Applied Mathematical Modelling 71 (2019) 45–59

Contents lists available at ScienceDirect

Applied Mathematical Modelling

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

A balanced whale optimization algorithm for constrained

engineering design problems
Huiling Chen a, Yueting Xu a, Mingjing Wang a, Xuehua Zhao b,∗
a
Department of Computer Science, Wenzhou University, Wenzhou 325035, China
b
School of Digital Media, Shenzhen Institute of Information Technology, Shenzhen 518172, China

a r t i c l e i n f o a b s t r a c t

Article history: In this study, two novel effective strategies composed of Lévy flight and chaotic local
Received 7 June 2018 search are synchronously introduced into the whale optimization algorithm (WOA) to
Revised 24 January 2019
guide the swarm and further promote the harmony between the inclusive exploratory
Accepted 4 February 2019
and neighborhood-informed capacities of the conventional technique and investigate the
Available online 12 February 2019
core searching capabilities of WOA in dealing with optimization tasks. However, the con-
Keywords: ventional WOA may simply be stuck at local optima or the global best may not be ob-
Whale optimization algorithm tained successfully when tackling more complex optimization landscapes, including the
Lévy flight multimodal and high dimensional scenarios. To substantiate the efficacy of the enhanced
Chaotic local search method, it is compared to a set of well-regarded variants of particle swarm optimization
Fixed-dimension functions and differential evolution. The used benchmark problems are composed of unimodal, mul-
Complex optimization tasks timodal, and fixed-dimensions multimodal functions. Additionally, the proposed balanced
Welded beam
method is applied to realize three practical, well-known mathematical models such as ten-
sion/compression spring, welded beam, pressure vessel design, three-bar truss design, and
I-beam design problems. The experimental results and analysis reveal that the proposed
algorithm can outperform other competitors in terms of the convergence speed and the
quality of solutions. Promisingly, the proposed method can be treated as an effective and
efficient auxiliary tool for more complex optimization models and scenarios.
© 2019 Published by Elsevier Inc.

1. Introduction

WOA is one of the latest swarm-based algorithms that has gained extensive attention in last year [1]. It is detected that
this method can show high performance on many complex optimization tasks. Diversification of the search space (explo-
ration) and intensification of the best agent (exploitation) are two contradictory points which must be taken into consider-
ation when constructing or employing a metaheuristic algorithm [2–5]. Therefore, a fact that cannot be ignored when using
swarm-based optimizers is that there should be a stable balance between exploitation and exploration inclinations and it
can be often enhanced [6–13]. In this study, two effective strategies composed of Lévy flight (LF) and chaotic local search
(CLS) are embedded into the original WOA and then, a novel BWOA algorithm is proposed to further reduce the possibil-
ity of local optima (LO) stagnation and premature convergence (PC) drawbacks, which can guarantee that a better balance
between the exploitation and exploration trends can be fulfilled.

∗
Correspondence author.
E-mail address: [email protected] (X. Zhao).

https://doi.org/10.1016/j.apm.2019.02.004
0307-904X/© 2019 Published by Elsevier Inc.
46 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

LF adopts a random walk, where the step length is described by a heavy-tailed probability distribution [14]. Due to
its good control ability for local search, this concept has attracted much attention in optimization literature. For instance,
Heidari and Pahlavani improved the efficacy of grey wolf optimizer with this concept [15]. Ali et al. [16] proposed using
this method to enhance the search capabilities of the cultural algorithm. Sharma et al. [17] explored the search capability of
an artificial bee colony with LF strategy. Xie et al. [18] adopted the LF-based trajectory and differential operator to relieve
the shortcomings of a sluggish convergence rate and undesirable accuracy of bat algorithm. CLS is a local search approach
based on the chaos theory, in which, the agents can show randomicity and ergodicity [19]. Chaos theory has also gained
tremendous attention in metaheuristic filed. Alatas et al. [20] presented utilizing chaotic maps for parameter adaptation
of particle swarm optimization (PSO). Jia et al. [21] investigated the efficacy of memetic differential evolution (DE) with
chaotic local search. Hosseini et al. [22] also introduced the chaos into the PSO to avoid the search from being trapped in
local optima.
In several works, it has been verified that the conventional WOA or its variants show competitive, preferable, or promis-
ing performance compared to other well-established optimizers [23]. A fact is that the efficacy of the original method can be
still improved, due to the disadvantages of stagnation or premature convergence with theoretical possibility. Furthermore,
much attention has been paid to the LF and CLS to enhance metaheuristic algorithms. Therefore, in this work, the LF and
CLS are introduced into the original WOA, synchronously, and a novel BWOA algorithm is proposed to further alleviate the
PC and LO stagnation. In order to validate the effectiveness of BWOA, this algorithm is applied to nine benchmark functions,
involving unimodal, multimodal, and fixed-dimension multi-modal functions, and compared to other well-regarded methods
such as MFO, and well-known variants of PSO and DE such as ALCPSO [24], CLPSO [25], jDE [26], and SaDE [27]. As the ex-
perimental results demonstrate, these two effective strategies can improve the efficacy of WOA and the proposed BWOA has
a relatively better performance compared to other competitors. Furthermore, the proposed algorithm is employed to deal
with five practical mathematical modeling problems such as tension/compression spring (TCS), welded beam (WB), pressure
vessel design (PV), three-bar truss design (TBT), and I-beam design (IBD) problems. The experimental results expose that
the BWOA can be an effective tool for these modeling cases and similar practical problems, as well.
The rest of this paper is organized as follows: Section 2 describes the proposed BWOA. The experiments design is ex-
hibited in Section 3. The experimental results are reported in Section 4. Section 5 presents some discussions on this study.
Finally, the concluding remarks and future works are summarized in Section 6.

2. The balanced WOA with LF and CLS

The full description of the basic WOA is provided in the original paper [1]. In this section, we propose the balanced
variant of WOA, which is called BWOA. To develop the BWOA, LF strategy and CLS mechanism are both synchronously
introduced into the WOA to further relieve the PC and LO stagnation drawbacks, which are the core disadvantages of the
WOA.
The Lévy distribution is defined as follows:

L(z ) ∼ z−1−β , 0 < β ≤ 2 (1)

A    
z= 1 /β
, A ∼ N 0, σ 2 , B ∼ N 0, σ 2 (2)
|B|
  2 /β
(1 + β ) sin (π β /2 )
δ2 = · (3)
β  [(1 + β )/2] 2(β −1)/2
where z is the step size and β means the Lévy index, which can control the stability [28], A/B ∼ N(0, σ 2 ) indicates that the
samples originate from a Gaussian distribution, whose mean and variance are zero and δ 2 , respectively, and  (·) represents
the Gamma function. As it is shown in [29, 30], the parameter β is set to 3/2. To further reduce the possibility of LO
stagnation and PC of conventional WOA, LF strategy was used to update the positions of agents, while the parameter β is
set to 3/2 in this study.
As stated in the original article, the agents often swim synchronously around the quarry, which the mathematical model
is stated as follows:

X ∗ (t ) − AD if p < 0.5
X (t + 1 ) = (4)
D ebl cos (2π l ) + X ∗ (t ) if p ≥ 0.5

where A and C are calculated by A = 2a · r − a and C = 2 · r, respectively, where a is linearly decreased from 2 to 0, r
is generated in [0, 1] randomly. X(t), X∗ (t), and X (t) are all agent’s position vectors and they have the same number of
dimensions, which is depended on the pending task, each dimension means one value of an unknown parameter in the
pending task within limits. X(t) is a current search agent (a position vector), X∗ (t) means the current leader vector, which
has the best objective value, t is the current iteration. D is calculated by D = |C · X (t) − X(t)| and X (t) is a random position
vector chosen from the current population. D is calculated by D = |X∗ (t) − X(t)|, which means the distance of an agent to
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 47

the leader, b defines the shape of a logarithmic spiral and it is a constant, and l is generated between [−1,1] randomly. p is
a random value inside [0, 1] to control the position updating mechanisms.
The LF mechanism is used to generate a new agent corresponding to the current swarm when the position of i-th agent
X(t + 1) is updated. The new candidate solution can be formulated as follows:

X (t + 1 ) + (X ∗ (t ) − AD ) · levy(β )  if p < 0.5
Xl (t + 1 ) = (5)
X (t + 1 ) + D ebl cos (2π l ) + X ∗ (t ) · levy(β ) if p ≥ 0.5
where Xl (t + 1) is the new state of ith search individual X(t) and that · is the dot product (entry-wise multiplications). In
order to guarantee the excellence of the agents, individuals with fitter values are kept.

Xl (t + 1 ) F (Xl (t + 1 ) ) > F (X(t + 1 ) )
X (t + 1 ) = (6)
X (t + 1 ) otherwise
According to Eq. (6), the better agents in each iteration can retain to be greatly enriched for the next iterations and the
other ones are discarded.
The combination of CLS mechanism and the original WOA not only makes the algorithm more capable to avoid falling
into LO but also it enhances the searching capacities and make a better harmony between exploratory and exploitative
patterns. To decrease the occurrence of LO stagnation and achieve better results, CLS is introduced into the WOA algorithm.
As is shown in [19], the logistic chaotic function can be also employed to construct a chaotic WOA. The logistic chaotic
function is generated as follows:
ϑk+1 = μϑk (1 − ϑk ), k = 1, 2, . . . , n; ϑ1 ∈ (0, 1), ϑ1 = 0.25, 0.5, and 0.75 (7)
where μ denotes a regulator parameter and n is the number of agents. It is not difficult to substantiate that by μ = 4, we
experience chaos-based patterns.
In some studies, local search is often used to construct the solutions of the objective function. However, it is not to be
ignored that common local search may lead to rapid convergence or entrapment into LO. The CLS, owing to its randomicity
and ergodicity, can effectively overcome these shortcomings. The chaotic local search can be generated as follows:
Xnew (t ) = (1 − S )X (t ) + Sϑc (8)
where Xnew (t) means a new vector of X(t) drawn by CLS, ϑc is calculated by the formula ϑc = lu + ϑk (lb − lu ), where ϑk is
generated by the used signal (see Eq. (7)), in which ϑ1 is randomly created in [0,1], and then, individual X(t) is mapped into
the search space [lb , lu ], S is the contraction factor, which is determined as follows [21]:
S = 1 − |(I − 1 )/I|m (9)
where I represents iteration and m controls the shrinking speed. For the efficient incorporation of CLS, the agents that will
experience the CLS should be properly nominated. Hence, only the fittest agent is selected to be guided by the CLS strategy.
The CLS chooses the best agent’s position as the base, and then forms the new position of the best agent. The candidate
solution CX(t) of the fittest agent can be determined as follows:
CX (t ) = (1 − S )X ∗ (t ) + Sϑc (10)
In order to ensure the excellence of the leader, candidate position with fitter objective values will be retained during the
CLS process. Hence, the quality of the agent population is improved, and consequently, the risk of stagnation in LO can be
relieved.
The detailed procedure of BWOA can be seen in Algorithm 1. It can be seen from Algorithm 1 that the parameter p
and value of |A| controls the position updating. In this study, using any condition, the LF-based phase will be employed to
reconstruct search agents in order to obtain agents with better fitness and then, CLS maps the agents originated from the
former step. The Flowchart of BWOA is also shown in Fig. 1.
Note that the computational complexity of the developed BWOA method mainly depends on six producers: initialization,
fitness evaluation, sorting, agents updating, LF strategy, and CLS mechanism. Considering the population of agents with
N search agents and the sorting mechanism, the computational complexity of initialization producers is O(N)+O(N × logN).
Updating the positions of all agents is O(g × N × D), where g is the maximum number of generations and D is the dimension.
The computational complexity of updating the positions of all agents by LF strategy is O(g × N × D) and the CLS strategy for
the best agent is O(g × D). Hence, the computational complexity of BWOA is O(N + N × logN)+O (g × D(2 × N + 1)).

3. Experimental studies and comparisons

In order to strictly investigate the capabilities of the enhanced BWOA, several experiments are performed to solve various
mathematical benchmark cases.

3.1. System details

In this study, any involved algorithm is tested using MATLAB R2014b software under a Windows 10 with Intel (R) Core
(TM) i7-4790 CPU @ 3.6 GHz with 8GB of RAM.
48 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

Algorithm 1 Pseudo-code of proposed BWOA.

Inputs: population size N, number of iterations M

Outputs: the best search agent and its fitness value
Initialize the agents’ population Xl (i = 1, 2, …, N)
Initialize the parameters A, a, C, I, and p;
Calculate the fitness of each agent;
Save the best search agent X∗
while (j<M)
for i = 1 to N
Generate a, A, C, L, and p
if (p<0.5)
if (|A| < 1)
Update the position of the current agent by Eq. (1);
else if (|A| ≥ 1)
Update the position around a random agent by Eq. (6);
end if
else if (p>0.5)
Update the position with the spiral-shaped path by Eq. (3);
end if
Update the position by LF strategy and then obtain a better agent;
end for
Adopt the CLS strategy for the best agent X∗ ;
Make sure that every search agent is not beyond the boundary;
Update X∗ if there is a better search agent;
end while
return X∗

3.2. Continuous benchmark set

For comparison purposes, a well-regarded benchmark set with four unimodal, multi-modal, and fixed-dimension multi-
modal functions is chosen from 23 common benchmarks [31, 32]. Evidently, the unimodal benchmark functions only have
one global optimum, which can be used to evaluate the exploitative tendencies of algorithms. The multimodal functions

Fig. 1. Flowchart of balanced BWOA.

 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 49

Table 1
Description of benchmark functions.

Name Function Dim. Range fmin

 
Schwefel 2.22 F1 (x ) = ni=1 |xi | + ni=1 |xi | 30 [−10,10] 0
n−1 2
F2 (x ) = i=1 [100(xi+1 − x2i ) + (xi − 1 ) ]
2
Rosenbrock 30 [−30, 30] 0
 −1
F3 (x ) = πn {10sin (π y1 ) + ni=1 (yi − 1 )2 [1 + 10sin2 (π yi+1 )]⎧
+ ( yn − 1 ) }
2 2
Penalized 30 [−50,50] 0


⎨k ( xi − a )
m
xi > a
+ ni=1 u(xi , 10, 100, 4), yi = 1 + 14 (xi + 1 ), u(xi , a, k, m) = 0 −a < xi < a
⎩k(−x − a )m xi < −a
i
7 4
F4 (x ) = − i=1 j=1 [(X j − ai j )(X j − ai j ) + ci ]
T −1
Shekel7 4 [0,10] −10.4028

Table 2
Parameters setting for compared methods.

Method Parameters

BWOA a1 =[2 0]; a2 =[−2 −1]; b = 1; m = 2500

WOA a1 =[2 0]; a2 =[−2 −1]; b = 1
MFO b = 1; t=[−1 1]; a ∈ [−1 −2]
ALCPSO w = 0.4; c1 = 2; c2 = 2; lifespan = 60;T = 2
CLPSO w=[0.2 0.9]; c = 1.496
jDE tau1 =0.1; tau2 =0.1
SaDE numst=4; learngen=50

often include several LO, which means these functions are useful when the exploration capability of an algorithm needs to
be evaluated. The formulation of used functions and their brief descriptions are provided in Table 1.

3.3. Compared optimizers and details

To substantiate the efficiency of BWOA, other common algorithms are also conducted on the used functions. Therefore,
the MFO, ALCPSO [24], CLPSO [25], jDE [26], SaDE [27], and the original WOA are utilized to optimize these benchmark tasks
and the experimental results are compared to those obtained by BWOA. Table 2 reports the parameters of used techniques.
For fair comparisons, the parameters of involved solvers are all set in a relatively fair way. Therefore, for all methods, the
dimension (D) of solution space is set to 30, we set the swarm size to be 20, and the maximum number of iterations is set
to 10 0 0. All methods are tested under the same condition and operating system. The initial parameters are set based on the
acclaimed settings in prior papers and other best parameters are set using test and trial. All settings are exposed in Table 2.
It is noteworthy that each solution of the experiments is executed according to the average results over 30 runs to reduce
stochastic error in this study.
The numerical results of these methods in terms of the average value (AVG) and standard deviation (STD) of the function
error rates are attained to assess the potentials of associated techniques, and the best result of each task is marked in bold.
The function error value for each search agent X  is represented by f(X) − f(X∗ ), where X∗ is the optimum solution of the
objective function [31, 32]. In addition, the non-parametric statistical Wilcoxon sign rank test [33] at 0.05 significant level
is used to estimate the statistically significant difference between the proposed method and other competitors. It should
be noted that the symbols of “+”, “=” and “-” indicate that BWOA is better than, equal to, and inferior to other peers,
respectively. Apart from these, the Friedman test is also utilized to rank the average performance of all the selected methods
for further statistical comparison, and ARV (the average ranking value) is reported in comparison results.

3.4. Impact of parameter m on the performance of BWOA

As mentioned in the study [21], the final accuracy of found solutions strongly depends on the parameter m (as shown
in Eq. (9)), which controls the shrinking speed of the feature space. To obtain better convergence leanings, the effective
choice of the parameter m is investigated. In these experiments, the number of iterations and swarm size are set to 10 0 0
and 20, respectively, and the dimension of the search space is set to 30. Also, F1-F4 as four different kinds of functions
are used as the benchmark problems. This experiment performs BWOA with different m values over four functions. The
results of each parameter value are executed 30 times independently and the numerical results are summarized in Table 3.
Fig. 2 also shows the variation of fitness results by changing the values of m. As per results in Table 3 and Fig. 2, we can
see that different landscapes have dissimilar responses to m values. In overall, in terms of the results of ARV, the adjustment
parameter m can be set to 2500 as the best choice.
50 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

Table 3
Outcomes of BWOA with altered m values.

m F1 F2 F3 F4 ARV

100 8.14E−03 1.01E − 04 7.27E − 06 −3.79E − 05 4.89

500 5.08E − 03 1.88E − 04 2.76E − 06 −7.16E − 05 4.15
10 0 0 9.48E − 03 1.11E − 04 2.23E − 06 −5.68E − 05 4.03
1500 1.91E − 03 7.54E − 05 1.82E − 06 −4.86E − 05 3.54
20 0 0 5.22E − 03 1.32E − 04 4.97E − 06 −6.61E − 05 3.95
2500 6.08E − 03 9.57E − 05 9.26E − 07 −7.71E − 05 3.45
30 0 0 5.97E − 03 1.01E − 04 4.55E − 06 −6.77E − 05 3.98

Fig. 2. Variation of results based on m values for different test functions.

3.5. Comparison of BWOA with other methods

In order to verify the efficiency of BWOA, mathematical benchmark tasks are explored by different evolutionary methods
in this section. Furthermore, several well-established methods including the MFO, ALCPSO, CLPSO, jDE, SaDE, and WOA are
all adopted as competitors for these mathematical tasks.
The AVG. index, STD. index and overall ranks of different methods are compared in detail. It is noteworthy that each
ranking rate is calculated using the average results of 30 runs.
The competitive results of algorithms in dealing with four cases are tabulated in Table 4. As per results in Table 4,
we see that the proposed BWOA outperforms other competitors on all four tasks in terms of AVG and STD rates. From
Table 4, one can see that the chaotic BWOA provides the lowest AVG values for all benchmark functions. The locality-
informed BWOA also has a better convergence performance, and it obtains the optimum solution for F1. In addition, the
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 51

Table 4
Comparison of results for different algorithms.

Fun Item MFO CLPSO ALCPSO jDE SaDE WOA BWOA

F1 AVG 3.63E + 01 1.71E + 01 1.94E − 02 1.26E − 08 3.39E − 04 1.04E − 96 0.00E + 00

STD 1.85E + 01 2.66E + 00 5.34E − 02 3.94E − 08 1.54E − 03 3.92E − 96 0.00E + 00
F2 AVG 2.68E + 06 2.39E + 05 6.96E + 01 1.17E + 02 1.52E + 02 2.79E + 01 1.02E − 01
STD 1.46E + 07 9.28E + 04 8.57E + 01 1.96E + 02 8.72E + 01 5.66E − 01 7.60E − 02
F3 AVG 8.81E + 06 7.70E + 02 5.44E − 01 5.00E − 02 6.73E − 01 2.12E − 02 1.49E − 04
STD 4.67E + 07 1.91E + 03 8.15E − 01 1.23E − 01 9.77E − 01 2.17E − 02 5.98E − 05
F4 AVG 2.65E + 00 2.48E − 01 2.02E + 00 9.08E − 01 2.22E − 01 2.34E + 00 3.02E − 04
STD 3.36E + 00 8.01E − 01 2.99E + 00 2.38E + 00 1.22E + 00 3.25E + 00 4.11E − 04
+/=/- 3/1/0 4/0/0 3/1/0 3/1/0 4/0/0 4/0/0 –
ARV 5.4 6.216667 3.920833 2.95 3.654167 3.608333 2.25

Table 5
Best error values in dealing with all functions with high dimensions.

ID D MFO CLPSO ALCPSO jDE SaDE WOA BWOA

F1 100 1.85E + 02 4.98E + 02 3.16E + 01 3.10E + 00 1.12E + 01 5.08E − 95 0.00E + 00

10 0 0 6.55E + 04 6.55E + 04 6.55E + 04 6.55E + 04 1.10E + 03 6.48E − 90 0.00E + 00
20 0 0 6.55E + 04 6.55E + 04 6.55E + 04 6.55E + 04 6.55E + 04 1.78E − 91 0.00E + 00
F2 100 1.18E + 08 1.40E + 08 1.21E + 05 6.65E + 05 1.27E + 05 9.81E + 01 6.00E − 01
10 0 0 1.15E + 10 1.36E + 10 4.04E + 09 8.69E + 08 1.40E + 08 9.95E + 02 8.15E + 00
20 0 0 2.66E + 10 2.99E + 10 1.30E + 10 2.63E + 09 3.74E + 08 1.99E + 03 1.42E + 01
F3 100 2.06E + 08 1.91E + 08 9.24E + 04 4.56E + 05 8.97E + 03 4.31E − 02 1.21E − 04
10 0 0 2.80E + 10 3.30E + 10 8.54E + 09 9.56E + 08 3.19E + 07 1.22E − 01 7.82E − 05
20 0 0 6.55E + 10 7.59E + 10 2.87E + 10 3.39E + 09 9.89E + 07 1.06E − 01 7.81E − 05

proposed method provides the highest quality on F2, F3, and F4. With the help of comparison results, we see that well-
known jDE also performs well for these test tasks, overall.
In addition, Wilcoxon sign rank test [33] is also used to estimate the significant improvements of the BWOA over other
methods. In terms of the “+ / = / − ”, there are statistically significant differences between the results of BWOA and those
from other competitors. According to Wilcoxon’s sign rank test, we can observe that the developed algorithm is significantly
better than MFO, ALCPSO, and jDE on 3 out of 4 functions, and equal to them on one case, respectively. Moreover, the
results of BWOA are significantly better than those obtained by the well-known CLPSO, SaDE, and WOA in dealing with all
functions. The chaos-based BWOA has the best performance among all these competitors from a statistical point of view.
Therefore, it clearly indicates that BWOA is better than other peers. In other words, the proposed BWOA can also be treated
as an effective and efficient auxiliary optimizer for realizing numerical optimization cases.
The average performance of BWOA and other competitive algorithms with the non-parameters statistical Friedman test
is also ranked. As can be seen from Table 4, in terms of ARV index, the chaotic BWOA obtains the lowest average ranking
for these functions, followed by the well-established jDE, WOA, SaDE, ALCPSO, MFO, and CLPSO techniques. According to
the observations, it can be concluded that the chaos-based BWOA can still be the best approach for handling three different
kinds of benchmark functions in 30 dimensions.
To further illustrate the advantages of the proposed BWOA, the convergence behaviors are exhibited in Fig. 3. According
to the convergence trends depicted in Fig. 3, it can be verified that the proposed BWOA can give expressions to a very
fast convergence compared to all other techniques. This observation demonstrates that the LF and CLS can also improve the
convergence tendencies of original WOA. Boxplots of top optimizers are also shown in Fig. 4. The boxplots vividly reveal the
stability and better performance of BWOA compared to other top competitors in Table 4.

3.6. Scalability test

To further study the operators of BWOA and WOA, the scalability test is strictly conducted in this part. The scalability
test can evaluate the impact of dimension on both excellence of results and effectiveness of optimizers, simultaneously.
The influence of dimension on the excellence of results is examined to explore what happens for the success of the MFO,
ALCPSO, CLPSO, jDE, SaDE, WOA, and BWOA, once the dimension of feature space increases, dynamically. Therefore, three
different dimensions are explored here: 100, 1000, and 2000. The same conditions are constructed by utilizing 20 search
agents during 10 0 0 iterations. By increasing the dimensions, the average obtained values in each iteration are tabulated in
Table 5.
It can be seen from Table 5 that not only in low dimensions but also in higher dimensions, the BWOA algorithm can
reveal obvious advantages over other five algorithms as well as the conventional WOA in tackling all functions. WOA holds
similar statistical results compared to the BWOA on all dimensions of F3, which means that the combination of LF and
52 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

Fig. 3. Convergence trends for BWOA versus other optimizers.

CLS can improve the exploratory and exploitative capabilities of WOA, considerably. As an obvious fact, all functions are
more challenging to be solved when the dimensions of search space increase, continuously. As per experimental results in
Table 5, it can be concluded that the BWOA has achieved better results and it has shown better performance than WOA.

3.7. Constrained benchmark problems

In this section, the proposed BWOA is utilized to explain five constrained practical mathematical modeling test cases. Dif-
ferent mathematical problems often possess different constraints; therefore, a proper constraint handling method is needed.
As is described in [34], there are several kinds of penalty functions: co-evolutionary, static, adaptive, dynamic, and death
penalty. Among all these functions, the death penalty is the modest one that constructs a main objective value of the math-
ematical model to be processed and the infeasible solutions can be discarded automatically by the heuristic algorithms.
Therefore, the BWOA with a death penalty function is employed here to tackle three well-known problems in mathematical
modeling studies.

3.7.1. Tension/compression spring design problem

This mathematical modeling of the tension/compression spring design is aimed at minimizing the weight of TCS [1]. In
this modeling, three design variables must be optimized iteratively by the BWOA. These variables are composed of wire
diameter (d), mean coil diameter (D), and the number of active coils (N). The mathematical model for this problem is
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 53

Fig. 4. Boxplots for top optimizers.

formulated as follows:
Consider x = [x1 x2 x3 ] = [d D N]
Objective function :
f (x )min = x21 x2 x3 + 2x21 x2
Subject to
x32 x3
h1 (x ) = 1 − 71785x41
≤ 0,
4x22 −x1 x2
h2 (x ) = + 1
≤ 0,
(
12566 x2 x31 −x41 ) 5180x21
h3 (x ) = 1 − 140x3.45
x
x1
≤ 0,
2 3
x1 +x2
h4 (x ) = 1. 5
− 1 ≤ 0
Variable ranges :
0.05 ≤ x1 ≤ 2.00,
0.25 ≤ x2 ≤ 1.30,
2.00 ≤ x3 ≤ 15.0
This model solved by both mathematical optimization techniques and meta-heuristic methods [1]. He and Wang
[35] has presented using PSO to solve many problems, which can be formulated as constrained problems including ten-
sion/compression spring design. Coello [36] has utilized the GA to realize this problem and the fitness value of 0.0127048
was obtained. The original WOA has also been used for this case. The improved Harmony Search (IHS) [37] algorithm has
been investigated for this case. The RO [38] algorithm has been utilized to solve this task and the experimental results
revealed that the optimum weight can reach up to 0.0126788.
The comparisons of results are demonstrated in Table 6. The statistical results obtained using the BWOA method are
better than other answers described in the literature. The optimization results obtained by BWOA have demonstrated the
fact that the original WOA with LF and CLS strategies possess the obvious advantage over both common meta-heuristic
techniques and mathematical techniques. It can be seen from Table 6 that the original WOA cannot outperform IHS and PSO
54 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

Table 6
Results of BWOA versus other works for TCS design problem.

Method Best variables Best weight

d N D

PSO [35] 0.015728 11.244543 0.357644 0.0126747

GA [36] 0.051480 11.632201 0.351661 1.0127048
RO [38] 0.051370 11.762790 0.349096 0.0126788
IHS [37] 0.051154 12.076432 0.349871 0.0126706
Constraint correction method (Arora) [1] 0.050 0 0 0 14.250 0 0 0 0.315900 0.0128334
Mathematical optimization method (Belegundu) [1] 0.053396 901,854,0 0 0 0.399180 0.0127303
WOA 0.051207 12.0043032 0.345215 0.0126763
BWOA 0.051602 11.2441198 0.357488 0.0126654

on this optimization task, despite the performance of BWOA. Based on the above descriptions, the efficacy of the original
WOA has been vividly improved and the BWOA can be treated as an effective solver for the TCS design in practical cases.

3.7.2. Welded beam design problem

The objective of this model [39] is defined to obtain the minimum value of the fabrication cost for the WB. In this case,
the optimization constraints are bucking load (Pc ), shear stress (τ ), bending stress in the beam (θ ), and deflection rate (δ ).
We have four variables for the fabrication cost. This includes the height of the bar (t), the thickness of the weld (h), the
thickness of the bar (b), and length of the bar (l). The mathematical model of this problem can be described as:

Consider x = [x1 x2 x3 x4 ] = [h l t b]
Objective f (x )min = 1.10471x2 x21 + 0.04811x3 x4 (14.0 + x2 )
Subject to g1 (x ) = τ (x ) − τmax ≤ 0
g2 (x ) = σ (x ) − σmax ≤ 0
g3 (x ) = δ (x ) − δmax ≤ 0
g4 (x ) = x1 − x4 ≤ 0
g5 (x ) = P − PC (x ) ≤ 0
g6 (x ) = 0.125 − x1 ≤ 0
g7 (x ) = 1.10471x21 + 0.04811x3 x4 (14.0 + x2 ) − 5.0 ≤ 0
Variable ranges :
0.1 ≤ x1 ≤ 2,
0.1 ≤ x2 ≤ 10,
0.1 ≤ x3 ≤ 10
0.1 ≤ x4 ≤ 2
where

τ (x ) = (τ  )2 + 2τ  τ  2xR2 + (τ  )2
  
τ = √2Px x , τ  = MR , M = P L + x22 ,
 1 2
 
J
x22 x1 +x3 2
R= + ,
√4 2
 x 1 + x 3 2  
x22
J=2 2x 1 x 2 4
+ 2
,
σ (x ) = 6PL
x4 x3 2
, δ (x ) = 6P L3
Ex23 x4
,

x2 x6  E
4.013E 3 4
PC (x ) = L2
36
1− x3
2L 4G
,
P = 60 0 01b, L = 14in..δmax = 0.25 in..
E = 30 × 16 psi, G = 12 × 106 psi
τmax = 13600 psi, σmax = 30000 psi
This problem has attracted much attention recently. Likewise, Kaveh and Khayatazad adopted the RO [38] to tackle this
task. Kang and Zong [40] used HS to optimize this problem and the experimental results revealed that HS can obtain an op-
timum cost of 2.3807. The improved HS (IHS) [37] has also been explored. These mathematical methods including Davidon-
Fletcher-Powell, Richardson’s random method, and Simplex method have been employed by Radgsdell and Phillips [41].
The results of BWOA are compared with other solutions in the former works in Table 7. As per results in Table 7, the
BWOA obtains an optimum cost. This indicates that the fabrication cost of the WB can be 1.695620 when the four param-
eters are set as 0.20577, 3.2589, 9.028, and 0.20613. It also can be seen that BWOA can outperform all other methods. In
short, the improved WOA, BWOA, compared with other algorithms, can predominantly solve the problem and obtain the
minimum value for fabrication cost of the WB.
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 55

Table 7
Comparison of BWOA with literature for the WB case.

Technique Best variables Best cost

h l t b

RO [38] 0.203687 3.528467 9.004233 0.207241 1.735344

HS [40] 0.2442 6.2231 8.2915 0.2433 2.3807
IHS [37] 0.20573 3.47049 9.03662 0.20573 1.7248
Richardson’s random method [41] 0.4575 4.7313 5.0853 0.6600 4.1185
Simple method [41] 0.2792 5.6256 7.7512 0.2796 2.5307
Davidon–Fletcher–Powell method [41] 0.2434 6.2552 8.2915 0.2444 2.3841
WOA 0.205396 3.484293 9.037426 0.206276 1.730499
BWOA 0.205829 3.251922 9.034556 0.205829 1.695620

Table 8
Results of BWOA versus peers in literature for PV design case.

Algorithm Optimum variables Optimum cost

Ts Th R L

IHS [37] 1.1250 0 0 0.6250 0 0 58.29015 43.69268 7197.7300

PSO [35] 0.812500 0.437500 42.091266 176.746500 6061.0777
GA [43] 0.937500 0.50 0 0 0 0 48.3290 0 0 112.6790 0 0 6410.3811
ES [44] 0.812500 0.437500 42.098087 176.640518 6059.7456
Lagrangian multiplier [42] 1.1250 0 0 0.6250 0 0 58.2910 0 0 43.690 0 0 0 7198.0428
Branch-and-bound [45] 1.1250 0 0 0.6250 0 0 47.70 0 0 0 0 117.710 0 0 8129.1036
WOA 0.812500 0.437500 42.098209 176.638998 6059.7410
BWOA 1.258663 0.621865 65.179120 10.198737 5347.6889

3.7.3. PV design problem

This mathematical modeling is aimed at minimizing the total cost of the cylindrical PV, which is closely related to ma-
terial, forming and welding [42]. The split ends of the PV are covered, whereas the head part has a hemispherical figure. In
PV, the depth of the shell (Ts ) and head (Th ), the internal radius (R), and the extent of the section, minus the head (L), are
variables to be optimized. The formulation can be described as follows:

Consider x = [x1 x2 x3 x4 ] = [Ts Th R L]

Objective : f (x )min = 0.6224Ts RL + 1.7781R2 Th + 3.1661Ts2 L + 19.84Th2 L
Subject to g1 (x ) = −Ts + 0.0193R ≤ 0,
g2 (x ) = −Th + 0.00954R ≤ 0,
g3 (x ) = −π LR2 − 43 π R3 + 12960 0 0 ≤ 0,
g4 (x ) = L − 240 ≤ 0,
Variable ranges :
0 ≤ x1 ≤ 99,
0 ≤ x2 ≤ 99,
10 ≤ x3 ≤ 200,
10 ≤ x4 ≤ 200.

This optimization task has been solved by many meta-heuristic methods. He and Wang [35] proposed using PSO to
solve this problem and the optimum cost reached up to 6061.07777. Deb [43] proposed utilizing GA to try this optimization
problem. The optimum cost of 6410.3811 was fulfilled. In addition, ES [44], IHS [37], and the mathematical methods [42,
45] have been all adopted to treat this task.
The optimization results of reported algorithms in the literature when solving the pressure design are shown in Table 8.
It can be observed from Table 8 that the BWOA obtains a minimum value of 5347.6889, which means that the total cost of
the cylindrical PV can be at a minimum when the TS , Th , R, and L are set as 1.258663, 0.621865, 65.179120, and 10.198737.
Among all of these algorithms, BWOA can find the best feasible optimal design. Therefore, the BWOA can provide powerful
assistance for the PV design.

3.7.4. Three-bar truss design

Three-bar truss design is a structural optimization problem in the field of civil engineering. In order to minimize the
weight subject to stress, deflection, and buckling constraints, the two parameters should be manipulated. The TBTD problem
56 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

Table 9
Comparison results of the TBTD problem.

Method Optimal values for variables Best weight

x1 x2

PSO-DE [46] 0.7886751 0.4082482 263.8958433

MBA [48] 0.7885650 0.4085597 263.8958522
MFO [31] 0.788244770931922 0.40946695784741 263.895979682
DEDS [47] 0.78867513 0.40824828 263.8958434
CS [50] 0.78867 0.40902 263.9716
Ray and Sain [49] 0.795 0.395 264.3
WOA 0.789050544 0.407187512 263.8959474
BWOA 0.788666327 0.408273202 263.8958435

has a difficult constrained search space; therefore, it has been mostly utilized.
Consider x = [x1x2√
] = [A1 A2 ]
Minimize (x ) = 2 2x1 + x2 ∗l
√
Subjectto g1 (x ) = √ 2x 1 + x 2 P −
2x21 +2x1 x2
σ ≤ 0,
g2 (x ) = √ x2
2x21 +2x1 x2
P− σ ≤0
g3 (x ) = √ 1
2x2 +x1
P − ≤ σ 0,
Variable range 0 ≤ x1 , x2 ≤ 1
This problem has been solved in the literature by many researchers as a benchmark optimization problem. Likewise,
Liu et al. [46] employed the hybrid algorithm PSO-DE to solve this problem and the experimental results revealed that the
PSO-DE can obtain an optimum cost of 263.8958433. The differential evolution with dynamic stochastic selection (DEDS)
[47], Mine blast algorithm (MBA) [48], MFO, Ray and Saini [49], and Cuckoo search (CS) [50] have been investigated for this
case. The results of the listed six algorithms are shown in Table 9. In order to evaluate the performance of the BWOA, the
original WOA was also employed for this case. Experiments of the enhanced BWOA and WOA were conducted in the same
programming language and by the same computing platform to make fair comparisons.
Table 9 shows that the BWOA algorithm is able to find a similar structure compared to those of PSO-DE, and DEDS. In
short, it can be observed that the developed BWOA, PSO-DE, and DEDS can achieve to the best place. The best weight of the
TBTD can be 263.8958435 by BWOA when the two parameters are set as 0.788666327 and 0.408273202, while the MBA,
WOA, MFO, CS, and Ray and Sain are in the next places, respectively.

3.7.5. I-beam design problem

Another structural optimization problem of the I-beam design problem is defined to obtain the minimum value of the
vertical deflection. We have four variables for the optimal vertical deflection. This includes length, height, and two thick-
nesses of this problem. Formulation of this problem can be described as follows:
 
Consider x = [x1 x2 x3 x4 x5 ] = b h tw t f
Minimize f (x ) = t (h−2t )3 50
bt 3
00
h−t 2
w
+2bt f ( 2 )
f f f
12 + 6
Subject to g1 (x ) = 2btw + tw h − 2t f ≤ 0
Variable ranges :
10 ≤ x1 ≤ 50,
10 ≤ x2 ≤ 99,
0.9 ≤ x3 ≤ 5,
0.9 ≤ x4 ≤ 5.
This problem was again solved using BWOA and compared to those of Symbiotic Organisms Search (SOS) [51], Improved
ARSM (IARSM) [52], CS and LARSM [52] in the literature. The basic WOA has also been explored, and Table 10 shows the
experimental results on this problem.
It can be observed from Table 10 that the BWOA obtains a minimum value of 0.00625958, which means that the total
cost of IBD problem can be at a minimum when the b, h, tw , and tf are set as 50, 80, 1.76470588 and 5, respectively. Based on
the above descriptions, the BWOA performs effectively when solving this problem and it provides the best optimal design.
The basic WOA has been vividly improved and the BWOA can be treated as an effective solver for the IBD design in practical
cases.
The effectiveness of BWOA in dealing with other constrained problems is influenced by the numerical method utilized
for constraint handling. Among all possible constraint handling approaches, exterior penalty-based techniques are the most
effective. Among this family, those methods that work based on adaptive, automatic and dynamic updating schemes es-
tablish the most promising results for tackling constrained optimization problems. One open direction is to validate the
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 57

Table 10
Comparison results of BWOA with literature for the IBD problem.

Algorithm Optimal values for variables Optimum cost

b h tw tf

MFO [31] 50 80 1.7647 5 0.0066259

SOS [51] 50 80 0.9 2.32179 0.0130741
ARSM [52] 37.05 80 1.71 2.31 0.0157
CS [50] 50 80 0.9 2.321675 0.0130747
IARSM [52] 48.42 79.99 0.9 2.4 0.131
WOA 49.99799 80 1.7647477 5 0.00662619
BWOA 50 80 1.76470588 5 0.00625958

effectiveness of the BWOA by checking other constraint handling schemes. Study of different constraint handling methods,
improvements and extensions will be done in future works and it is beyond the scope of this work.

4. Discussion on results

In this study, two novel effective strategies, LF and CLS, are synchronously introduced to the WOA to promote the inclu-
sive exploration and confined exploitation features of the basic version and to thoroughly investigate the searching capability
of both balanced BWOA and basic WOA. In addition, the experimental results obtained by BWOA are compared to the other
common optimizers on a set of benchmark cases and the results can credibly verify that the LF and CLS used in BWOA can
improve the core characteristics of conventional WOA. With regard to the comparison results in Section 3, several reasons
can be accounted for the observations where the BWOA has fulfilled competitive and better solutions in dealing with three
different kinds of tasks.
Based on LF strategy in BWOA, it can enhance the possibility of exploration trends to search different regions of the so-
lution space, uniformly, and then, gradually convert from exploration to exploitation during the whole optimization process.
Therefore, the LF can offer assistance to BWOA in possessing a better equilibrium between the inclusive and neighborhood-
informed searching trends in dealing with studied problems. Nevertheless, the studied optimizers without LF such as MFO,
DA, WOA, and PSO, cannot perform well. In terms of the LO stagnation, the LF-motivated searching agents and time-varying
chaotic patterns of BWOA can assist current agents in escaping from the LO and stagnation disadvantages. Furthermore,
the agents in BWOA can perform their searching missions based on various exploratory and exploitative patterns because
these search agents can shift toward the best and perform well based on some random operators and various LF-triggered
trajectories. Based on CLS in BWOA, the best agent is guided towards better vicinities and the diversity of the population is
strengthened. As it is shown in Fig. 3, the efficiency and robustness of the BWOA, with their improved convergence curves in
contrast to other optimizers, is successful. A proper reason can account for this phenomenon that the introduced strategies
have contributed to BWOA to enable it for upholding a reasonable equilibrium between the diversification and intensifica-
tion proclivities. Furthermore, the special movements of LF, called triggered logarithmic shrinking spiral-shaped motions, can
further enhance the above-described exploration and exploitation patterns of BWOA. Based on the above descriptions and
analyses, the LF in BWOA has a constructive effect, which can lead to a better balance between global exploratory and local
exploitative trends and the higher capability in escaping from the LO stagnation. As another merit of BWOA, we observed
that the CLS also assists the BWOA to fulfill a reasonable balance between the diversification and intensification inclinations
during the whole optimization process. Chaotic local search patterns have guided the leader and enriched the exploitation
tendencies of the BWOA in last steps and make it more exploitative in the case of finding a high-quality solution. To sum-
marize, the WOA has two core characteristics that should be balanced to experience an efficient performance: exploration
and exploitation. Initially, the whale optimizer has an intensive exploration operator, but it suffers from the lake of a sta-
ble balance between its core searching trends. Chaotic local search has assisted the base method in guiding the leader and
therefore, it boosts the exploitation propensities of the proposed method in last steps. Chaos-based pseudo-random patterns
also help the BWOA to have more varied jumping lengths during the exploitation process. The LF-based operator also en-
riches the exploitation characteristics of the whale optimizer. In BWOA, only improved results are recorded to be enriched
in the next iterations, which also enhance the accuracy of the search. Therefore, the BWOA can be used as an effective and
efficient auxiliary tool for real-world optimization tasks in practical cases.
We see that the BWOA has possessed competitive and progressive results in comparison with other optimizers in this
study. A fact that cannot be ignored is that the BWOA may still have more room to be competitive enough with more
current optimizers, as it can be understood from “no free lunch” (NFL) theorem [53]. Therefore, the performance of the
BWOA should be thoroughly investigated for more problems such as IEEE CEC 2017 test cases.

5. Conclusion and future works

The WOA algorithm has shown an acceptable efficacy in dealing with a wide range of application areas. However, in
terms of the global exploration or local exploitation capacity of whale optimizer itself, there is much room for improvement
58 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59

in many practical cases. In this study, the enhanced whale optimizer with several new operators is proposed. The BWOA
algorithm employs the Lévy flight mechanism and chaotic local search strategy, which makes the conventional whale op-
timizer to not only have a better search capability but also faster convergence rates. The BWOA was tested on a well-
studied set of benchmark problems in comparison with several other optimization algorithms. The strict experimental re-
sults demonstrate that the BWOA has outperformed other competitors, which means that the exploratory and exploitative
capacities of the basic whale optimizer have been strengthened, significantly. In addition, The BWOA also was utilized to
study five practical problems in mathematical modelings. The final experimental results show that the BWOA can supply
powerful assistance for three mathematical designs and it has the potential to be very effective in tackling more practical
problems with complex search spaces as well.
In future works, there are lots of aspects in WOA, which worth to be further explored. On one hand, the proposed method
may be applied to more practical cases to exert its full effective merits, especially in the areas of engineering optimization
control. On another hand, the proposed method can be reconstructed as a fusion procedure with other meta-heuristics, and
can also be explored as a multi-objective algorithm based on Pareto theory in order to further investigate its performance,
thoroughly. Furthermore, a binary BWOA version can also be proposed for solving discrete optimization tasks. In addition,
another feasible future work is to propose a parallel computing framework based on BWOA. The proposed method can deal
with more sophisticated and difficult constrained optimization problems with other constraint handling methods. In short,
the searching behaviors of the WOA still deserves further exploration and how to apply this proposal into more application
scenarios also needs to be ceaselessly tried.

Acknowledgments

This research is funded by the Science and Technology Plan Project of Wenzhou of China under Grant no. ZG2017019,
Zhejiang Provincial Natural Science Foundation of China under Grant No. LY17F020012, Guangdong Natural Science Foun-
dation under Grant No. 2018A030313339, Characteristic Innovation Projects of Universities in Guangdong under Grant no.
2017GKTSCX063, MOE (Ministry of Education in China) Youth Fund Project of Humanities and Social Sciences Grant No.
17YJCZH261, National Natural Science Foundation of China Grant No. 61471133 and No. 61871475. The authors would like to
thank Ali Asghar Heidari for revising the manuscript.

References

[1] S. Mirjalili, A. Lewis, The whale optimization algorithm, Adv. Eng. Softw. 95 (2016) 51–67.
[2] M.M. Mafarja, S. Mirjalili, Hybrid whale optimization algorithm with simulated annealing for feature selection, Neurocomputing 260 (2017) 302–312.
[3] X. Zhao, X. Zhang, Z. Cai, X. Tian, X. Wang, Y. Huang, H. Chen, L. Hu, Chaos enhanced grey wolf optimization wrapped ELM for diagnosis of
paraquat-poisoned patients, Comput. Biol. Chem. 78 (2019) 481–490.
[4] Q. Zhang, H. Chen, J. Luo, Y. Xu, C. Wu, C. Li, Chaos enhanced bacterial foraging optimization for global optimization, IEEE Access 6 (2018)
64905–64919.
[5] J. Luo, H. Chen, Q. zhang, Y. Xu, H. Huang, X. Zhao, An improved grasshopper optimization algorithm with application to financial stress prediction,
Appl. Math. Model. 64 (2018) 654–668.
[6] R. Abbassi, A. Abbassi, A.A. Heidari, S. Mirjalili, An efficient salp swarm-inspired algorithm for parameters identification of photovoltaic cell models,
Energy Convers. Manag. 179 (2019) 362–372.
[7] H. Faris, M.M. Mafarja, A.A. Heidari, I. Aljarah, A.-Z. Ala’M, S. Mirjalili, H. Fujita, An efficient binary Salp Swarm Algorithm with crossover scheme for
feature selection problems, Knowl. Based Syst. 154 (2018) 43–67.
[8] A.A. Heidari, R.A. Abbaspour, A.R. Jordehi, An efficient chaotic water cycle algorithm for optimization tasks, Neural Comput. Appl. 28 (2017) 57–85.
[9] A.A. Heidari, H. Faris, I. Aljarah, S. Mirjalili, An efficient hybrid multilayer perceptron neural network with grasshopper optimization, Soft Comput.
(2018) 1–18, doi:10.10 07/s0 050 0- 018- 3424- 2.
[10] M. Mafarja, I. Aljarah, A.A. Heidari, A.I. Hammouri, H. Faris, A.-Z. Ala’M, S. Mirjalili, Evolutionary population dynamics and grasshopper optimization
approaches for feature selection problems, Knowl. Based Syst. 145 (2018) 25–45.
[11] W. Deng, H. Zhao, L. Zou, G. Li, X. Yang, D. Wu, A novel collaborative optimization algorithm in solving complex optimization problems, Soft Comput.
21 (2017) 4387–4398.
[12] W. Deng, H. Zhao, X. Yang, J. Xiong, M. Sun, B. Li, Study on an improved adaptive PSO algorithm for solving multi-objective gate assignment, Appl.
Soft Comput. 59 (2017) 288–302.
[13] W. Deng, R. Yao, H. Zhao, X. Yang, G. Li, A novel intelligent diagnosis method using optimal LS-SVM with improved PSO algorithm, Soft Comput. (2017)
1–18, doi:10.10 07/s0 050 0- 017- 2940- 9.
[14] Y. Ling, Y. Zhou, Q. Luo, Lévy flight trajectory-based whale optimization algorithm for global optimization, IEEE Access 5 (2017) 6168–6186.
[15] A.A. Heidari, P. Pahlavani, An efficient modified grey wolf optimizer with Lévy flight for optimization tasks, Appl. Soft Comput. 60 (2017) 115–134.
[16] M.Z. Ali, N.H. Awad, R.G. Reynolds, P.N. Suganthan, A balanced fuzzy Cultural Algorithm with a modified Levy flight search for real parameter opti-
mization, Inf. Sci. 447 (2018) 12–35.
[17] H. Sharma, J.C. Bansal, K.V. Arya, Opposition based lévy flight artificial bee colony, Memet. Comput. 5 (2013) 213–227.
[18] J. Xie, Y. Zhou, H. Chen, A novel bat algorithm based on differential operator and Lévy flights trajectory, Comput. Intell. Neurosci. 2013 (2013) 1–13.
[19] M. Wang, H. Chen, B. Yang, X. Zhao, L. Hu, Z.N. Cai, H. Huang, C. Tong, Towards an optimal kernel extreme learning machine using a chaotic moth-flame
optimization strategy with applications in medical diagnoses, Neurocomputing 267 (2017) 69–84.
[20] B. Alatas, E. Akin, A.B. Ozer, Chaos embedded particle swarm optimization algorithms, Chaos, Solitons Fractals 40 (2009) 1715–1734.
[21] D. Jia, G. Zheng, M. Khurram Khan, An effective memetic differential evolution algorithm based on chaotic local search, Inf. Sci. 181 (2011) 3175–3187.
[22] S.M. Hosseini Bamakan, H. Wang, T. Yingjie, Y. Shi, An effective intrusion detection framework based on MCLP/SVM optimized by time-varying chaos
particle swarm optimization, Neurocomputing 199 (2016) 90–102.
[23] H. Zhao, S. Guo, H. Zhao, Energy-related CO2 emissions forecasting using an improved LSSVM model optimized by whale optimization algorithm,
Energies 10 (2017) 1–15.
[24] W.N. Chen, J. Zhang, Y. Lin, N. Chen, Z.H. Zhan, H.S.H. Chung, Y. Li, Y.H. Shi, Particle swarm optimization with an aging leader and challengers, IEEE
Trans. Evolut. Comput. 17 (2013) 241–258.
[25] J.J. Liang, A.K. Qin, P.N. Suganthan, S. Baskar, Comprehensive learning particle swarm optimizer for global optimization of multimodal functions, IEEE
Trans. Evolut. Comput. 10 (2006) 281–295.
 H. Chen, Y. Xu and M. Wang et al. / Applied Mathematical Modelling 71 (2019) 45–59 59

[26] J. Brest, S. Greiner, B. Bošković, M. Mernik, V. Zumer, Self-adapting control parameters in differential evolution: a comparative study on numerical
benchmark problems, IEEE Trans. Evolut. Comput. 10 (2006) 646–657.
[27] A.K. Qin, P.N. Suganthan, Self-adaptive differential evolution algorithm for numerical optimization, Evolutionary Computation, in: Proceedings of the
IEEE Congress, 1782, 2005, pp. 1785–1791.
[28] H. Haklı, H. Uğuz, A novel particle swarm optimization algorithm with Levy flight, Appl. Soft Comput. J. 23 (2014) 333–345.
[29] X. Li, P. Niu, J. Liu, Combustion optimization of a boiler based on the chaos and lèvy flight vortex search algorithm, Appl. Math. Model. 58 (2018) 3–18.
[30] X.S. Yang, Flower pollination algorithm for global optimization, in: Proceedings of the International Conference on Unconventional Computation and
Natural Computation, 2013, pp. 240–249.
[31] S. Mirjalili, Moth-flame optimization algorithm: a novel nature-inspired heuristic paradigm, Knowl. Based Syst. 89 (2015) 228–249.
[32] S. Mirjalili, Dragonfly algorithm: a new meta-heuristic optimization technique for solving single-objective, discrete, and multi-objective problems,
Neural Comput. Appl. 27 (2016) 1053–1073.
[33] S. García, A. Fernández, J. Luengo, F. Herrera, Advanced nonparametric tests for multiple comparisons in the design of experiments in computational
intelligence and data mining: experimental analysis of power, Inf. Sci. 180 (2010) 2044–2064.
[34] C.A.Coello Coello, Theoretical and numerical constraint-handling techniques used with evolutionary algorithms: a survey of the state of the art, Com-
put. Methods Appl. Mech. Eng. 191 (2002) 1245–1287.
[35] Q. He, L. Wang, An effective co-evolutionary particle swarm optimization for constrained engineering design problems, Eng. Appl. Artif. Intell. 20
(2007) 89–99.
[36] C.A.C. Coello, Use of a Self-Adaptive Penalty Approach for Engineering Optimization Problems (20 0 0).
[37] M. Mahdavi, M. Fesanghary, E. Damangir, An improved harmony search algorithm for solving optimization problems, Appl. Math. Comput. 188 (2007)
1567–1579.
[38] A. Kaveh, M. Khayatazad, A new meta-heuristic method: ray optimization, Comput. Struct. 112–113 (2012) 283–294.
[39] C.A.C. Coello, Use of a self-adaptive penalty approach for engineering optimization problems, Comput. Ind. 41 (20 0 0) 113–127.
[40] S.L. Kang, W.G. Zong, A new meta-heuristic algorithm for continuous engineering optimization: harmony search theory and practice, Comput. Methods
Appl. Mech. Eng. 194 (2005) 3902–3933.
[41] K.M. Ragsdell, D.T. Phillips, Optimal design of a class of welded structures using geometric programming, J. Eng. Ind. 98 (1976) 97.
[42] B.K. Kannan, S.N. Kramer, An augmented Lagrange multiplier based method for mixed integer discrete continuous optimization and its applications to
mechanical design, J. Mech. Des. 116 (1994) 405–411.
[43] K. Deb, GeneAS: A Robust Optimal Design Technique for Mechanical Component Design, Springer Berlin Heidelberg, 1997.
[44] E. Mezura-Montes, C.A.C. Coello, An empirical study about the usefulness of evolution strategies to solve constrained optimization problems, Int. J.
Gen. Syst. 37 (2008) 443–473.
[45] E. Sandgren, Nonlinear integer and discrete programming in mechanical design optimization, J. Mech. Des. 112 (1990) 223–229.
[46] H. Liu, Z. Cai, Y. Wang, Hybridizing particle swarm optimization with differential evolution for constrained numerical and engineering optimization,
Appl. Soft Comput. 10 (2010) 629–640.
[47] M. Zhang, W. Luo, X. Wang, Differential evolution with dynamic stochastic selection for constrained optimization, Inf. Sci. 178 (2008) 3043–3074.
[48] A. Sadollah, A. Bahreininejad, H. Eskandar, M. Hamdi, Mine blast algorithm: a new population based algorithm for solving constrained engineering
optimization problems, Appl. Soft Comput. 13 (2013) 2592–2612.
[49] T. Ray, P. Saini, Engineering design optimization using a swarm with an intelligent information sharing among individuals, Eng. Optim. 33 (2001)
735–748.
[50] A.H. Gandomi, X.-S. Yang, A.H. Alavi, Cuckoo search algorithm: a metaheuristic approach to solve structural optimization problems, Eng. Comput. 29
(2013) 17–35.
[51] M.-Y. Cheng, D. Prayogo, Symbiotic organisms search: a new metaheuristic optimization algorithm, Comput. Struct. 139 (2014) 98–112.
[52] G.G. Wang, Adaptive response surface method using inherited latin hypercube design points, J. Mech. Des. 125 (2003) 210–220.
[53] D.H. Wolpert, W.G. Macready, No free lunch theorems for optimization, IEEE Trans. Evolut. Comput. 1 (1997) 67–82.