.63745 7.24E-03 PDP = t=1 ∑T (17)
t=1 [PLoad (t)]
∑T
where, t=1 [PLoad (t)] denotes total demand and PInv− Load (t) is the
vertical solar radiation radiated to the PV, θPV stands for the angle of the
delivered power of electrolyzer into the demand. So, the reliability
array with respect to the horizon as well, and ηMPPTPV denotes PV MPPT
constraint is presented as follows:
efficiency.
The amount of power obtained from a wind turbine is determined as LPSP ≤ LPSPmax (18)
follows: max
Where, LPSP refers to the upper value of reliability constraint
⎧ ( )μ
⎪
⎪ V − Vci (1)%).
⎪ P × ; Vci ≤ V ≤ Vn
⎪ n,wt
⎨ Vn − Vci Also, the problem constraints including number of PV arrays (NPV ),
PWT = (12) wind resources (NWT ), electrolyzer capacity (PEL ), hydrogen mass in the
⎪
⎪ Pn,wt ; Vn ≤ V ≤ Vci
⎪
⎪
⎩ tank (MHST ), fuel cell capacity (PFC ), inverter delivered power (PInv ) and
0V ≤ Vci , Vco ≤ V PVs angle relative to the horizon (θPV ) are given as follows [67]:
Where, PWT denotes power of WT, V clears wind speed, Vci , Vn and min
NPV max
≤ NPV ≤ NPV
Vco refer to the cut-in, rated and cut-out wind speed, and Pn,wt refers to
the rated power of WT. min
NWT max
≤ NWT ≤ NWT
Depending on the needs of the system’s load, extra electricity is sent
to the electrolyzer to produce hydrogen. The hydrogen generated is Pmin max
EL ≤ PEL ≤ PEL
directed to the hydrogen storage tank, where it is pressure-stored. The
energy of the hydrogen storage tank (HST) at time t (eHST(t)) is min
MHST max
≤ MHST ≤ MHST
computed by
eHST(t) = eHST(t − 1) + PEL− − PHST− × ηHST (13) Pmin max
FC ≤ PFC ≤ PFC
HST (t)Δt FC (t)Δt

eHST(t) Pmin max

Inv ≤ PInv ≤ PInv
MHST (t) = (14)
HHV H2
θmin max
PV ≤ θPV ≤ θPV (19)
Where, eHST(t − 1) is tank hydrogen energy at time (t-1), PEL− HST (t)
denotes the output injected power from electrolyzer to tank at time t, The size issue is applied to a supply of annual demand. By attaining
PHST− FC (t) is the equivalent power equivalent to the delivered hydrogen the lowest OF and adhering to the limitations, the aim is to identify the
ideal variables. The SAA is used to size problems, and there are 30,000

280
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 19. Convergence curve for CBD.

Fig. 20. Block diagram of the hybrid energy system.

4.4.8. Economic load dispatch problem

Table 26 The operation of electrical systems are crucial, and one issue relating
Best statistical results of various algorithms for the hybrid energy system sizing to these topics is the economic load dispatch issue, whose main goal is to
problem.
reduce the overall operating costs of electrical power generation while
Methods Best (M$) Mean (M$) Worst (M$) Std. (M$) taking into account a variety of operational and system’s limitations.
SAA 2.7582 2.7615 2.7639 0.000264 The following formulation [47,89–92] can be used to express the
FPA [67] 2.8022 2.8022 2.8050 0.000395 objective function of the economic load dispatch issue:
∑
NG ∑
NG

assessments in total. Table 26 and Fig. 21 present the best solutions for minFELD = FCi (PGi ) = (ai + bi PGi + ci P2Gi )
this problem utilizing the SAA and convergence procedures for SAA. In
i=1 i=1

Table 27, the statistical findings are displayed in a variety of ways. Subject to:
Compared to the other techniques, the CVSO’s statistical outcomes were
∑
NG
competitive and superior. The results demonstrated that the SAA per­ PGi = PD + PLoss
formed better than the FPA [67] in obtaining a lower objective function i=1

(NPC). The outcome demonstrated the SAA’s superior efficiency

compared contrast with the FPA [67]. PGi,min ≤ PGi ≤ PGi,max

281
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 21. Convergence characteristic for the sizing of the hybrid energy system.

Table 27 Table 29
The best solutions for the hybrid energy sys­ The optimal solution to the economic load
tem sizing problem. dispatch.
Variables SAA Variables CVSO

NPV 203 P1 (MW) 220.96

NWT 14 P2 (MW) 206.13
PEL (kW) 110.27 P3 (MW) 288.42
MHST (kg) 149.14 P4 (MW) 243.38
PFC (kW) 49.85 P5 (MW) 275.22
PInv (kW) 47.78 P6 (MW) 246.25
θPV (deg) 38.04 P7 (MW) 287.39
Best (M$) 2.7582 P8 (MW) 240.36
P9 (MW) 430.95
P10 (MW) 273.54
Best ($/h) 623.8247
Table 28
Optimal statistical outcomes of different algorithms for the problem of economic
load dispatch. [47,89,90], and 2700 MW is taken as the overall power consumption.
Methods Best ($/h) Mean ($/h) Worst ($/h) Std. For the SAA, the maximum iteration, population, and repetition are 200,
50, and 30 respectively. Given in Tables 28 and 29, respectively, are the
AA [89] 623.9500
statistically significant results and optimum solution for employing the
– – –
ORCSA [90] 623.8608 623.8963 623.9353 –
DEPSO [47] 623.8300 623.9000 624.0800 – SAA to solve the 10-unit benchmark power system with Multi-Fuel (MF)
SPPO [91] 623.8279 – – – and Valve-Point Effects (VPE) restrictions. The outcomes are then
IPSO [92] 623.8730 – – – compared with those of AA, ORCSA, DEPSO, SPPO, and IPSO. The
SAA 623.8247 623.8377 623.8476 0.0043
findings demonstrated that the suggested strategy is more cost-effective
than other methods as shown in Fig. 22. Its superior competitiveness has
thus been verified.
PGi − P0Gi ⩽URGi &P0Gi − PGi ⩽DRGi
{ } { } 5. Conclusion
max PGi,min , P0Gi ⩽PGi ⩽min PGi,max , P0Gi + URGi (20)

Where, NG denotes number of generation units number, PD indicates This paper presented a new optimizer method named snow ava­
system load, PLoss is to the system loss, P0Gi is initial active power of the ith lanches algorithm (SSA) which is modeled by snow avalanches in na­
generator, DRGi and URGi are the down-ramp and up-ramp rate con­ ture. The proposed algorithm included four phases for simulation,
straints of the ith unit, respectively. avalanche due to mountain slope, human factors, weather in the region,
The input data for the benchmark 10-unit power system is taken from and also normal conditions. A thorough analysis of the suggested

282
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

Fig. 22. Convergence characteristic for solving the economic load dispatch problem.

algorithm’s performance in solving 23 traditional test functions was the work reported in this paper.
given. The suggested method was competitive against state-of-the-art
MHAs. References

• The simulations related to the classic and CEC-2019 benchmark test [1] S. Ekinci, D. Izci, R. Abu Zitar, A.R. Alsoud, L. Abualigah, Development of Lévy
flight-based reptile search algorithm with local search ability for power systems
functions demonstrated that the SSA finds the best solutions in a engineering design problems, Neural Comput. & Applic. 34 (22) (2022)
simple manner while being remarkably robust and effective. 20263–20283.
Compared to previous techniques, the SSA has produced more [2] S. Ekinci, D. Izci, M.R. Al Nasar, R. Abu Zitar, L. Abualigah, Logarithmic spiral
search based arithmetic optimization algorithm with selective mechanism and its
acceptable outcomes, according to the findings, by offering higher application to functional electrical stimulation system control, Soft. Comput. 26
exploration and exploitation power. Additionally, comparisons have (22) (2022) 12257–12269.
shown that the SAA converged more quickly than other algorithms [3] Z. Wang, D. Zhao, Y. Guan, Flexible-constrained time-variant hybrid reliability-
based design optimization, Struct. Multidiscip. Optim. 66 (4) (2023) 89.
and was not caught in the local optimum. [4] Y. Zheng, L. Li, L. Qian, B. Cheng, W. Hou, Y. Zhuang, Sine-SSA-BP Ship Trajectory
• The competitive capability of the suggested algorithm in resolving Prediction Based on Chaotic Mapping Improved Sparrow Search Algorithm,
six engineering problems, including the design of pressure vessels, Sensors 23 (2) (2023) 704.
[5] D.H. Wolpert, W.G. Macready, No free lunch theorems for optimization, IEEE
welded beams, tension/compression springs, three-bar trusses, gear
Trans. Evol. Comput. 1 (1) (1997) 67–82.
trains, and cantilever beams, was confirmed for additional analyses [6] B. Cao, J. Zhao, Y. Gu, Y. Ling, X. Ma, Applying graph-based differential grouping
and its superior capability is proved compared to the other well- for multiobjective large-scale optimization, Swarm Evol. Comput. 53 (2020),
known algorithms. 100626.
[7] A.E. Ezugwu, O.J. Adeleke, A.A. Akinyelu, S. Viriri, A conceptual comparison of
• The SAA proves its superiority in hybrid energy sizing by obtaining a several metaheuristic algorithms on continuous optimisation problems, Neural
lower cost of M$ 2.7492 than TGA, LOA, ABC, PSO, BA, and HHO Comput. & Applic. 32 (10) (2020) 6207–6251.
algorithms, which attain costs of M$ 2.7878, M$ 2.7556, M$ 2.8521, [8] Z. Meng, J.S. Pan, K.K. Tseng, PaDE: An enhanced Differential Evolution algorithm
with novel control parameter adaptation schemes for numerical optimization,
M$ 2.9503, M$ 3.0942, and M$ 2.9614, respectively. Moreover, the Knowl.-Based Syst. 168 (2019) 80–99.
effectiveness of the SAA is assessed in order to solve the economic [9] F. Miao, Y. Zhou, Q. Luo, Complex-valued encoding symbiotic organisms search
load dispatch issue, and the results revealed that the proposed algorithm for global optimization, Knowl. Inf. Syst. 58 (2019) 209–248.
[10] M. Fadaee, M.A.M. Radzi, Multi-objective optimization of a stand-alone hybrid
approach achieved a lower cost than other approaches, and its su­ renewable energy system by using evolutionary algorithms: A review, Renew.
perior competitive performance was proven. Sustain. Energy Rev. 16 (5) (2012) 3364–3369.
• Generally, the advantages of the SAA are low control parameters, [11] B. Cao, W. Zhang, X. Wang, J. Zhao, Y. Gu, Y. Zhang, A memetic algorithm based
on two_Arch2 for multi-depot heterogeneous-vehicle capacitated arc routing
simple structure and also easy implementation. Also, according to problem, Swarm Evol. Comput. 63 (2021), 100864.
the NFL theorem, an optimization algorithm cannot show superior [12] M. Wang, H. Chen, Chaotic multi-swarm whale optimizer boosted support vector
performance in solving all optimization problems. SAA, like other machine for medical diagnosis, Appl. Soft Comput. 88 (2020), 105946.
[13] S. Ekinci, D. Izci, L. Abualigah, R.A. Zitar, A modified oppositional chaotic local
meta-heuristic algorithms, has challenges and limitations, one of
search strategy based Aquila Optimizer to design an effective controller for vehicle
which is increasing the tolerance of convergence and increasing the cruise control system, J. Bionic Eng. (2023) 1–24.
computational cost in the conditions of increasing the dimensions [14] M. Jahannoush, S.A. Nowdeh, Optimal designing and management of a stand-alone
and complexity of the problem in a large size, which can be easily hybrid energy system using meta-heuristic improved sine–cosine algorithm for
Recreational Center, case study for Iran country, Appl. Soft Comput. 96 (2020),
solved by using special techniques and improving the performance of 106611.
the algorithm. This challenge is solved, and the improvement of SAA [15] Y. Duan, Y. Zhao, J. Hu, An initialization-free distributed algorithm for dynamic
is also suggested for future work. economic dispatch problems in microgrid: Modeling, optimization and analysis,
Sustainable Energy Grids Networks 34 (2023), 101004.
[16] B. Cao, J. Zhao, Z. Lv, P. Yang, Diversified personalized recommendation
optimization based on mobile data, IEEE Trans. Intell. Transp. Syst. 22 (4) (2020)
Declaration of Competing Interest 2133–2139.
[17] J. Zhang, C. Zhu, L. Zheng, K. Xu, ROSEFusion: random optimization for online
dense reconstruction under fast camera motion, ACM Transactions on Graphics
The authors declare that they have no known competing financial
(TOG) 40 (4) (2021) 1–17.
interests or personal relationships that could have appeared to influence

283
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

[18] Y. Zheng, X. Lv, L. Qian, X. Liu, An optimal bp neural network track prediction [49] C. Tang, Y. Zhou, Z. Tang, Q. Luo, Teaching-learning-based pathfinder algorithm
method based on a ga–aco hybrid algorithm, Journal of Marine Science and for function and engineering optimization problems, Appl. Intell. 51 (2021)
Engineering 10 (10) (2022) 1399. 5040–5066.
[19] L. Qian, Y. Zheng, L. Li, Y. Ma, C. Zhou, D. Zhang, A new method of inland water [50] Atashpaz-Gargari, E., & Lucas, C. (2007, September). Imperialist competitive
ship trajectory prediction based on long short-term memory network optimized by algorithm: an algorithm for optimization inspired by imperialistic competition.
genetic algorithm, Appl. Sci. 12 (8) (2022) 4073. In 2007 IEEE congress on evolutionary computation (pp. 4661-4667). Ieee.
[20] H. Garg, A hybrid PSO-GA algorithm for constrained optimization problems, Appl. [51] M. Dehghani, E. Trojovská, P. Trojovský, A new human-based metaheuristic
Math Comput. 274 (2016) 292–305. algorithm for solving optimization problems on the base of simulation of driving
[21] Z. Li, Y. Xia, H. Sahli, CSA-DE/EDA: a Novel Bio-inspired Algorithm for Function training process, Sci. Rep. 12 (1) (2022) 1–21.
Optimization and Segmentation of Brain MR Images, Cogn. Comput. 11 (6) (2019) [52] R. Moghdani, K. Salimifard, Volleyball premier league algorithm, Appl. Soft
855–868. Comput. 64 (2018) 161–185.
[22] D. Yu, Y. Wang, H. Liu, K. Jermsittiparsert, N. Razmjooy, System identification of [53] E.J. Hopfinger, Snow avalanche motion and related phenomena, Annu. Rev. Fluid
PEM fuel cells using an improved Elman neural network and a new hybrid Mech. 15 (1) (1983) 47–76.
optimization algorithm, Energy Rep. 5 (2019) 1365–1374. [54] S.P. Pudasaini, K. Hutter, Avalanche dynamics: dynamics of rapid flows of dense
[23] Z. Wang, Q. Luo, Y. Zhou, Hybrid metaheuristic algorithm using butterfly and granular avalanches, Springer Science & Business Media, 2007.
flower pollination base on mutualism mechanism for global optimization [55] M.E. Eglit, K.S. Demidov, Mathematical modeling of snow entrainment in
problems, Eng. Comput. 37 (2021) 3665–3698. avalanche motion, Cold Reg. Sci. Technol. 43 (1–2) (2005) 10–23.
[24] Jahannoosh, M., Nowdeh, S. A., Naderipour, A., Kamyab, H., Davoudkhani, I. F., & [56] C. Ancey, V. Bain, Dynamics of glide avalanches and snow gliding, Rev. Geophys.
Klemeš, J. J. New hybrid meta-heuristic algorithm for reliable and cost-effective 53 (3) (2015) 745–784.
designing of photovoltaic/wind/fuel cell energy system considering load [57] M. Abdel-Basset, L. Abdel-Fatah, A.K. Sangaiah, Metaheuristic algorithms: A
interruption probability. Journal of Cleaner Production, 278, 123406. comprehensive review, Computational Intelligence for Multimedia Big Data on the
[25] Davoodkhani, F., Nowdeh, S. A., Abdelaziz, A. Y., Mansoori, S., Nasri, S., & Alijani, Cloud with Engineering Applications (2018) 185–231.
M. (2020). A new hybrid method based on gray wolf optimizer-crow search [58] A.A. Ameen, T.A. Rashid, S. Askar, CDDO–HS: Child Drawing Development
algorithm for maximum power point tracking of photovoltaic energy system. Optimization-Harmony Search Algorithm, Appl. Sci. 13 (9) (2023) 5795.
In Modern Maximum Power Point Tracking Techniques for Photovoltaic Energy [59] A. Faramarzi, M. Heidarinejad, B. Stephens, S. Mirjalili, Equilibrium optimizer: A
Systems (pp. 421-438). Springer, Cham. novel optimization algorithm, Knowledge-Based Syst. 191 (2020), 105190.
[26] S. Ekinci, D. Izci, E. Eker, L. Abualigah, An effective control design approach based [60] Kaveh, A., M. Khanzadi, and M. Rastegar Moghaddam. “Billiards-inspired
on novel enhanced aquila optimizer for automatic voltage regulator, Artif. Intell. optimization algorithm; a new meta-heuristic method.” Structures. Vol. 27.
Rev. 56 (2) (2023) 1731–1762. Elsevier, 2020.
[27] S.A. Nowdeh, I.F. Davoudkhani, M.H. Moghaddam, E.S. Najmi, A.Y. Abdelaziz, [61] C.A.C. Coello, E.M. Montes, Constraint-handling in genetic algorithms through the
A. Ahmadi, F.H. Gandoman, Fuzzy multi-objective placement of renewable energy use of dominance-based tournament selection, Adv. Eng. Informatics 16 (3) (2002)
sources in distribution system with objective of loss reduction and reliability 193–203.
improvement using a novel hybrid method, Appl. Soft Comput. 77 (2019) [62] C.A.C. Coello, Use of a self-adaptive penalty approach for engineering optimization
761–779. problems, Comput. Ind. 41 (2) (2000) 113–127.
[28] Goldberg, D. E. (2013). Genetic algorithms. pearson education India. [63] C.A.C. Coello, N.C. Cortés, Hybridizing a genetic algorithm with an artificial
[29] Q. Luo, X. Yang, Y. Zhou, Nature-inspired approach: An enhanced moth swarm immune system for global optimization, Eng. Optim. 36 (5) (2004) 607–634.
algorithm for global optimization, Math. Comput. Simul 159 (2019) 57–92. [64] L. dos Santos Coelho, Gaussian quantum-behaved particle swarm optimization
[30] D. Simon, Biogeography-based optimization, IEEE Trans. Evol. Comput. 12 (6) approaches for constrained engineering design problems, Expert Syst. Appl. 37 (2)
(2008) 702–713. (2010) 1676–1683.
[31] M. Dorigo, M. Birattari, T. Stutzle, Ant colony optimization, IEEE Comput. Intell. [65] F. Huang, L. Wang, Q. He, An effective co-evolutionary differential evolution for
Mag. 1 (4) (2006) 28–39. constrained optimization, Appl. Math Comput. 186 (1) (2007) 340–356.
[32] G. Dhiman, V. Kumar, Spotted hyena optimizer: a novel bio-inspired based [66] E. Mezura-Montes, C.A.C. Coello, An empirical study about the usefulness of
metaheuristic technique for engineering applications, Adv. Eng. Softw. 114 (2017) evolution strategies to solve constrained optimization problems, Int. J. Gen Syst 37
48–70. (4) (2008) 443–473.
[33] Kennedy, J., & Eberhart, R. (1995, November). Particle swarm optimization. [67] K. E. Parsopoulos and M. N. Vrahatis, “Unified particle swarm optimization for
In Proceedings of ICNN’95-International Conference on Neural Networks (Vol. 4, solving constrained engineering optimization problems,” in International
pp. 1942-1948). IEEE. conference on natural computation, 2005, pp. 582–591.
[34] L. Abualigah, D. Yousri, M. Abd Elaziz, A.A. Ewees, M.A. Al-Qaness, A.H. Gandomi, [68] E. Mezura-Montes and B. Hernández-Ocana, “Bacterial foraging for engineering
Aquila optimizer: a novel meta-heuristic optimization algorithm, Comput. Ind. design problems: preliminary results,” Lab. Nac. Informática Av. (LANIA AC)-
Eng. 157 (2021), 107250. Universidad Juárez Autónoma Tabasco. México, 2008.
[35] J.O. Agushaka, A.E. Ezugwu, L. Abualigah, Dwarf mongoose optimization [69] G. Brammya, S. Praveena, N.S. Ninu Preetha, R. Ramya, B.R. Rajakumar, D. Binu,
algorithm, Comput. Methods Appl. Mech. Eng. 391 (2022), 114570. Deer hunting optimization algorithm: a new nature-inspired meta-heuristic
[36] K. Zhong, G. Zhou, W. Deng, Y. Zhou, Q. Luo, MOMPA: Multi-objective marine paradigm, Comput. J. (2019).
predator algorithm, Comput. Methods Appl. Mech. Eng. 385 (2021), 114029. [70] V.S. Aragón, S.C. Esquivel, C.A.C. Coello, A modified version of a T-Cell Algorithm
[37] N. Chopra, M.M. Ansari, Golden jackal optimization: A novel nature-inspired for constrained optimization problems, Int. J. Numer. Meth. Eng. 84 (3) (2010)
optimizer for engineering applications, Expert Syst. Appl. 198 (2022), 116924. 351–378.
[38] D. Karaboga, B. Basturk, On the performance of artificial bee colony (ABC) [71] Q. He, L. Wang, An effective co-evolutionary particle swarm optimization for
algorithm, Appl. Soft Comput. 8 (1) (2008) 687–697. constrained engineering design problems, Eng. Appl. Artif. Intel. 20 (1) (2007)
[39] X.S. Yang, Bat algorithm for multi-objective optimisation, International Journal of 89–99.
Bio-Inspired Computation 3 (5) (2011) 267–274. [72] H.S. Bernardino, H.J.C. Barbosa, A.C.C. Lemonge, L.G. Fonseca, “A new hybrid AIS-
[40] E. Rashedi, H. Nezamabadi-Pour, S. Saryazdi, GSA: a gravitational search GA for constrained optimization problems in mechanical engineering”, in, IEEE
algorithm, Inf. Sci. 179 (13) (2009) 2232–2248. Congress on Evolutionary Computation (IEEE World Congress on Computational
[41] B. Cao, J. Zhao, P. Yang, Y. Gu, K. Muhammad, J.J. Rodrigues, V.H.C. de Intelligence) 2008 (2008) 1455–1462.
Albuquerque, Multiobjective 3-D topology optimization of next-generation [73] S. Shadravan, H.R. Naji, V.K. Bardsiri, The Sailfish Optimizer: A novel nature
wireless data center network, IEEE Trans. Ind. Inf. 16 (5) (2019) 3597–3605. inspired metaheuristic algorithm for solving constrained engineering optimization
[42] M. Ghasemi, I.F. Davoudkhani, E. Akbari, A. Rahimnejad, S. Ghavidel, L. Li, problems, Eng. Appl. Artif. Intel. 80 (2019) 20–34.
A novel and effective optimization algorithm for global optimization and its [74] S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer, Adv. Eng. Softw. 69
engineering applications: Turbulent Flow of Water-based Optimization (TFWO), (2014) 46–61.
Eng. Appl. Artif. Intel. 92 (2020), 103666. [75] A.-R. Hedar, M. Fukushima, Derivative-free filter simulated annealing method for
[43] B. Alatas, ACROA: artificial chemical reaction optimization algorithm for global constrained continuous global optimization, J. Glob. Optim. 35 (4) (2006)
optimization, Expert Syst. Appl. 38 (10) (2011) 13170–13180. 521–549.
[44] A. Kaveh, T. Bakhshpoori, Water evaporation optimization: a novel physically [76] S. Akhtar, K. Tai, T. Ray, A socio-behavioural simulation model for engineering
inspired optimization algorithm, Comput. Struct. 167 (2016) 69–85. design optimization, Eng. Optim. 34 (4) (2002) 341–354.
[45] A.A. Hudaib, H.N. Fakhouri, Supernova optimizer: a novel natural inspired meta- [77] T. Ray, K.-M. Liew, Society and civilization: an optimization algorithm based on
heuristic, Mod. Appl. Sci. 12 (1) (2018) 32–50. the simulation of social behavior, IEEE Trans. Evol. Comput. 7 (4) (2003) 386–396.
[46] A.H. Kashan, An efficient algorithm for constrained global optimization and [78] S. He, E. Prempain, Q.H. Wu, An improved particle swarm optimizer for
application to mechanical engineering design: League championship algorithm mechanical design optimization problems, Eng. Optim. 36 (5) (2004) 585–605.
(LCA), Comput. Aided Des. 43 (12) (2011) 1769–1792. [79] A. Kaveh, A. Dadras, A novel meta-heuristic optimization algorithm: thermal
[47] A. Sadollah, A. Bahreininejad, H. Eskandar, M. Hamdi, Mine blast algorithm: A exchange optimization, Adv. Eng. Softw. 110 (2017) 69–84.
new population based algorithm for solving constrained engineering optimization [80] A.H. Gandomi, X.-S. Yang, A.H. Alavi, S. Talatahari, Bat algorithm for constrained
problems, Appl. Soft Comput. 13 (5) (2013) 2592–2612. optimization tasks, Neural Comput. & Applic. 22 (6) (2013) 1239–1255.
[48] P. Agrawal, T. Ganesh, A.W. Mohamed, A novel binary gaining–sharing [81] S.-F. Hwang, R.-S. He, A hybrid real-parameter genetic algorithm for function
knowledge-based optimization algorithm for feature selection, Neural Comput. & optimization, Adv. Eng. Informatics 20 (1) (2006) 7–21.
Applic. 33 (11) (2021) 5989–6008. [82] T. Ray, P. Saini, Engineering design optimization using a swarm with an intelligent
information sharing among individuals, Eng. Optim. 33 (6) (2001) 735–748.

284
K. Golalipour et al. Alexandria Engineering Journal 83 (2023) 257–285

[83] X.-B. Meng, H.-X. Li, X.-Z. Gao, An adaptive reinforcement learning-based bat [88] E. Mezura-Montes and C. A. C. Coello, “Useful infeasible solutions in engineering
algorithm for structural design problems, Int. J. Bio-Inspired Comput. 14 (2) optimization with evolutionary algorithms,” in Mexican international conference
(2019) 114–124. on artificial intelligence, 2005, pp. 652–662.
[84] C.A. Coello Coello, R.L. Becerra, Efficient evolutionary optimization through the [89] A.H. Gandomi, X.-S. Yang, A.H. Alavi, Cuckoo search algorithm: a metaheuristic
use of a cultural algorithm, Eng. Optim. 36 (2) (2004) 219–236. approach to solve structural optimization problems, Eng. Comput. 29 (1) (2013)
[85] H.N. Ghafil, K. Jármai, Dynamic differential annealed optimization: New 17–35.
metaheuristic optimization algorithm for engineering applications, Appl. Soft [90] S. Gupta, K. Deep, A memory-based grey wolf optimizer for global optimization
Comput. 93 (2020), 106392. tasks, Appl. Soft Comput. 93 (2020), 106367.
[86] H.S. Bernardino, H.J.C. Barbosa, A.C.C. Lemonge, “A hybrid genetic algorithm for [91] Z. Liu, T. Nishi, Multipopulation ensemble particle swarm optimizer for
constrained optimization problems in mechanical engineering”, in, IEEE Congress engineering design problems, Math. Probl. Eng. 2020 (2020).
on Evolutionary Computation 2007 (2007) 646–653. [92] S. Gupta, K. Deep, A hybrid self-adaptive sine cosine algorithm with opposition
[87] M. Montemurro, A. Vincenti, P. Vannucci, The automatic dynamic penalisation based learning, Expert Syst. Appl. 119 (2019) 210–230.
method (ADP) for handling constraints with genetic algorithms, Comput. Methods
Appl. Mech. Eng. 256 (2013) 70–87.

285