Energy Reports 7 (2021) 228–246

Contents lists available at ScienceDirect

Energy Reports
journal homepage: [Link]/locate/egyr

Delayed dynamic step shuffling frog-leaping algorithm for optimal

design of photovoltaic models
∗
Yi Fan a , Pengjun Wang a ,1 , , Ali Asghar Heidari b,c ,2 , Xuehua Zhao d , Hamza Turabieh e ,
∗
Huiling Chen f ,1 ,
a
College of Electrical and Electronic Engineering, Wenzhou University, Wenzhou 325035, China
b
School of Surveying and Geospatial Engineering, College of Engineering, University of Tehran, Tehran 1417466191, Iran
c
Department of Computer Science, School of Computing, National University of Singapore, Singapore 117417, Singapore
d
School of Digital Media, Shenzhen Institute of Information Technology, Shenzhen 518172, China
e
Department of Information Technology, College of Computers and Information Technology, P.O. Box 11099, Taif 21944, Taif University, Taif, Saudi
Arabia
f
College of Computer Science and Artificial Intelligence, Wenzhou University, Wenzhou, Zhejiang 325035, China

article info a b s t r a c t

Article history: Developing an accurate mathematical model is an essential tool for studying and optimizing the
Received 9 September 2020 performance of the photovoltaic cell system (PV). Transforming the PV problem into an optimization
Received in revised form 29 November 2020 problem in which meta-heuristic algorithms excel provides an alternative approach to identifying the
Accepted 7 December 2020
PV model’s parameters. The memetic evolution mechanism and the shuffling strategy included in the
Available online 21 December 2020
shuffling frog-leaping algorithm (SFLA) provide safeguards for solving nonlinear, multimodal, and high-
Keywords: dimensional problems. However, the low convergence accuracy is an essential drawback of the SFLA
Swarm-intelligence algorithm to solve the PV problem. This paper proposes a delayed dynamic step mechanism based
Shuffled frog-leaping algorithm on the SFLA algorithm’s characteristics to overcome this disadvantage, called the DDSFLA algorithm.
Delayed dynamic step mechanism The results of testing 23 benchmark functions and extracting the single diode model, the double-
PV models
diode model, and the PV module show that the DDSFLA algorithm has a faster convergence speed and
Parameter extraction
higher convergence accuracy and exhibits strong optimization stability under the special conditions
of different temperatures or light intensities. The results suggest that the proposed algorithm can be
used as an effective method to handle PV models’ parameter extraction. The extra resources and online
user guidance for this research will be provided at [Link]
© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license ([Link]

1. Introduction directly (Su et al., 2019). However, as a side effect, the energy
disaster, shortage, and ecological greenhouse gasses have encour-
Energy, information science, and biotechnology can be consid- aged us to strive for new clean energy (Guo et al., 2020). As there
ered the main supports of contemporary life in both west and is a galaxy, solar energy is the most available, pollution-free, and
east of the world (Peng et al., 2020; Chen et al., 2017, 2018). The most potent energy source instead of oil and other fossil-based
matter of fact is that any form of energy is the most ultimate resources and to avoid climate change (Wang et al., 2020c; Zhang
and vital requirement of human life (Cheng et al., 2016) and the et al., 2020c). The modern world has become inextricably linked
basis of the technology developed countries such as China and to fossil fuels (Zhu et al., 2018a). The emergence of a number of
Singapore (Liu et al., 2019). In any industry, from transportation environmental problems such as global warming (He et al., 2018),
to food factories, energy consumption in different forms, such air pollution, species extinction, and extreme weather (Chen et al.,
as gas or electricity (Liu et al., 2020a), happens indirectly or 2020a) has led to the realization that fossil fuels’ overuse may
be the root cause of this situation. To continue to meet people’s
∗ Corresponding authors. energy needs on the premise of environmental protection, the
E-mail addresses: 184511088183@[Link] (Y. Fan), photovoltaic cell system (PV) that converts solar energy into
wangpengjun@[Link] (P. Wang), as_heidari@[Link],
electricity has received widespread attention.
aliasgha@[Link], t0917038@[Link] (A.A. Heidari),
zhaoxh@[Link] (X. Zhao), [Link]@[Link] (H. Turabieh), Any mathematical problem has some unknown variables and
[Link]@[Link] (H. Chen). some constraint (at least bound of variables) that we need to
1 These authors contributed equally to this work. develop an efficient decision-making model (Liu et al., 2016; Liu
2 [Link] and Liu, 2020; Wu et al., 2020a; Zeng et al., 2020; Liu et al.,

[Link]
2352-4847/© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license ([Link]
nc-nd/4.0/).
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

2020c) as an outcome of optimization attempts. Such models and the advantage of simple structure and ability to avoid local
decisions are not only limited to energy field but also dependent optima, and effectively tackling optimization cases with a harsh
topics and technologies such as fuzzy optimization (Chen et al., landscape, these techniques are useful to be applied to many ar-
2019b), robust optimization (Qu et al., 2020), traffic manage- eas (Sun et al., 2019b; Song et al., 2020; Wang et al., 2020b; Wang
ment (Liu et al., 2020b), neural networks and deep learning (Li and Chen, 2020; Xu et al., 2019; Zhao et al., 2019; Wang et al.,
et al., 2019a; Mousavi et al., 2020; Ni et al., 2020; Qian et al., 2017; Shen et al., 2016). Therefore, to reach higher quality results
2020a,b; Qiu et al., 2019; Xu et al., 2018; Yang et al., 2019; and more exciting performances, it is accepted to propose more
Zhang et al., 2020b), multiobjective optimization (Cao et al., 2019, problem-centric enhanced and progressive single-objective and
2020b), location-based services (LBS) (Li et al., 2019b), wireless many-objective approaches to augment their efficacy in dealing
sensor networks (Fu et al., 2020b,a, 2019), honeynet potency (Ren with various real-world problems (Cao et al., 2020d).
et al., 2020), monitoring (Li et al., 2020a; Sun et al., 2018, 2019a; Eusuff et al. (2006) simulate frog populations’ natural behavior
Wen et al., 2017; Xie et al., 2018; Zhang et al., 2019a), computer- searching for food to create the shuffled frog-leaping algorithm
aided design (Singh et al., 2011), and IoT (Lv and Xiu, 2020). For (SFLA) to address water distribution system design issues and
dealing with these models, there are two logical ways we need to groundwater model calibration problems. The local search mech-
choose: exact and deterministic solvers that deal with complete anism and the global information exchange strategy included
information of the surface of the variable space (Ran et al., 2020; in the SFLA algorithm provide safeguards for solving nonlinear,
Xiong et al., 2020; Zeng et al., 2019b,a; Zhu et al., 2018b) or multimodal and high dimensional problems. Such as: static syn-
requisite gradient and subgradient statistics (Long et al., 2015; chronous compensator system (Hasanien, 2014), optimal bidding
Shi et al., 2020a; Zhang et al., 2019b), and other way is to uti- strategy problem (Kumar and Kumar, 2014), channel equalization
lize approximated and algorithms with stochastic operations that problem (Panda et al., 2014), time-cost-resource optimization
perform according to a projected qualitative metric of the target problems (Ashuri and Tavakolan, 2015), two-agent hybrid flow
problems (Zhao and Li, 2020). Designing accurate mathematical shop scheduling problem (Lei and Guo, 2015), task scheduling
models is an important tool for studying and optimizing the problems (Tripathy et al., 2015), flexible job shop scheduling
performance of the photovoltaic cell system. The single diode problem (Lei et al., 2017), and spatial modeling of landslide
model (SDM) and the double diode model (DDM) are the most susceptibility (Chen et al., 2019a).
widely used practical application models. Since both the SDM Although the SFLA algorithm has been successfully applied, it
and the DDM are implicit transcendental equations, proposing shows poor optimization when dealing with more multifaceted
a method that can accurately extract the model parameters has problems. Therefore, some researchers have proposed different
become a research interest. conceptions to overcome the SFLA algorithm’s shortcomings to
Many researchers have proposed numerous techniques for solve the problem better. Niknam et al. (2011) proposed a strong
identifying model parameters for the PV system’s nonlinear mutation that increases population diversity to ensure that the
current–voltage characteristics. For example, a nonlinear least- SFLA algorithm solves the optimal power flow problem. Bhat-
squares optimization algorithm (Easwarakhanthan et al., 1986). tacharjee and Sarmah (2014) used a small probability operator
In some cases, it is not feasible to reach acceptable and practical to improve the SFLA algorithm’s ability to avoid premature mat-
results with complex spaces throughout a reasonable time using uration, thus solving knapsack problems. Luo and Chen (2014)
general mathematical approaches (Cao et al., 2020c). The strong enabled the SFLA algorithm to better handle multi-depots ve-
performance of meta-heuristic algorithms, as a stochastic-based hicle routing problems, adding a power-law extremal optimiza-
branch of methods, in the field of optimization problems provides tion neighborhood search mechanism. Jadidoleslam and Ebrahimi
another way to identify PV models’ parameters. Meta-heuristic (2015) addressed the reliability constrained generation expansion
algorithms are a class of techniques proposed for optimization planning problem by adding a deviation angle to expand the local
problems based on natural phenomena (Cao et al., 2020a; Yang search space of the SFLA algorithm. Zhou et al. (2015) introduced
and Chen, 2019; Chen et al., 2020b). The abundance of natural chaos theory and clone selection into the SFLA algorithm to
phenomena provides a steady stream of ideas for researchers. reduce the peak-to-average power ratio. The cascade reservoir
The ease with which the core concepts can be understood and operation optimization problem was solved by a hybrid cloud
the ease with which the code can be written makes it possible model algorithm and SFLA algorithm (Sun et al., 2016). Zhang
to apply the algorithm to a wide range of practical problems. A et al. (2016) proposed the concept of Gaussian variation and
wide variety of meta-heuristic algorithms have been proposed random-beating free frogs to improve the global search perfor-
in the literature for the problem of parameter identification mance of the SFLA algorithm. Zhao et al. (2016) introduced chaos
of PV models (Ridha et al., 2021; Chen et al., 2019c; Abbassi operators because of its fundamental behaviors (Wang et al.,
et al., 2020; Zhang et al., 2020d; Liu et al., 2020d; Wang et al., 2020) and convergence factors in SFLA to solve the problem of
2020a). Examples of these methods are not limited to genetic bearing early fault diagnosis. They used chaos because of its
algorithm (GA) (Jervase et al., 2001), particle swarm optimization property that it has a sensitive state to the initial condition, and
(PSO) (Kiani et al., 2020), differential evolution (DE) (Liang et al., by this merit, it augments sampled patterns’ data (Shi et al.,
2020), bacterial foraging optimizer (BFO) (Rajasekar et al., 2013), 2017b; Wu et al., 2019, 2020b; Shi et al., 2018, 2017a; Xiong et al.,
Harris hawks optimization (HHO)3 (Chen et al., 2020d; Jiao et al., 2016). Sharma and Pant (2017) combined an opposition-based
2020; Ridha et al., 2020), slime mould algorithm (SMA)4 (Li learning mechanism with an SFLA algorithm to implement wire-
et al., 2020b) backtracking search algorithm (BSA) (Zhang et al., less networks’ optimal location management problem. Hsu and
2020a). We can describe them as emerging evolutionary com- Chiang (2019) proposed four mechanisms to optimize the SFLA
puting tools for stochastic optimization, which has inspired re- algorithm: (1) multiple group strategies, (2) shuffling strategies,
searchers’ widespread attention for handling the optimization (3) self-adaptive jumping, and (4) adaptive variation strategies.
case (Cao et al., 2020e; Liu et al., 2015). These methods have Hany Hasanien (2015) applied the SFLA algorithm to SDM
parameter identification. However, the basic SFLA algorithm was
3 The info of HHO algorithm, its background, and codes are available for found to be prone to premature maturation. When the num-
public use at [Link] ber of search iterations is low, the local search mechanism and
4 The info of SMA algorithm, its background, and codes are available for the global information exchange strategy enable the SFLA algo-
public use at [Link] rithm to exhibit faster convergence. However, when increasing
229
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

the number of iterations by a certain amount, the SFLA algo-

rithm is no longer capable of further improving the convergence
accuracy. Compared to the results in other literature, the SFLA
algorithm has no more significant advantages. Therefore, this
paper proposes a delayed dynamic step mechanism based on
the characteristics of the SFLA algorithm. The concept of this
mechanism is to wait for the SFLA algorithm to search iteratively
for a certain number of times and then combine it with a dynamic
step mechanism to search further for the global optimal solution,
thus achieving further improvement in convergence accuracy.
The innovations throughout the paper can be summarized as
Fig. 1. The equivalent circuit of double-diode model.
follows:

• Proposing delay dynamic step mechanism for shuffling frog-

leaping algorithm.
• The variant SFLA algorithm performs better than some
swarm-intelligence algorithms.
• The variant SFLA algorithm exhibits strong robustness to PV
models under different conditions.

The paper can be divided into the following sections: Section 2

constructs the characteristics of the SDM, DDM, and PV modules
and the adaptation functions for evaluating the performance of
the algorithms; the original SFLA algorithm and the new DDS-
FLA algorithm are introduced in Section 3; Section 4 provides
the experimental results and specific analysis of the benchmark Fig. 2. The equivalent circuit of the single-diode model.
functions and the PV problem, and the summary and future
perspectives are presented in Section 5.

2. Photovoltaic model description

In practical applications, a single-diode model (SDM) and

double-diode model (DDM) are commonly used to describe the
nonlinear voltage–current characteristics of photovoltaic systems.
This section describes the properties of each of these two models.

2.1. Property

The equivalent circuit of the DDM is depicted in Fig. 1. The

Fig. 3. Photovoltaic cell module equivalent circuit.
output current (IL ) of the DDM is related to the photo generated
current (Iph ), the two diode currents (Id1 , Id2 ), and the shunt
resistor current (Ish ), as in Eq. (1) (Wolf and Rauschenbach, 1963):
q · (VL + RS · IL )
[ ( ) ]
V L + R S · IL
− Isd2 · exp −1 − (5)
n2 · k · T RSh
IL = Iph − Id1 − Id2 − ISh (1)
From Eq. (5), it can be seen that there are seven parameters
where, according to Shockley’s equation (Shockley and Field,
(Iph , Isd1 , Isd2 , Rsh , RS , n1 , n2 ) to be extracted from the DDM.
1952), the diffusion current (Isd1 ), the saturation current (Isd2 ),
the output voltage (VL ), IL , the diffusion diode ideal factor(n1 ), The equivalent circuit of the SDM is depicted in Fig. 2. SDM is
the recombination diode ideal factor (n2 ), the series resistance a simpler structure than the DDM (Humada et al., 2016). The IL is
(RS ), the Boltzmann’s constant (k = 1.3806503×10^(−23) J/K), as in Eq. (6):
the cell temperature in Kelvin (T ), and the electron charge (q =
q · (VL + RS · IL )
[ ( ) ]
1.60217646×10^(−19) C) make up the Id1 and Id2 , as in Eqs. (2) V L + R S · IL
IL = Iph − Isd · exp −1 − (6)
and (3): n·k·T RSh
q · (VL + RS · IL )
[ ( ] )
Id1 = Isd1 · exp −1 (2) From Eq. (7), it can be seen that the SDM has five parameters (Iph ,
n1 · k · T Isd , RS , Rsh , n) that need to be identified.
q · (VL + RS · IL )
[ ( ) ]
The equivalent circuit of the PV module is depicted in Fig. 3.
Id2 = Isd2 · exp −1 (3)
n2 · k · T PV module is made up of multiple solar cells connected in series
The shunt resistor current (Ish ) is calculated from the VL , IL , RS , and parallel. NP represents the number of parallel strings, and
and shunt resistance (Rsh ), as in Eq. (4): NS represents the number of cells in series, as expressed in Eq.
V L + R S · IL (7) (Soon and Low, 2012; AlRashidi et al., 2011):
Ish = (4)
q · (VL /NS + RS · IL /NP )
[ ( ) ]
RSh
IL /NP = Iph − Isd · exp −1
Take Eqs. (2), (3), and (4) into Eq. (1) to derive the output n·k·T
current expression (5) (Humada et al., 2016): VL /NS + RS · IL /NP
− (7)
q · (VL + RS · IL )
[ ( ) ]
RSh
IL = Iph − Isd1 · exp −1
n1 · k · T
230
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

2.2. Fitness function to step 4.5 if the current new location has a better fitness value;
otherwise, go to step 4.3.
The meta-heuristic algorithm’s task is to extract the parame-
ters of the SDM and the DDM to minimize the error between the S = min[rand ∗ (PB − PW ), Smax ] (11)
estimated current and the measured current. Thus the root mean S = max[rand ∗ (PB − PW ), −Smax ] (12)
square error (RMSE) can be defined as a fitness function, as in Eq.
(8) (Xu and Wang, 2017): newposition1 = PW + S (13)

 N
1 ∑ where rand denotes a random number obeying a uniform distri-
RMSE (X) = √ F (VL , IL , X )2 (8) bution within the range [0, 1], Smax denotes the maximum jump
N
i=1 step.
where, N represents the number of estimated data. Step 4.3: The worst host position is updated according to
For SDM, F (VL , IL , X ) is expressed as Eq. (9): Eqs. (14), (15), and (16) using the global optimal performance
[ ( ) ] of the host information. Substitute and jump to step 4.5 if the
q·(VL +RS ·IL )
⎧
⎪
⎨ FSDM ( VL , I L , X ) = I ph − I sd · exp n·k·T
− 1 current new location has a better fitness value; otherwise, go to
(9) step 4.4.
− VL +R RS ·IL − IL
Sh
⎪
X = (Iph , Isd , RS , Rsh , n);
⎩
S = min[rand ∗ (PX − PW ), Smax ] (14)
For DDM, F (VL , IL , X ) is expressed as Eq. (10): S = max[rand ∗ (PX − PW ) , −Smax ] (15)
newposition2 = PW + S
[ ( ) ]
q·(VL +RS ·IL )
⎧
⎨ FDDM (VL , IL , X ) = Iph − Isd1 · exp
⎪ −1 (16)
⎪ n1 ·k·T

(10)
[ ( ) ]
−Isd2 · exp q·(VnL +·kR·TS ·IL ) − 1 − VL +R RS ·IL − IL Step 4.4: A random approach is used to update the worst host
2 Sh
⎪
location.
⎪
X = (Iph , Isd1 , Isd2 , Rsh , RS , n1 , n2 )
⎩
Step 4.5: Reorder the n hosts and jump back to step 4.1 until
3. Specific improvement ideas for SFLA the condition of the maximum number of iterations N is satisfied.
Step 5: Reorder the updated populations and renew the opti-
3.1. Basic concept of SFLA mal host position PX .
Step 6: Determine if the search stop condition has been
Eusuff et al. (2006) proposed a memetic meta-heuristic algo- reached; otherwise, return to step 3.
rithm called a shuffled frog-leaping algorithm (SFLA). Similar to
the other stochastic solvers (Cao et al., 2020f), there is a nature-
Algorithm 1 Pseudo-code of DDSFLA algorithm
inspired idea behind this method. Frogs are considered hosts of
Initialization: upper bound (ub), lower bound (lb), the
memes that exchange information with others to change their
maximum number of iterations (T ), dimension (dim), fitness
rocks’ location to search the most food-rich areas in the swamp.
function (fobj), number of memeplexes (m), number of frogs
First, a population of frogs is randomly generated and sorted into
(n), population size (F=m*n), number of selecting frogs (q),
multiple memeplexes in the swamp. Second, the frogs in each
number of memetic evolution iterations (N1 , N2 ), constant (α ,
memeplex are selected by triangular probability distributions to
β ), the impact factor (γ ), delay time (D),
form new submemeplexes. The host, with the worst performance
1. for i = 1 : F
in each submemeplex optimizes its meme and changes its stone’s
2. positioni,dim (= unifrnd(lb), ub, dim);
position by exchanging information with the best performing
3. Costi = fobj positioni,dim ;
host in the entire population or the best performing host in the
4. end for
submemeplex. Finally, all submemeplexes are shuffled to form a
5. [New Cost , Order] = sort (Cost ) ;
new population after completing memetic evolution and contin-
6. position = postion (Order ) ;
ues to seek out the most food-rich areas of the swamp iteratively.
7. PX = postion (1) ; BestCost = New Cost (1) ;
The essential steps are as follows:
8. While t < T
Step 1: Within the search area, the population F is randomly
9. if t < D delayTime = 0; else delayTime = 1; endif
generated.
10. The position is divided into m memeplexes;
Where F = m × n represents the number of memeplexes, n
11. for j = 1 : m
denotes the number of hosts.
12. P = 2 ∗ n∗(n+1) ) ;
n+1−(1 : n
Step 2: The fitness values of all memes are calculated and then
sorted. The location PX of the global optimal performance host is 13. for iN1 = 1 : N1
saved. 14. f or i = 1 : qQ (i) = randsample(n, 1, true, P); end
Step 3: The population F is divided into m memeplexes based 15. New Position = memeplexj (Q ); New MeCt =
on the ranking results. memeplexCostj (Q );
Step 4: All memeplexes perform memetic evolution within the 16. for iN2 = 1 : N2
number of N iterations. 17. [New Cost2, Order2] = sort(New MeCt);
Step 4.1: The hosts are assigned a weight based on the triangu- 18. New Position = New Position(Order2);
lar probability distribution, i.e. Pj = 2(n + 1 − j)/n(n + 1), j = 1, 19. PB = New Position (1) ; PW = New Position (end) ; PW Cost =
New Cost2 (end) ;
. . . , n. The q frogs are randomly selected from n hosts depended on
20. flag1 = false; flag2 =[ false; ( β )]
weights to form submemeplex, and the current best-performing
host PB and worst-performing host PW are recorded. 21. S new = delayTime ∗ α − t ∗ T ∗ S old + γ ∗ rand∗
Step 4.2: The worst host position is updated according to (PB − PW ) ;
Eqs. (11), (12), and (13) using the host information of the best 22. New PW 1 = PW + S new ;
performance in the current submemeplex. Substitute and jump
231
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

23. PW 1 Cost = fobj (New PW 1 ) ; Table 1

24. if PW 1 Cost < PW CostPW = New PW 1 ; PW Cost < The information of unimodal functions.
Number fmin Bound Formula
PW 1 Cost ; elseflag1 = ture; end ∑n
25. if flag1 F1 0 [−100, 100] f1 (x) = i=1 x2i
( β )]
26. S new = delayTime ∗ α − t ∗ T ∗ S old + γ ∗ rand∗
[ ∑n ∏n
F2 0 [−10, 10] f2 (x) = i=1 | xi | + i=1 |xi |
(PX − PW ) ; F3 0 [−100, 100] f3 (x) =
∑n ∑i 2
i=1 ( xj )
27. New PW 2 = PW + S new ; j−1

28. PW 2 Cost = fobj (New PW 2 ) ; F4 0 [−100, 100] f4 (x) = maxi (|xi | , 1 ≤ i ≤ n)

29. if PW 2 Cost < PW CostPW = New PW 2 ; PW Cost < F5 0 [−30, 30] f5 (x) =
∑n−1
i=1 [100(xi+1 − x2i )2 + (xi − 1)2 ]
PW 2 Cost ; elseflag2 = ture; end ∑n
F6 0 [−100, 100] f6 (x) = i=1 ( [xi + 0.5])2
30. endif ∑n
31. if flag2PW = unifrnd (lb, ub) ; PW Cost = fobj (PW ) ; end F7 0 [−1.28, 1.28] f7 (x) = i=1 ix4i + random[0, 1)
32. New Position (end) = PW ; New Cost2 (end) = PW Cost ;
33. New MeCt = New Cost2; S old = snew ;
34. endfor 4. Experimental testing and analysis
35. memeplexj (Q ) = New Position; memeplexCostj (Q )
= NewMeCt ; 4.1. Benchmark functions
36. endfor
37. endfor
38. position = memeplexes1 : m ; Cost = memeplexCost1 : m ; 23 benchmark functions are used to evaluate the optimization
39. [New Cost , Order ] = sort (Cost ) ; performance of the DDSFLA algorithm. The 23 benchmark func-
40. position = postion (Order ) ; tions are divided into F1–F7 for unimodal functions (in Table 1),
41. PX = postion (1) ; BestCost = New Cost (1) ; F8–F13 for multi-modal functions (in Table 2), and F14–F23 for
42. t = t + 1; fixed-dimensional multimodal functions (in Table 3). For com-
43. end while parison, the SFLA (Eusuff et al., 2006), IJAYA (Yu et al., 2017),
PSO (Kennedy and Eberhart, 1995), WOA (Mirjalili and Lewis,
3.2. DDSFLA 2016), SSA (Mirjalili et al., 2017), SCA (Mirjalili, 2016b), DA (Mir-
jalili, 2016a), MFO (Mirjalili, 2015), FA (Yang, 2010a), and BA
In this subsection, a delayed dynamic step mechanism is pro- (Yang, 2010b) algorithms are chosen.
posed to improve the convergence accuracy of the SFLA algorithm Wilcoxon signed-rank test (Derrac et al., 2011) at 0.05 sig-
for dealing with the PV problem. Multiple memeplexes and shuf- nificance level and Freidman’s test (García et al., 2010) are used
fling approaches provide the exploration of the SFLA algorithm. for assessing the algorithm’s optimization performance. Also, the
Each memeplex performs memetic evolution to ensure the ex- mean and standard deviation of 30 tests are normalized to the [0,
ploitation capability of the SFLA algorithm. The SFLA algorithm’s 1] range to enable comparison of test results.
characteristics provide faster convergence in the initial search To ensure fairness of the experiment (Chen et al., 2020c; Lv
stages, but the convergence accuracy cannot be further improved. and Qiao, 2020; Shi et al., 2020b), the following are listed as
Therefore, (1): the first step is to ensure that the SFLA algorithm generic parameter settings. The maximum number of evaluations
adopts its approach to search by reserving the initial search is 20,000, the population size is 30, each algorithm has been inde-
phase. (2): a dynamic step mechanism is added in the part of pendently tested 30 times, and the running software is MATLAB
updating the host with the worst performance in each submeme- R2016a. The parameters involved in the comparison algorithm
plex to improve the convergence accuracy of the original SFLA are set according to the original paper. The parameters m, n, q,
algorithm. The specific formula (17) is as follows: N of the original SFLA algorithm are consistent with the DDSFLA
algorithm. Table 4 lists the parameters of the DDSFLA algorithm.
β
[ ( )]
S new = delaytime ∗ α − t ∗ ∗ S old +γ ∗ rand ∗ (P − PW ) (17) In Table 5, the Avg and Std results obtained by optimizing 23
T functions are normalized to evaluate the overall performance of
where α and β are constants, t represents the current number of the algorithm. The best (0) represents the number of normalized
iterations, T denotes the maximum number of iterations, γ means function results with a best value 0, and worst (1) denotes the
the impact factor, rand is a random number that obeys a uniform number of normalized function results with a worst value 1. The
distribution between 0 and 1, P indicates the global optimal host DDSFLA algorithm has the highest number of best (0) for Avg
location or the current submemeplex optimal host location. The and Std, while the SFLA, PSO, DA, and BA algorithms have none.
dynamic step mechanism introduces information from the old IJAYA results show the average and sum of Avg and Std. Both
step into the current step and uses the impact factor γ to amplify are minimums. WOA and SSA exhibit approximate optimization
the influence of other good hosts on the worst host, thereby capabilities.
improving the convergence accuracy of the SFLA algorithm. The As can be seen in Table 6, the DDSFLA algorithm ranks first.
pseudo-code is shown in Algorithm 1, and the flowchart is shown ‘‘+/-/=’’ indicates the number of better, worse, and similar calcu-
in Fig. 4. After initializing the necessary parameters, a population lations for the DDSFLA algorithm compared to other algorithms.
of frogs is generated uniformly over the search space. The fitness Compared to SFLA, PSO, SSA, SCA, DA, FA, and BA, the number of
value of the population is calculated and ordered according to its ‘‘+’’ is more than 15.
size. The search proceeds to an iterative phase after saving the The convergence curves in Fig. 5 show that the DDSFLA algo-
optimal individual information. Once the delay time is set, the rithm converges with higher accuracy for the unimodal functions
population is divided into m clusters. Follow the core part of the F1, F4, and F5. The results for the multimodal functions F8, F11,
SFLA algorithm to find a better population location. When the F13 indicate that the DDSFLA algorithm is in a local dilemma.
delay condition is satisfied, the individuals are further searched The convergence curves of the fixed-dimensional multimodal
using Eq. (17). Until the search stop condition is met, the final functions F15, F17, F20 present that the DDSFLA algorithm has
result is output. better convergence performance.
232
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 2
The information of multimodal functions.
ID fmin Bound Formula
∑n √
F8 −418.9829*5 [−500, 500] f8 (x) = i=1 −xi sin( |xi |)
∑n
F9 0 [−5.12, 5.12] f9 (x) = [x2i − 10 cos (2π xi ) + 10]
i=1
( √ ∑ )
n
f10 (x) = −20exp −0.2 1n 2
− exp( 1n ni=1 cos(2π xi )) + 20 + e
∑
F10 0 [−32, 32] i=1 xi

∑n ∏n ( )
x
F11 0 [−600, 600] f11 (x) = 1
4000 i=1 x2i − i=1 cos √i +1
i
{ } ∑
π
∑n−1
f12 (x) = 10 sin (π y1 ) + (yi − 1)2 1 + 10 sin2 (π yi+1 ) + (yn − 1)2 + ni=1 u(xi , 10, 100, 4)
[ ]
F12 0 [−50, 50] n i=1

∑n ∑n
f13 (x) = 0.1 sin2 (3π x1 ) + − 1)2 [1 + sin2 (3π xi + 1)] + (xn − 1)2 [1 + sin2 (2π xn )] + u(xi , 5, 100, 4)
{
F13 0 [−50, 50] i=1 (xi i=1

Table 3
The information of fix-dimension multimodal functions.
ID fmin Bound Formula
( )−1
∑25
F14 1 [−65, 65] f14 (x) = 1
+ 1
500 j=1 j+∑2
i=1 xi −aij (
6
)
∑11 x1 (bi 2 +bi x2 ) 2
F15 0.00030 [−5, 5] f15 (x) = i=1 [ai − ]
bi 2 +bi x3 +x4

F16 −1.0316 [−5, 5] f16 (x) = 4x1 − 2.1x1 +

2 4 1
x 6
3 1
+ x1 x2 − 4x2 2 + 4x2 4
( )2
5.1
f17 (x) = x2 − 5 1
( )
F17 0.398 [−5, 5] 4π 2
x1 2 + x
π 1
−6 + 10 1 − 8π
cos x1 + 10

f (x) = 1 + (x1 + x2 + 1)2 19 − 14x1 + 3x1 2 − 14x2 + 6x1 x2 + 3x2 2 × [30 + (2x1 − 3x2 )2 ×
[ ( )]
F18 3 [−2, 2]
(18 2 2
)
18 − 32x1 + 12x1 + 48x2 − 36x1 x2 + 27x2 ]
∑4 ∑3
F19 −3.86 [1, 3] f19 (x) = − ci exp(− 2
j=1 aij (xj − pij ) )
i=1

∑4 ( ∑6 )2 )
f20 (x) = − i=1 ci exp − j=1 aij xj − pij
(
F20 −3.32 [0, 1]
∑5 [ ]−1
F21 −10.1532 [0, 10] f21 (x) = − i=1 (X − ai ) (X − ai )T + ci
∑7 [ ]−1
F22 −10.4028 [0, 10] f22 (x) = − i=1 (X − ai ) (X − ai )T + ci
∑10 [ ]−1
F23 −10.5363 [0, 10] f23 (x) = − i=1 (X − ai ) (X − ai )T + ci

4.2. PV parameter extraction problem Table 4

The parameters of the DDSFLA algorithm.
Item Parameter
The DDSFLA algorithm is tested separately on two sets of
data (Easwarakhanthan et al., 1986) and three sets of experi- Memeplexes (m) 3
Frogs (n) 10
mental data from the manufacturer’s datasheet (Merchaoui et al.,
Selecting frogs (q) 6
2018) to verify the delay dynamic step mechanism’s effect on Memetic evolution iterations (N1 , N2 ) 5/3
the original SFLA algorithm in dealing with the PV problem. The Impact factor (γ ) 2
first data set is a 26-pair current–voltage value measured by a Constant (α , β ) 0.5/0.3
Delay time (D) 1000
commercial 57-mm diameter RTC France silicon solar cell at an
irradiance of 1000 W/m2 and temperature of 33 ◦ C. The DDSFLA
algorithm extracts SDM and DDM parameters on the first set of
data, respectively. The second data set is the 25 current–voltage and current values and the estimated current value, estimated
pairs measured by a Photo-watt-PWP 201 PV panel of 36 poly- power value, relative error, and absolute error. Fig. 6 plots the
crystalline silicon cells in series with an irradiance of 1000 W/m2 convergence curves for all algorithms.
and temperature of 45 ◦ C. The DDSFLA algorithm extracts the In Table 8, the fitness value (9.8630E−04) obtained by the
parameters of the PV module on the second set of data. The three DDSFLA algorithm is smaller than any other algorithm. The four
data sets include SM55, KC200GT, and ST40 at different levels of
metrics results after 30 independent tests of all algorithms in
irradiance and temperature. The DDSFLA algorithm extracts SDM
Table 9 indicate that the DDSFLA algorithm performs more con-
and DDM parameters on experimental data. To demonstrate the
sistently than other algorithms. As can be seen from Table 10, the
competitiveness of the DDSFLA algorithm, it is compared with
estimated data of the SDM identified by the DDSFLA algorithm
basic SFLA, IJAYA, GOTLBO (Chen et al., 2016), PSO, CS (Yang and
Deb, 2009), WOA, SSA, SCA, GWO (Mirjalili et al., 2014), and DA. are consistent with the measured data. The absolute error for
each group of currents does not exceed 2.5089E−03. The relative
4.2.1. Identify single-diode model on the RTC France error for currents is between −1.9990E−02 and 1.4375E−01. The
Table 7 contains the range of values for the parameters to DDSFLA algorithm in Fig. 6 maintains the same search speed as
be identified in the SDM. Table 8 lists the accurate parameter the basic SFLA algorithm at the initial search stage. However, as
values extracted by all algorithms and the fitness values. The the number of evaluations increases, the delayed dynamic step-
four metrics to assess the performance of the algorithm are ping mechanism helps the host converge to a more high-precision
shown in Table 9. Table 10 contains the measured data’s voltage region.
233
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 4. Flowchart of DDSFLA algorithm.

Table 5
The normalized results of eleven algorithms.
DDSFLA SFLA IJAYA PSO WOA SSA
Avg Std Avg Std Avg Std Avg Std Avg Std Avg Std
Average 0.1943 0.3488 0.5644 0.7226 0.0174 0.0187 0.4153 0.3367 0.0844 0.1915 0.0804 0.1945
Sum 4.4697 8.0219 12.9806 16.6202 0.3999 0.4298 9.5511 7.7431 1.9418 4.4043 1.8494 4.4742
Best (0) 8 7 0 0 7 6 0 0 2 2 2 3
Worst (1) 1 5 4 7 0 0 4 4 0 0 0 0
SCA DA MFO FA BA
Avg Std Avg Std Avg Std Avg Std Avg Std
Average 0.1975 0.1052 0.3133 0.3672 0.1071 0.2466 0.5578 0.3539 0.3830 0.3885
Sum 4.5419 2.4187 7.2054 8.4457 2.4622 5.6708 12.8291 8.1397 8.8099 8.9346
Best (0) 2 2 0 0 6 6 0 1 0 0
Worst (1) 3 0 2 2 0 0 8 2 1 3

Table 6
The results of the Wilcoxon signed rank test and Freidman’s test.
DDSFLA SFLA IJAYA PSO WOA SSA SCA DA MFO FA BA
+/ − / = ∼ 19/2/2 10/10/3 17/3/3 9/7/7 14/4/5 15/5/3 17/3/3 7/2/14 17/3/3 17/2/4
ARV 2.8457 8.2812 4.2826 8.4942 4.2645 4.1341 5.4891 8.0507 3.2130 8.9449 8
Rank 1 9 5 10 4 3 6 8 2 11 7

234
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 5. Convergence curves of eleven algorithms on 23 benchmark functions.

Table 7 algorithm. Table 14 contains the measured data and the estimated
The range of parameters (model: SDM, data: RTC France). data. Fig. 7 maps the convergence curves for all algorithms.
SDM RTC France As shown in Table 12, the DDSFLA algorithm evaluates the
lb ub optimal adaptation value (9.8434E−04) of the DDM compared
Iph (A) 0 1 to other algorithms. The statistical results of the four metrics
Isd (µA) 0 1 in Table 13 show that the DDSFLA algorithm has not only high
RS () 0 0.5
Rsh () 0 100
search accuracy but also good stability. In Table 14, the maximum
n 1 2 current absolute error is 2.519E−03 and the current relative error
range between −2.0069E−02 and 1.4564E−01. The convergence
curves in Fig. 7 show that the DDSFLA algorithm exhibits a strong
convergence performance compared to other algorithms.
4.2.2. Identify double-diode model on the RTC France
Table 11 contains the range of values for the seven parameters
4.2.3. Identify the photovoltaic cell module on the Photo-watt-PWP
to be identified in the DDM. Table 12 shows the specified parame- 201
ters extracted by each algorithm as well as the adaptation values. The range of values for the PV module parameters is different
Table 13 presents four indicators to assess the performance of the from both the SDM and DDM, listed in Table 15. Table 16 provides

Table 8
The accurate parameters and fitness values (model: SDM, data: RTC France).
Iph (A) Isd (µA) RS () Rsh () n RMSE
DDSFLA 0.7608 0.3191 0.0364 53.3770 1.4800 9.8630E−04
SSA 0.7607 0.3189 0.0364 54.0730 1.4799 9.8705E−04
IJAYA 0.7607 0.3295 0.0363 55.4290 1.4832 9.8871E−04
GOTLBO 0.7608 0.3102 0.0365 51.9710 1.4771 9.8997E−04
PSO 0.7607 0.3505 0.0361 56.0570 1.4894 9.9826E−04
GWO 0.7601 0.3253 0.0365 61.5080 1.4818 1.0741E−03
CS 0.7609 0.2860 0.0366 51.3020 1.4691 1.1593E−03
WOA 0.7603 0.2964 0.0374 64.4360 1.4723 1.4339E−03
SFLA 0.7602 0.3212 0.0374 75.3620 1.4802 1.8791E−03
DA 0.7620 0.9463 0.0313 57.4660 1.5988 2.8764E−03
SCA 0.7762 0.2657 0.0384 44.3620 1.4560 1.8246E−02

235
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 6. The convergence of DDSFLA and other algorithms (model: SDM, data: Fig. 7. The convergence of DDSFLA and other algorithms (model: DDM, data:
RTC France). RTC France).

Table 9 Table 11
Four kinds of evaluation indexes (model: SDM, data: RTC France). The range of parameters (model: DDM, data: RTC France).
Algorithm max min mean std DDM RTC France
DDSFLA 1.3056E−03 9.8630E−04 1.0819E−03 8.6464E−05 lb ub
SFLA 1.7469E−01 1.8791E−03 8.4421E−02 5.3543E−02 Iph (A) 0 1
IJAYA 1.5539E−03 9.8871E−04 1.1202E−03 1.5379E−04 Isd1 (µA) 0 1
GOTLBO 1.3951E−03 9.8997E−04 1.1486E−03 1.3573E−04 Isd2 (µA) 0 1
PSO 3.8151E−02 9.9826E−04 9.3253E−03 1.4666E−02 RS () 0 0.5
CS 1.8041E−03 1.1593E−03 1.4137E−03 1.6466E−04 Rsh () 0 100
WOA 5.2786E−02 1.4339E−03 1.9243E−02 1.7361E−02 n1 1 2
SSA 8.5066E−03 9.8705E−04 3.1614E−03 2.3006E−03 n2 1 2
SCA 2.2291E−01 1.8246E−02 5.5944E−02 4.6079E−02
GWO 4.4505E−02 1.0741E−03 1.2458E−02 1.5504E−02
DA 5.7047E−02 2.8764E−03 2.0513E−02 1.9840E−02

From the results in Tables 16 and 17, evaluating the PV

module, the DDSFLA algorithm has the highest accuracy, the
the exact parameters extracted by all algorithms, along with the GOTLBO algorithm has the smallest mean value, and the CS
algorithm has the optimal standard deviation and maximum
adaptation values. Table 17 presents the results of the four eval- value. Table 18 indicates that the current’s absolute error does
uation metrics for all algorithms. Table 18 shows the measured not exceed 4.8715E−03, with a relative error range of between
−4.2430E−02 and 5.0677E−02. As can be seen from the conver-
data and the estimated data. Fig. 8 plots the convergence curves
gence curve in Fig. 8, the DDSFLA algorithm exhibits the fastest
for all algorithms. convergence rate.

Table 10
Measured data and estimated data (model: SDM, data: RTC France).
Measured data Estimated current data Estimated power data
Vm Im Ie IAEI REI Pe IAEP REP
1 −0.2057 0.7640 0.7641 1.1543E−04 −1.5106E−04 −0.1572 2.3744E−05 −1.5108E−04
2 −0.1291 0.7620 0.7627 6.8169E−04 −8.9380E−04 −0.0985 8.8006E−05 −8.9460E−04
3 −0.0588 0.7605 0.7614 8.6553E−04 −1.1368E−03 −0.0448 5.0893E−05 −1.1381E−03
4 0.0057 0.7605 0.7602 3.4347E−04 4.5184E−04 0.0043 1.9578E−06 4.5164E−04
5 0.0646 0.7600 0.7591 9.4925E−04 1.2506E−03 0.0490 6.1322E−05 1.2490E−03
6 0.1185 0.7590 0.7580 9.6846E−04 1.2776E−03 0.0898 1.1476E−04 1.2760E−03
7 0.1678 0.7570 0.7571 7.5209E−05 −9.9341E−05 0.1270 1.2620E−05 −9.9351E−05
8 0.2132 0.7570 0.7561 8.7990E−04 1.1637E−03 0.1612 1.8759E−04 1.1623E−03
9 0.2545 0.7555 0.7551 4.3793E−04 5.7999E−04 0.1922 1.1145E−04 5.7965E−04
10 0.2924 0.7540 0.7536 3.6258E−04 4.8110E−04 0.2204 1.0602E−04 4.8087E−04
11 0.3269 0.7505 0.7514 8.6569E−04 −1.1522E−03 0.2456 2.8299E−04 −1.1535E−03
12 0.3585 0.7465 0.7473 8.3363E−04 −1.1155E−03 0.2679 2.9886E−04 −1.1167E−03
13 0.3873 0.7385 0.7401 1.6066E−03 −2.1708E−03 0.2866 6.2224E−04 −2.1755E−03
14 0.4137 0.7280 0.7274 6.1455E−04 8.4488E−04 0.3009 2.5424E−04 8.4417E−04
15 0.4373 0.7065 0.7070 4.9153E−04 −6.9523E−04 0.3092 2.1494E−04 −6.9572E−04
16 0.4590 0.6755 0.6753 1.8703E−04 2.7695E−04 0.3100 8.5847E−05 2.7688E−04
17 0.4784 0.6320 0.6308 1.2016E−03 1.9049E−03 0.3018 5.7485E−04 1.9013E−03
18 0.4960 0.5730 0.5720 1.0337E−03 1.8073E−03 0.2837 5.1271E−04 1.8040E−03
19 0.5119 0.4990 0.4996 6.3350E−04 −1.2679E−03 0.2558 3.2429E−04 −1.2695E−03
20 0.5265 0.4130 0.4137 6.5628E−04 −1.5865E−03 0.2178 3.4553E−04 −1.5890E−03
21 0.5398 0.3165 0.3175 9.9710E−04 −3.1405E−03 0.1714 5.3823E−04 −3.1504E−03
22 0.5521 0.2120 0.2121 1.2504E−04 −5.8945E−04 0.1171 6.9033E−05 −5.8979E−04
23 0.5633 0.1035 0.1022 1.2855E−03 1.2577E−02 0.0576 7.2414E−04 1.2421E−02
24 0.5736 −0.0100 −0.0087 1.2568E−03 1.4375E−01 −0.0050 7.2091E−04 1.2568E−01
25 0.5833 −0.1230 −0.1255 2.5089E−03 −1.9990E−02 −0.0732 1.4634E−03 −2.0398E−02
26 0.5900 −0.2100 −0.2084 1.5634E−03 7.5006E−03 −0.1230 9.2241E−04 7.4448E−03

236
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 12
The accurate parameters and fitness values (model: DDM, data: RTC France).
Iph (A) Isd1 (µA) RS () Rsh () n1 Isd2 (µA) n2 RMSE
DDSFLA 0.7608 0.2931 0.0365 54.3710 1.4730 0.2271 2.0000 9.8434E−04
GOTLBO 0.7607 0.2427 0.0365 54.9130 1.6685 0.1972 1.4479 9.8708E−04
IJAYA 0.7607 0.2656 0.0366 52.7410 1.4664 0.0955 1.6935 9.9029E−04
PSO 0.7606 0.3495 0.0361 61.3690 1.4897 0.0950 2.0000 1.0412E−03
SSA 0.7603 0.2341 0.0370 59.8600 1.6764 0.1848 1.4400 1.0676E−03
GWO 0.7612 0.0926 0.0367 53.7900 1.3912 0.8229 1.7544 1.0761E−03
CS 0.7610 0.8913 0.0369 55.8260 1.9266 0.1952 1.4411 1.2656E−03
WOA 0.7606 0.0106 0.0345 63.9400 1.9999 0.4795 1.5222 1.2946E−03
SFLA 0.7632 0.4549 0.0338 46.3390 1.5373 0.2159 1.6308 2.5690E−03
DA 0.7618 0.0753 0.0374 63.8180 1.3684 0.4456 1.6901 3.7966E−03
SCA 0.7797 0.0932 0.0355 78.0850 1.6436 0.8180 1.5857 1.4297E−02

Table 13
Four kinds of evaluation indexes (model: DDM, data: RTC France).
Algorithm max min mean std
DDSFLA 1.4725E−03 9.8434E−04 1.1071E−03 1.3014E−04
SFLA 2.6409E−01 2.5690E−03 1.1049E−01 6.3386E−02
IJAYA 1.7056E−03 9.9029E−04 1.1448E−03 1.7796E−04
GOTLBO 2.1566E−03 9.8708E−04 1.3389E−03 3.1055E−04
PSO 3.8151E−02 1.0412E−03 1.1268E−02 1.5156E−02
CS 3.2133E−03 1.2656E−03 2.2676E−03 5.0373E−04
WOA 8.7552E−02 1.2946E−03 1.7165E−02 2.1022E−02
SSA 6.9348E−03 1.0676E−03 3.4591E−03 1.5350E−03
SCA 2.2287E−01 1.4297E−02 4.7143E−02 3.5150E−02
GWO 5.1522E−02 1.0761E−03 1.2993E−02 1.5527E−02
DA 6.5302E−02 3.7966E−03 2.2022E−02 1.9586E−02

4.2.4. CPU time Fig. 8. The convergence of DDSFLA and other algorithms (model: PV, data:
This section tests the CPU time consumed by 11 algorithms Photo-watt-PWP 201).
on SDM, DDM, and PV to measure the computational cost of the
DDSFLA algorithm on the PV identification problem. Table 19 and
Fig. 9 record the average computational time for all algorithms on 4.2.5. Analysis of the number of different populations
the three models.
It can be seen from the data that GWO, SCA, and IJAYA al- Under the condition that the number of evaluations decreases,
gorithms spend the least amount of time on SDM, DDM, and and the population increases slightly, this section tests the algo-
PV, respectively. Compared to the SFLA algorithm, the DDSFLA
rithm’s performance to identify photovoltaic model parameters.
algorithm takes less time on SDM and close to 25 s on both DDM
and PV. The experiments are divided into a total of 5 groups, and the

Table 14
Measured data and estimated data (model: DDM, data: RTC France).
Measured data Estimated current data Estimated power data
Vm Im Ie IAEI REI Pe IAEP REP
1 −0.2057 0.7640 0.7641 4.5505E−05 −5.9558E−05 −0.1572 9.3604E−06 −5.9562E−05
2 −0.1291 0.7620 0.7626 6.3796E−04 −8.3651E−04 −0.0985 8.2360E−05 −8.3721E−04
3 −0.0588 0.7605 0.7614 8.4578E−04 −1.1109E−03 −0.0448 4.9732E−05 −1.1121E−03
4 0.0057 0.7605 0.7602 3.4138E−04 4.4909E−04 0.0043 1.9459E−06 4.4889E−04
5 0.0646 0.7600 0.7591 9.2769E−04 1.2221E−03 0.0490 5.9929E−05 1.2206E−03
6 0.1185 0.7590 0.7581 9.3013E−04 1.2270E−03 0.0898 1.1022E−04 1.2255E−03
7 0.1678 0.7570 0.7571 1.2674E−04 −1.6740E−04 0.1271 2.1267E−05 −1.6743E−04
8 0.2132 0.7570 0.7562 8.2002E−04 1.0844E−03 0.1612 1.7483E−04 1.0832E−03
9 0.2545 0.7555 0.7551 3.7650E−04 4.9859E−04 0.1922 9.5818E−05 4.9834E−04
10 0.2924 0.7540 0.7537 3.0812E−04 4.0882E−04 0.2204 9.0095E−05 4.0865E−04
11 0.3269 0.7505 0.7514 9.0383E−04 −1.2029E−03 0.2456 2.9546E−04 −1.2043E−03
12 0.3585 0.7465 0.7474 8.4698E−04 −1.1333E−03 0.2679 3.0364E−04 −1.1346E−03
13 0.3873 0.7385 0.7401 1.5905E−03 −2.1491E−03 0.2866 6.1602E−04 −2.1537E−03
14 0.4137 0.7280 0.7273 6.5877E−04 9.0572E−04 0.3009 2.7253E−04 9.0491E−04
15 0.4373 0.7065 0.7069 4.2849E−04 −6.0612E−04 0.3091 1.8738E−04 −6.0649E−04
16 0.4590 0.6755 0.6752 2.5507E−04 3.7774E−04 0.3099 1.1708E−04 3.7760E−04
17 0.4784 0.6320 0.6307 1.2596E−03 1.9970E−03 0.3018 6.0259E−04 1.9930E−03
18 0.4960 0.5730 0.5719 1.0704E−03 1.8715E−03 0.2837 5.3090E−04 1.8680E−03
19 0.5119 0.4990 0.4996 6.2267E−04 −1.2463E−03 0.2558 3.1874E−04 −1.2478E−03
20 0.5265 0.4130 0.4137 6.6878E−04 −1.6167E−03 0.2178 3.5211E−04 −1.6193E−03
21 0.5398 0.3165 0.3175 1.0253E−03 −3.2289E−03 0.1714 5.5344E−04 −3.2394E−03
22 0.5521 0.2120 0.2122 1.5816E−04 −7.4547E−04 0.1171 8.7319E−05 −7.4603E−04
23 0.5633 0.1035 0.1022 1.2585E−03 1.2309E−02 0.0576 7.0892E−04 1.2160E−02
24 0.5736 −0.0100 −0.0087 1.2712E−03 1.4564E−01 −0.0050 7.2918E−04 1.2712E−01
25 0.5833 −0.1230 −0.1255 2.5190E−03 −2.0069E−02 −0.0732 1.4694E−03 −2.0480E−02
26 0.5900 −0.2100 −0.2085 1.5378E−03 7.3768E−03 −0.1230 9.0729E−04 7.3227E−03

237
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 15 Table 17
The range of parameters (model: PV, data: Photo-watt-PWP 201). Four kinds of evaluation indexes (model: PV, data: Photo-watt-PWP 201).
PV module Photowatt-PWP 201 Algorithm max min mean std
lb ub DDSFLA 2.9281E−03 2.4252E−03 2.4974E−03 1.1000E−04
Iph (A) 0 2 SFLA 2.8482E+00 8.7187E−03 4.7818E−01 6.7841E−01
Isd (µA) 0 50 IJAYA 5.2319E−03 2.4255E−03 2.5694E−03 5.0979E−04
RS () 0 2 GOTLBO 2.8962E−03 2.4253E−03 2.4898E−03 1.0046E−04
Rsh () 0 2000 PSO 3.1526E−01 2.4566E−03 1.6608E−01 1.3611E−01
n 1 50 CS 2.6101E−03 2.4344E−03 2.5225E−03 3.8370E−05
WOA 7.8391E−01 2.4793E−03 1.1750E−01 1.7488E−01
SSA 3.0125E−01 2.4386E−03 2.3932E−02 7.3523E−02
SCA 2.7436E−01 1.0742E−02 1.8422E−01 1.0624E−01
GWO 7.2697E−02 2.6009E−03 6.1604E−03 1.2658E−02
DA 4.3836E−01 6.2886E−03 6.6814E−02 8.8488E−02

data for changes in temperature or irradiance under certain con-

ditions. When the temperature or irradiance dynamically differs,
the range of the model parameters also changes, as shown in
Table 21. ISC is calculated from Eq. (18), where T and G denote dy-
namic temperature and irradiance, respectively, α represents the
temperature factor of ISC , and the subscript STC means standard
conditions.
G
ISC (G, T ) = ISCSTC × + α × (T − TSTC ) (18)
GSTC
Table 22 contains the accurate parameters and fitness values
of the SDM and DDM extracted by the DDSFLA algorithm on
the ST40 dataset at an irradiance of 1000 W/m2 at different
temperatures of 25 ◦ C, 40 ◦ C, 50 ◦ C, and 70 ◦ C, respectively.
At a temperature of 70 ◦ C, the DDSFLA algorithm estimates the
smallest fitness values for SDM and DDM than for other tem-
Fig. 9. The curves of CPU time of eleven algorithms on SDM, DDM, PV.
peratures. The fitness values of DDM are better than those of
SDM for irradiance of 600 W/m2 , 400 W/m2 , and 200 W/m2 at
a temperature of 25 ◦ C, in Table 23. In Table 24, the results of
the SDM and DDM data from the DDSFLA algorithm tested on
specific data are presented in Table 20. Figs. 10–12 plot the
the SM55 dataset at an irradiance of 1000 W/m2 at temperatures
convergence curves of all algorithms under different conditions.
of 25 ◦ C, 40 ◦ C, and 60 ◦ C, respectively, are similar. Table 25
From the three figures, it can be seen that changing the size
shows that the fitness value of DDM with an irradiance of 800
of the population and the number of evaluations has an enor-
W/m2 is better than that of SDM. Tables 26 and 27 indicates that
mous impact on the initial solution of the GOTLBO algorithm.
the DDSFLA algorithm optimizes SDM and DDM on the KC200GT
The optimization performance of the DDSFLA algorithm on the
SDM, DDM, and PV models does not change with the popula- dataset with similar results. Figs. 13–17, and 18 plot the fitting
tion size and the number of evaluations. In Table 20, the SFLA curve between the measured and estimated data. Graphically,
algorithm’s ability to optimize the SDM decreases as the number both SDM and DDM optimized by the DDSFLA algorithm can be
of evaluations decreases to 5000. the CS algorithm’s performance used as mathematical models of real PV systems.
in handling the three models worsens with population size and
the number of evaluations. The results of several other algorithms 5. Conclusions and future directions
were not significantly different.
In this paper, a new delayed dynamic step mechanism is intro-
4.2.6. Evaluating the DDSFLA algorithm on the manufacturer’s data duced to improve the parameter accuracy of the SFLA algorithm
To further test the optimization performance of the DDSFLA for extracting SDM and DDM. As the SFLA algorithm has its own
algorithm under special conditions, this section conducts experi- fast convergence speed in dealing with photovoltaic problems,
ments to extract the parameters of SDM and DDM on ST40, SM55, it can be precocious and does not have its own ability to jump
and KC200GT datasets. All three datasets contain current–voltage out of the local optimum. In order to correct the shortcomings

Table 16
The accurate parameters and fitness values (model: PV, data: Photo-watt-PWP 201).
Iph (A) Isd (µA) RS () Rsh () n RMSE
DDSFLA 1.0306 3.4473 1.2023 971.2500 48.6040 2.4252E−03
GOTLBO 1.0305 3.5214 1.2000 992.8000 48.6860 2.4253E−03
IJAYA 1.0304 3.4732 1.2019 992.5300 48.6320 2.4255E−03
CS 1.0306 3.4980 1.2014 978.8900 48.6610 2.4344E−03
SSA 1.0301 3.8129 1.1912 1071.9000 48.9930 2.4386E−03
PSO 1.0296 4.0059 1.1865 1180.5000 49.1840 2.4566E−03
WOA 1.0313 2.9027 1.2189 826.9100 47.9560 2.4793E−03
GWO 1.0283 4.7360 1.1698 1696.3000 49.8420 2.6009E−03
DA 1.0263 4.9062 1.1169 1111.3000 50.0000 6.2886E−03
SFLA 1.0312 0.6430 1.3945 1050.7000 42.8580 8.7187E−03
SCA 1.0381 4.4850 1.2345 1971.4000 49.6190 1.0742E−02

238
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 18
Measured data and estimated data (model: PV, data: Photo-watt-PWP 201).
Measured data Estimated current data Estimated power data
Vm Im Ie IAEI REI Pe IAEP REP
1 0.1248 1.0315 1.0292 2.3394E−03 2.2731E−03 0.1284 2.9196E−04 2.2680E−03
2 1.8093 1.0300 1.0274 2.5962E−03 2.5270E−03 1.8589 4.6974E−03 2.5206E−03
3 3.3511 1.0260 1.0257 2.5235E−04 2.4602E−04 3.4374 8.4566E−04 2.4596E−04
4 4.7622 1.0220 1.0241 2.0983E−03 −2.0489E−03 4.8770 9.9923E−03 −2.0531E−03
5 6.0538 1.0180 1.0223 4.2707E−03 −4.1776E−03 6.1886 2.5854E−02 −4.1952E−03
6 7.2364 1.0155 1.0199 4.4006E−03 −4.3148E−03 7.3804 3.1845E−02 −4.3335E−03
7 8.3189 1.0140 1.0163 2.3285E−03 −2.2911E−03 8.4547 1.9371E−02 −2.2964E−03
8 9.3097 1.0100 1.0105 4.6260E−04 −4.5781E−04 9.4071 4.3066E−03 −4.5802E−04
9 10.2160 1.0035 1.0006 2.8973E−03 2.8956E−03 10.2220 2.9600E−02 2.8872E−03
10 11.0450 0.9880 0.9845 3.4647E−03 3.5191E−03 10.8740 3.8267E−02 3.5068E−03
11 11.8020 0.9630 0.9595 3.4742E−03 3.6208E−03 11.3240 4.1002E−02 3.6077E−03
12 12.4930 0.9255 0.9229 2.6395E−03 2.8602E−03 11.5290 3.2976E−02 2.8520E−03
13 13.1230 0.8725 0.8726 1.3505E−04 −1.5476E−04 11.4520 1.7722E−03 −1.5478E−04
14 13.6980 0.8075 0.8073 1.8447E−04 2.2849E−04 11.0590 2.5269E−03 2.2844E−04
15 14.2220 0.7265 0.7284 1.8749E−03 −2.5741E−03 10.3590 2.6665E−02 −2.5807E−03
16 14.7000 0.6345 0.6372 2.6653E−03 −4.1831E−03 9.3660 3.9179E−02 −4.2006E−03
17 15.1350 0.5345 0.5362 1.7235E−03 −3.2141E−03 8.1155 2.6084E−02 −3.2245E−03
18 15.5310 0.4275 0.4295 2.0042E−03 −4.6662E−03 6.6707 3.1127E−02 −4.6881E−03
19 15.8930 0.3185 0.3188 2.5118E−04 −7.8802E−04 5.0659 3.9920E−03 −7.8864E−04
20 16.2230 0.2085 0.2074 1.1443E−03 5.5187E−03 3.3639 1.8564E−02 5.4884E−03
21 16.5240 0.1010 0.0961 4.8715E−03 5.0677E−02 1.5884 8.0497E−02 4.8233E−02
22 16.7990 −0.0080 −0.0084 3.5448E−04 −4.2430E−02 −0.1403 5.9549E−03 −4.4311E−02
23 17.0500 −0.1110 −0.1110 4.9634E−05 4.4735E−04 −1.8917 8.4625E−04 4.4715E−04
24 17.2790 −0.2090 −0.2092 2.3843E−04 −1.1395E−03 −3.6155 4.1199E−03 −1.1408E−03
25 17.4890 −0.3030 −0.3008 2.1773E−03 7.2380E−03 −5.2609 3.8078E−02 7.1860E−03

Table 19
CPU time of eleven algorithms on SDM, DDM, PV.
DDSFLA SFLA IJAYA GOTLBO PSO CS WOA SSA SCA GWO DA
SDM 19.70 21.52 20.58 23.29 21.24 21.25 21.05 20.88 19.65 18.96 49.39
DDM 25.49 25.12 18.89 20.08 18.78 18.93 18.34 18.31 18.27 18.93 54.41
PV 25.48 24.61 18.11 20.37 18.89 18.95 18.77 18.73 18.85 18.21 50.15

Table 20
The different population size and the number of evaluations.
Algorithm Model RMSE
Pop Num Pop Num Pop Num Pop Num Pop Num
30 20,000 40 10,000 50 10,000 50 5000 60 5000
SDM 9.8630E−04 9.8718E−04 9.8675E−04 9.8611E−04 9.8816E−04
DDSFLA DDM 9.8434E−04 9.8531E−04 9.8523E−04 9.8585E−04 9.9481E−04
PV 2.4252E−03 2.4252E−03 2.4270E−03 2.4251E−03 2.4255E−03
SDM 1.8791E−03 1.4732E−03 2.2452E−03 4.0286E−03 3.9844E−03
SFLA DDM 2.5690E−03 2.2094E−03 4.3422E−03 4.2726E−03 2.9694E−03
PV 8.7187E−03 1.2352E−02 4.1238E−03 3.7355E−03 1.6435E−02
SDM 9.8871E−04 9.8756E−04 1.0011E−03 9.9667E−04 9.8690E−04
IJAYA DDM 9.9029E−04 9.8603E−04 9.8437E−04 1.0320E−03 1.0702E−03
PV 2.4255E−03 2.4271E−03 2.4261E−03 2.4328E−03 2.4343E−03
SDM 9.8997E−04 1.0154E−03 1.0487E−03 1.0606E−03 1.0280E−03
GOTLBO DDM 9.8708E−04 1.1652E−03 1.1582E−03 1.4502E−03 1.7724E−03
PV 2.4253E−03 2.4280E−03 2.4425E−03 2.5079E−03 2.4619E−03
SDM 9.9826E−04 1.0681E−03 9.8604E−04 1.0338E−03 9.8920E−04
PSO DDM 1.0412E−03 9.8540E−04 1.1275E−03 9.9884E−04 1.0305E−03
PV 2.4566E−03 2.4943E−03 2.4813E−03 2.4388E−03 2.6083E−03
SDM 1.1593E−03 1.7892E−03 2.1224E−03 2.8912E−03 5.1740E−03
CS DDM 1.2656E−03 2.2103E−03 3.2314E−03 7.4312E−03 6.2480E−03
PV 2.4344E−03 2.7158E−03 2.6414E−03 6.4492E−03 5.0479E−03
SDM 1.4339E−03 1.6068E−03 1.9992E−03 2.2621E−03 1.9900E−03
WOA DDM 1.2946E−03 1.4802E−03 1.6752E−03 1.6078E−03 3.8534E−03
PV 2.4793E−03 2.6837E−03 2.5847E−03 2.5740E−03 2.6193E−03
SDM 9.8705E−04 1.0198E−03 1.0253E−03 1.1644E−03 1.5276E−03
SSA DDM 1.0676E−03 9.9885E−04 1.3786E−03 1.1859E−03 1.6358E−03
PV 2.4386E−03 2.5456E−03 2.5306E−03 2.6209E−03 3.1980E−03
SDM 1.8246E−02 1.5962E−02 1.8644E−02 2.5522E−02 2.2971E−02
SCA DDM 1.4297E−02 2.1011E−02 1.5855E−02 1.8537E−02 1.9817E−02
PV 1.0742E−02 1.9344E−02 8.5130E−03 3.4531E−02 2.0368E−02

(continued on next page)

239
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 20 (continued).
Algorithm Model RMSE
Pop Num Pop Num Pop Num Pop Num Pop Num
30 20,000 40 10,000 50 10,000 50 5000 60 5000
SDM 1.0741E−03 1.7606E−03 1.8132E−03 2.1847E−03 1.3231E−03
GWO DDM 1.0761E−03 1.4150E−03 2.0242E−03 1.8178E−03 1.9804E−03
PV 2.6009E−03 2.6236E−03 2.6301E−03 2.6193E−03 2.6211E−03
SDM 2.8764E−03 5.0025E−03 2.3789E−03 4.4671E−03 7.1741E−03
DA DDM 3.7966E−03 6.7820E−03 4.8233E−03 5.3872E−03 1.4649E−02
PV 6.2886E−03 7.2301E−03 1.1078E−02 1.3343E−02 1.4717E−02

Table 21 Table 24
The range of parameters (models: DDM and SDM, data: manufacturer’s data). The accurate parameters and fitness values (models: SDM and DDM, data: SM55,
DDM SDM Manufacturer’s data irradiance : 1000 W/m2 , temperature: 25, 40, 60 ◦ C).

lb ub SM55 Temperature/1000 W/m2

Iph (A) 0 2 ISC 25 ◦ C 40 ◦ C 60 ◦ C

Isd1 (µA), Isd2 (µA) Isd (µA) 0 100 Iph (A) 3.4442 3.4659 3.4918
RS () 0 2 Isd (µA) 0.3673 1.5603 9.5917
Rsh () 0 5000 RS () 0.3014 0.3010 0.3037
SDM
n1 ,n2 n 1 4 Rsh () 817.3500 727.0500 670.0800
n 1.4616 1.4476 1.4408
RMSE 4.0399E−03 4.2267E−03 4.2987E−03
of the original SFLA algorithm while retaining its own advan- Iph (A) 3.4408 3.4654 3.4942
Isd1 (µA) 0.0050 2.2913 4.5098
tages, the initial stage of the search is reserved to allow the
RS () 0.3228 0.2981 0.3224
original SFLA algorithm to take advantage of its fast convergence Rsh () 1007.0000 840.8800 512.3000
DDM
speed; this is then combined with a dynamic step mechanism n1 1.2135 1.5798 1.3696
to improve the convergence accuracy of the SFLA algorithm. The Isd2 (µA) 0.5545 0.3760 8.0344
n2 1.5443 1.3672 1.7588
DDSFLA algorithm ranks first among the 23 benchmark func- RMSE 3.8780E−03 4.8240E−03 3.8634E−03
tions tested. In the parameter experiments of extracting SDM
and DDM on RTC France and Photo-watt-PWP 201 data sets, the
DDSFLA algorithm finds optimal identification results with faster

Table 22
The accurate parameters and fitness values (models: SDM and DDM, data: ST40, irradiance: 1000 W/m2 , temperature: 25, 40, 50,
70 ◦ C).
ST40 Temperature/1000 W/m2
25 ◦ C 40 ◦ C 50 ◦ C 70 ◦ C
Iph (A) 2.6689 2.6750 2.6893 2.6926
Isd (µA) 3.4629 9.9366 22.5890 86.2730
RS () 1.0522 1.0892 1.1339 1.1271
SDM
Rsh () 533.5300 528.2700 332.5400 361.6100
n 1.8542 1.7990 1.7450 1.7250
RMSE 3.2413E−03 2.9059E−03 2.0086E−03 7.8166E−04
Iph (A) 2.6740 2.6780 2.6908 2.6925
Isd1 (µA) 0.1194 0.0244 11.9880 47.6920
RS () 1.1443 1.1125 1.1584 1.1280
Rsh () 404.3400 426.2800 320.5500 365.7900
DDM
n1 1.5236 3.8562 1.6649 1.6890
Isd2 (µA) 6.0520 7.2341 40.8810 42.2010
n2 2.1115 1.7550 2.4318 1.8040
RMSE 1.0600E−03 1.7390E−03 1.7498E−03 7.8965E−04

Table 23
The accurate parameters and fitness values (models: SDM and DDM, data: ST40, irradiance: 200, 400, 600, 800, 1000 W/m2 , temperature: 25 ◦ C).
ST40 Irradiance/25◦ C
1000 W/m2 800 W/m2 600 W/m2 400 W/m2 200 W/m2
Iph (A) 2.6691 2.1335 1.6016 1.0654 0.5323
Isd (µA) 3.3877 2.4077 3.0015 3.3489 2.2975
RS () 1.0533 1.0619 1.0095 0.9668 0.9404
SDM
Rsh () 518.0600 417.8300 406.5000 405.0200 355.8800
n 1.8512 1.8095 1.8404 1.8612 1.8131
RMSE 3.1258E−03 2.4908E−03 1.6714E−03 1.2387E−03 5.5314E−04
Iph (A) 2.6733 2.1347 1.6037 1.0667 0.5331
Isd1 (µA) 9.4311 0.1615 0.9455 2.3934 0.9842
RS () 1.1370 1.1451 1.1261 1.0350 1.2627
Rsh () 420.2900 399.9900 370.6100 379.9800 350.4300
DDM
n1 2.2925 1.5428 1.6989 1.8136 1.7007
Isd2 (µA) 0.2889 6.6545 10.3860 0.0005 15.0840
n2 1.5910 2.1724 2.7491 3.9998 3.2986
RMSE 1.1129E−03 1.3481E−03 7.5533E−04 7.7968E−04 4.6398E−04

240
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Table 25
The accurate parameters and fitness values (models: SDM and DDM, data: SM55, irradiance: 200, 400, 600, 800, 1000 W/m2 , temperature: 25 ◦ C).
SM55 Irradiance/25 ◦ C
1000 W/m2 800 W/m2 600 W/m2 400 W/m2 200 W/m2
Iph (A) 3.4454 2.7560 2.0675 1.3812 0.6918
Isd (µA) 0.3774 0.4736 0.3474 0.2042 0.1644
RS () 0.3012 0.2668 0.2760 0.3083 0.2461
SDM
Rsh () 762.3800 670.3400 579.7200 474.0900 443.7900
n 1.4642 1.4847 1.4582 1.4116 1.3908
RMSE 4.0415E−03 3.4230E−03 2.2507E−03 1.1451E−03 5.3719E−04
Iph (A) 3.4477 2.7593 2.0683 1.3810 0.6918
Isd1 (µA) 0.1302 0.1201 0.3664 0.0236 0.0008
RS () 0.3259 0.3302 0.2686 0.2679 0.2499
Rsh () 552.6400 502.2900 553.9000 483.3100 443.4200
DDM
n1 1.3791 1.3707 1.4631 3.4535 3.9730
Isd2 (µA) 0.4534 0.3910 0.0944 0.2667 0.1621
n2 1.7796 1.7977 2.6771 1.4352 1.3896
RMSE 1.7701E−03 9.6884E−04 2.3408E−03 1.3899E−03 5.3530E−04

The DDSFLA algorithm provides a new solution to the problem of

Table 26
identifying photovoltaic parameters.
The accurate parameters and fitness values (models: SDM and DDM, data:
KC200GT, temperature: 25, 50, 75 ◦ C, irradiance: 1000 W/m2 ). In future work, we intend to further improve the ease of
KC200GT Temperature/1000 W/m2 setting the DDSFLA algorithm’s parameters. Since the DDSFLA
25 ◦ C 50 ◦ C 75 ◦ C algorithm has a high number of parameters that need to be
Iph (A) 8.2064 8.2878 8.3720
continuously adjusted according to the experimental process, it
Isd (µA) 0.0113 0.2442 2.1001 is not conducive to expanding its scope. Therefore, we hope to
RS () 0.3152 0.3224 0.3375 find a better way to set the parameters needed for the algorithm.
SDM
Rsh () 4873.2000 4922.8000 1653.6000
n 1.1606 1.1595 1.1197
RMSE 1.2085E−02 7.3942E−03 5.6265E−03
CRediT authorship contribution statement
Iph (A) 8.2159 8.2873 8.3684
Isd1 (µA) 1.4214 0.0002 0.2117
RS () 0.3049 0.3239 0.3439 Yi Fan: Writing - original draft, Writing - review & editing,
Rsh () 1636.6000 4960.9000 4727.9000 Software, Visualization, Investigation. Pengjun Wang: Writing -
DDM
n1 3.4624 1.3244 1.0017
original draft, Writing - review & editing, Software, Visualiza-
Isd2 (µA) 0.0170 0.2308 4.9166
n2 1.1838 1.1558 1.2676 tion, Investigation. Ali Asghar Heidari: Formal analysis, Writing -
RMSE 1.6581E−02 6.8052E−03 5.8943E−03 original draft, Writing - review & editing, Software, Visualization,
Investigation. Xuehua Zhao: Writing - original draft, Writing -
review & editing, Software, Visualization, Investigation. Hamza
convergence speed and higher convergence accuracy compared Turabieh: Writing - original draft, Writing - review & editing,
to other algorithms. The DDSFLA algorithm optimization yields Software, Visualization, Investigation. Huiling Chen: Conceptu-
the smallest adaptation values for the three models, evaluating alization, Methodology, Formal analysis, Investigation, Writing -
review & editing, Funding acquisition, Supervision.
the estimated current and power relative and absolute errors
in a reasonable interval. The computational cost of the DDS-
FLA algorithm does not increase much compared to the original Declaration of competing interest
SFLA algorithm. Increasing the population size and decreasing
the number of assessments had little effect on the DSSFLSA al- The authors declare that they have no known competing finan-
gorithm. The DDSFLA algorithm exhibits a stable optimization cial interests or personal relationships that could have appeared
capability under different temperatures or irradiance conditions. to influence the work reported in this paper.

Table 27
The accurate parameters and fitness values (models: SDM and DDM, data: KC200GT, irradiance: 200, 400, 600, 800, 1000 W/m2 , temperature: 25 ◦ C).
KC200GT Irradiance/25◦ C
1000 W/m2 800 W/m2 600 W/m2 400 W/m2 200 W/m2
Iph (A) 8.2084 6.5621 4.9279 3.2816 1.6444
Isd (µA) 0.0150 0.0054 0.0114 0.0100 0.0025
RS () 0.3093 0.3244 0.3097 0.2660 0.2307
SDM
Rsh () 4712.3000 3236.9000 1385.0000 1371.5000 784.5700
n 1.1765 1.1203 1.1638 1.1559 1.0791
RMSE 1.4491E−02 1.0695E−02 6.0222E−03 5.5300E−03 2.2721E−03
Iph (A) 8.2077 6.5603 4.9242 3.2815 1.6446
Isd1 (µA) 0.0142 0.0082 3.3506 0.0073 0.0028
RS () 0.3112 0.3145 0.2798 0.2840 0.2131
Rsh () 4959.3000 4901.1000 3329.3000 1303.9000 777.1600
DDM
n1 1.1733 1.1428 3.9335 1.1379 1.0861
Isd2 (µA) 0.0189 0.0087 0.0307 0.5906 0.0009
n2 2.5104 3.3666 1.2245 3.9482 3.9498
RMSE 1.3929E−02 1.3290E−02 1.1652E−02 4.7342E−03 2.4132E−03

241
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 10. Different population and number of evaluations on SDM.

Fig. 11. Different population and number of evaluations on DDM.

Fig. 12. Different population and number of evaluations on PV.

242
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 13. The fitting curves between measured and estimated data for SDM and DDM for ST40 at different temperature.

Fig. 14. The fitting curves between measured and estimated data for SDM and DDM for ST40 at different irradiance.

Fig. 15. The fitting curves between measured and estimated data for SDM and DDM for SM55 at different temperature.

Fig. 16. The fitting curves between measured and estimated data for SDM and DDM for SM55 at different irradiance.

243
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Fig. 17. The fitting curves between measured and estimated data for SDM and DDM for KC200GT at different temperature.

Fig. 18. The fitting curves between measured and estimated data for SDM and DDM for KC200GT at different irradiance.

Acknowledgments Cao, B., et al., 2020c. Hybrid microgrid many-objective sizing optimization with
fuzzy decision. IEEE Trans. Fuzzy Syst..
Cao, B., et al., 2020d. A many-objective optimization model of industrial internet
This research is supported by the Zhejiang Provincial Nat- of things based on private blockchain. IEEE Network 34 (5), 78–83.
ural Science Foundation of China (LJ19F020001), Science and Cao, B., et al., 2020e. Multiobjective evolution of fuzzy rough neural network via
Technology Plan Project of Wenzhou, China (2018ZG012), and Na- distributed parallelism for stock prediction. IEEE Trans. Fuzzy Syst. 28 (5),
tional Natural Science Foundation of China (62076185, U1809209, 939–952.
Cao, B., et al., 2020f. Quantum-enhanced multiobjective large-scale optimization
71803136, 61471133), and Guangdong Natural Science Founda- via parallelism. Swarm Evol. Comput. 57, 100697.
tion, China (2018A030313339), MOE (Ministry of Education in Chen, H., Chen, A., Xu, L., Xie, H., Qiao, H., Lin, Q., Cai, K., 2020a. A deep learning
China) Youth Fund Project of Humanities and Social Sciences CNN architecture applied in smart near-infrared analysis of water pollution
(17YJCZH261), Scientific Research Team Project of Shenzhen In- for agricultural irrigation resources. Agric. Water Manag. 240, 106303.
Chen, H., Fan, D.L., Fang, L., Huang, W., Huang, J., Cao, C., ...., Zeng, L., 2020b.
stitute of Information Technology, China (SZIIT2019KJ022), Taif Particle swarm optimization algorithm with mutation operator for particle
University Researchers Supporting Project Number filter noise reduction in mechanical fault diagnosis. Int. J. Pattern Recognition
(TURSP-2020/125), Taif University, Taif, Saudi Arabia. We appre- Artif. Intell. 2058012. [Link]
ciate the efforts and constructive comments of respected editor Chen, Y., He, L., Guan, Y., Lu, H., Li, J., 2017. Life cycle assessment of greenhouse
gas emissions and water-energy optimization for shale gas supply chain
and reviewers that enhanced this research quality. planning based on multi-level approach: Case study in Barnett, Marcellus,
Fayetteville, and Haynesville shales. Energy Conversion Manag. 134, 382–398.
References Chen, Y., He, L., Li, J., Zhang, S., 2018. Multi-criteria design of shale-gas-water
supply chains and production systems towards optimal life cycle economics
and greenhouse gas emissions under uncertainty. Comput. Chem. Eng. 109,
Abbassi, A., et al., 2020. Parameters identification of photovoltaic cell models
216–235.
using enhanced exploratory salp chains-based approach. Energy 198, 117333.
Chen, X., et al., 2016. Parameters identification of solar cell models using
AlRashidi, M.R., et al., 2011. A new estimation approach for determining the I–V
generalized oppositional teaching learning based optimization. Energy 99,
characteristics of solar cells. Sol. Energy 85 (7), 1543–1550.
170–180.
Ashuri, B., Tavakolan, M., 2015. Shuffled frog-leaping model for solving time- Chen, W., et al., 2019a. Applying population-based evolutionary algorithms and
cost-resource optimization problems in construction project planning. J. a neuro-fuzzy system for modeling landslide susceptibility. Catena 172,
Comput. Civ. Eng. 29 (1). 212–231.
Bhattacharjee, K.K., Sarmah, S.P., 2014. Shuffled frog leaping algorithm and its Chen, H., et al., 2019b. A fuzzy optimization strategy for the implementation of
application to 0/1 knapsack problem Kaushik Kumar. Appl. Soft Comput. 19, RBF LSSVR model in vis–NIR analysis of pomelo maturity. IEEE Trans. Ind.
252–263. Inf. 15 (11), 5971–5979.
Cao, Y., Li, Y., Zhang, G., Jermsittiparsert, K., Nasseri, M., 2020a. An efficient Chen, H., et al., 2019c. An opposition-based sine cosine approach with local
terminal voltage control for PEMFC based on an improved version of whale search for parameter estimation of photovoltaic models. Energy Convers.
optimization algorithm. Energy Rep. 6, 530–542. Manage. 195, 927–942.
Cao, B., Zhao, J., Gu, Y., Ling, Y., Ma, X., 2020b. Applying graph-based differential Chen, H., et al., 2020c. Model-based method with nonlinear ultrasonic system
grouping for multiobjective large-scale optimization. Swarm Evol. Comput. identification for mechanical structural health assessment. Trans. Emerg.
53, 100626. Telecommun. Technol. n/a, e3955. [Link]
Cao, B., Zhao, J., Yang, P., Gu, Y., Muhammad, K., Rodrigues, J.J., de Albuquerque, Chen, H., et al., 2020d. Parameters identification of photovoltaic cells and mod-
2019. Multiobjective 3-D topology optimization of next-generation wireless ules using diversification-enriched harris hawks optimization with chaotic
data center network. IEEE Trans. Ind. Inform. 16 (5), 3597–3605. drifts. J. Cleaner Prod. 244, 118778.

244
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Cheng, X., He, L., Lu, H., Chen, Y., Ren, L., 2016. Optimal water resources Liu, Y., Yang, C., Sun, Q., 2020b. Thresholds based image extraction schemes in
management and system benefit for the Marcellus shale-gas reservoir in big data environment in intelligent traffic management. IEEE Trans. Intell.
Pennsylvania and West Virginia. J. Hydrology 540, 412–422. Transp. Syst. 1–9.
Derrac, J., et al., 2011. A practical tutorial on the use of nonparametric statistical Liu, S., Yu, W, Chan, FTS, Niu, B, 2020c. A variable weight-based hybrid approach
tests as a methodology for comparing evolutionary and swarm intelligence for multi-attribute group decision making under interval-valued intuitionistic
algorithms. Swarm Evol. Comput. 1 (1), 3–18. fuzzy sets. Int. J. Intell. Syst. 1–38. [Link]
Easwarakhanthan, T., et al., 1986. Nonlinear minimization algorithm for deter- Liu, J., et al., 2015. A novel differential search algorithm and applications for
mining the solar cell parameters with microcomputers. Int. J. Sol. Energy 4 structure design. Appl. Math. Comput. 268, 246–269.
(1), 1–12. Liu, E., et al., 2019. Formation mechanism of trailing oil in product oil pipeline.
Eusuff, M., Lansey, K.E., Pasha, F., 2006. Shuffled frog-leaping algorithm: a Processes 7 (1), 7.
memetic meta-heuristic for discrete optimization. Eng. Optim. 38 (2), Liu, Y., et al., 2020d. Horizontal and vertical crossover of harris hawk optimizer
129–154. with nelder–mead simplex for parameter estimation of photovoltaic models.
Fu, X., et al., 2019. WSNs-assisted opportunistic network for low-latency message Energy Convers. Manage. 223, 113211. [Link]
forwarding in sparse settings. Future Gener. Comput. Syst. 91, 223–237. 2020.113211.
Fu, X., et al., 2020a. Environment-fusion multipath routing protocol for wireless Long, Q., Wu, C., Wang, X., 2015. A system of nonsmooth equations solver based
sensor networks. Inf. Fusion 53, 4–19. upon subgradient method. Appl. Math. Comput. 251, 284–299.
Fu, X., et al., 2020b. Topology optimization against cascading failures on wireless Luo, J., Chen, M.-R., 2014. Improved shuffled frog leaping algorithm and its multi-
sensor networks using a memetic algorithm. Comput. Netw. 107327. phase model for multi-depot vehicle routing problem. Expert Syst. Appl. 41
García, S., et al., 2010. Advanced nonparametric tests for multiple comparisons (5), 2535–2545.
in the design of experiments in computational intelligence and data mining: Lv, Z., Qiao, L., 2020. Deep belief network and linear perceptron based cognitive
Experimental analysis of power. Inform. Sci. 180 (10), 2044–2064. computing for collaborative robots. Appl. Soft Comput. 92, 106300.
Guo, J., et al., 2020. Does air pollution stimulate electric vehicle sales? Empirical Lv, Z., Xiu, W., 2020. Interaction of edge-cloud computing based on SDN and
evidence from twenty major cities in China. J. Cleaner Prod. 249, 119372. NFV for next generation IoT. IEEE Internet Things J. 7 (7), 5706–5712.
Hasanien, H.M., 2014. Shuffled frog leaping algorithm-based static synchronous Merchaoui, M., Sakly, A., Mimouni, M.F., 2018. Particle swarm optimisation with
compensator for transient stability improvement of a grid-connected wind adaptive mutation strategy for photovoltaic solar cell/module parameter
farm. Iet Renew. Power Gener. 8 (6), 722–730. extraction. Energy Convers. Manage. 175, 151–163.
Hasanien, H.M., 2015. Shuffled frog leaping algorithm for photovoltaic model Mirjalili, S., 2015. Moth-flame optimization algorithm: A novel nature-inspired
identification. IEEE Trans. Sustain. Energy 6 (2), 509–515. heuristic paradigm. Knowl.-Based Syst. 89, 228–249.
He, L., Chen, Y., Li, J., 2018. A three-level framework for balancing the tradeoffs Mirjalili, S., 2016a. Dragonfly algorithm: a new meta-heuristic optimization tech-
among the energy, water, and air-emission implications within the life-cycle nique for solving single-objective, discrete, and multi-objective problems.
shale gas supply chains. Resources Conservation Recycling 133, 206–228. Neural Comput. Appl. 27 (4), 1053–1073.
Mirjalili, S., 2016b. SCA: A Sine cosine algorithm for solving optimization
Hsu, H.-P., Chiang, T.-L., 2019. An improved shuffled frog-leaping algorithm for
problems. Knowl.-Based Syst. 96, 120–133.
solving the dynamic and continuous berth allocation problem (DCBAP). Appl.
Mirjalili, S., Lewis, A., 2016. The whale optimization algorithm. Adv. Eng. Softw.
Sci.-Basel 9 (21).
95, 51–67.
Humada, A.M., et al., 2016. Solar cell parameters extraction based on single and
Mirjalili, S., Mirjalili, S.M., Lewis, A., 2014. Grey wolf optimizer. Adv. Eng. Softw.
double-diode models: A review. Renew. Sustain. Energy Rev. 56, 494–509.
69, 46–61.
Jadidoleslam, M., Ebrahimi, A., 2015. Reliability constrained generation expansion
Mirjalili, S., et al., 2017. Salp swarm algorithm: A bio-inspired optimizer for
planning by a modified shuffled frog leaping algorithm. Int. J. Electr. Power
engineering design problems. Adv. Eng. Softw. 114, 163–191.
Energy Syst. 64, 743–751.
Mousavi, A.A., Zhang, C., Masri, S.F., Gholipour, G., 2020. Structural damage
Jervase, J.A., Bourdoucen, H., Allawati, A., 2001. Solar cell parameter extraction
localization and quantification based on a CEEMDAN Hilbert transform neural
using genetic algorithms. Meas. Sci. Technol. 12 (11), 1922–1925.
network approach: A model steel truss bridge case study. Sensors 20 (5),
Jiao, S., et al., 2020. Orthogonally adapted harris hawks optimization for
1271.
parameter estimation of photovoltaic models. Energy 203, 117804.
Ni, T., et al., 2020. Non-intrusive online distributed pulse shrinking-based
Kennedy, J., Eberhart, R., 1995. Particle swarm optimization. In: Proceedings of
interconnect testing in 2.5D IC. IEEE Trans. Circuits Syst. II: Express Briefs
ICNN’95 - International Conference on Neural Networks.
67 (11), 2657–2661.
Kiani, A.T., et al., 2020. Optimal PV parameter estimation via double exponential
Niknam, T., et al., 2011. A modified shuffle frog leaping algorithm for
function-based dynamic inertia weight particle swarm optimization. Energies
multi-objective optimal power flow. Energy 36 (11), 6420–6432.
13 (15), 4037. Panda, S., Sarangi, A., Panigrahi, S.P., 2014. A new training strategy for neural
Kumar, J.V., Kumar, D.M.V., 2014. Generation bidding strategy in a pool based network using shuffled frog-leaping algorithm and application to channel
electricity market using shuffled frog leaping algorithm. Appl. Soft Comput. equalization. Aeu-Int. J. Electron. Commun. 68 (11), 1031–1036.
21, 407–414. Peng, S., et al., 2020. Analysis of particle deposition in a new-type rectifying
Lei, D., Guo, X., 2015. A shuffled frog-leaping algorithm for hybrid flow shop plate system during shale gas extraction. Energy Sci. Eng. 8 (3), 702–717.
scheduling with two agents. Expert Syst. Appl. 42 (23), 9333–9339. Qian, J., Feng, S., Li, Y., Tao, T., Han, J., Chen, Q., Zuo, C., 2020a. Single-
Lei, D., Zheng, Y., Guo, X., 2017. A shuffled frog-leaping algorithm for flexible shot absolute 3D shape measurement with deep-learning-based color fringe
job shop scheduling with the consideration of energy consumption. Int. J. projection profilometry. Opt. Lett. 45 (7), 1842–1845.
Prod. Res. 55 (11), 3126–3140. Qian, J., Feng, S., Tao, T., Hu, Y., Li, Y., Chen, Q., Zuo, C., 2020b. Deep-learning-
Li, C., Sun, L., Xu, Z., Wu, X., Liang, T., Shi, W., 2020a. Experimental investigation enabled geometric constraints and phase unwrapping for single-shot
and error analysis of high precision FBG displacement sensor for structural absolute 3D shape measurement. APL Photonics 5 (4), 046105.
health monitoring. Int. J. Struct. Stab. Dyn. 20 (6), 2040011. [Link] Qiu, T., Shi, X., Wang, J., Li, Y., Qu, S., Cheng, Q., Sui, S., 2019. Deep learning: A
10.1142/S0219455420400118. rapid and efficient route to automatic metasurface design. Adv. Sci. 6 (12),
Li, T., Xu, M., Zhu, C., Yang, R., Wang, Z., Guan, Z., 2019a. A deep learning 1900128.
approach for multi-frame in-loop filter of HEVC. IEEE Trans. Image Process. Qu, S., Han, Y., Wu, Z., al, et., 2020. Consensus modeling with asymmetric
28 (11), 5663–5678. cost based on data-driven robust optimization. Group Decis Negot http:
Li, X., Zhu, Y., Wang, J., 2019b. Highly efficient privacy preserving location- //[Link]/10.1007/s10726-020-09707-w.
based services with enhanced one-round blind filter. IEEE Trans. Emerg. Top. Rajasekar, N., Krishna Kumar, N., Venugopalan, R., 2013. Bacterial foraging
Comput. 1. algorithm based solar PV parameter estimation. Sol. Energy 97, 255–265.
Li, S., et al., 2020b. Slime mould algorithm: A new method for stochastic Ran, W., Liu, S., Zhang, Z., 2020. A polling-based dynamic order-picking system
optimization. Future Gener. Comput. Syst. 111, 300–323. considering priority orders. Complexity 2020, 4595316.
Liang, J., et al., 2020. Parameters estimation of solar photovoltaic models Ren, J., Zhang, C., Hao, Q., 2020. A theoretical method to evaluate honeynet
via a self-adaptive ensemble-based differential evolution. Sol. Energy 207, potency. Future Gener. Comput. Syst. 116, 76–85. [Link]
336–346. [Link].2020.08.021.
Liu, S., Chan, F.T., Ran, W., 2016. Decision making for the selection of cloud Ridha, H.M., et al., 2020. Boosted mutation-based harris hawks optimizer for
vendor: An improved approach under group decision-making with integrated parameters identification of single-diode solar cell models. Energy Convers.
weights and objective/subjective attributes. Expert Syst. Appl. 55, 37–47. Manage. 209, 112660.
Liu, L., Liu, S., 2020. Integrated production and distribution problem of perishable Ridha, H.M., et al., 2021. Multi-objective optimization and multi-criteria decision-
products with a minimum total order weighted delivery time. Mathematics making methods for optimal design of standalone photovoltaic system:
8 (2), 146. A comprehensive review. Renew. Sustain. Energy Rev. 135, 110202. http:
Liu, J., Liu, Y., Wang, X., 2020a. An environmental assessment model of con- //[Link]/10.1016/[Link].2020.110202.
struction and demolition waste based on system dynamics: a case study in Sharma, T.K., Pant, M., 2017. Opposition based learning ingrained shuffled
Guangzhou. Environ. Sci. Pollution Res. 27 (30), 37237–37259. frog-leaping algorithm. J. Comput. Sci. 21, 307–315.

245
Y. Fan, P. Wang, A.A. Heidari et al. Energy Reports 7 (2021) 228–246

Shen, L., et al., 2016. Evolving support vector machines using fruit fly Xiong, L., et al., 2016. Improved stability and H∞ performance for neutral
optimization for medical data classification. Knowl.-Based Syst. 96, 61–75. systems with uncertain Markovian jump. Nonlinear Anal. Hybrid Syst. 19,
Shi, K., et al., 2017a. Non-fragile sampled-data robust synchronization of uncer- 13–25.
tain delayed chaotic lurie systems with randomly occurring controller gain Xiong, Q., et al., 2020. A parallel algorithm framework for feature extraction of
fluctuation. ISA Trans. 66, 185–199. EEG signals on MPI. Comput. Math. Methods Med. 2020, 9812019.
Shi, K., et al., 2017b. Secondary delay-partition approach on robust performance Xu, M., Li, T., Wang, Z., Deng, X., Yang, R., Guan, Z., 2018. Reducing complexity
analysis for uncertain time-varying lurie nonlinear control system. Optim. of HEVC: A deep learning approach. IEEE Trans. Image Process. 27 (10),
Control Appl. Methods 38 (6), 1208–1226. 5044–5059.
Shi, K., et al., 2018. Nonfragile asynchronous control for uncertain chaotic Xu, S., Wang, Y., 2017. Parameter estimation of photovoltaic modules using a
lurie network systems with Bernoulli stochastic process. Internat. J. Robust hybrid flower pollination algorithm. Energy Convers. Manage. 144, 53–68.
Nonlinear Control 28 (5), 1693–1714. Xu, Y., et al., 2019. Enhanced moth-flame optimizer with mutation strategy for
Shi, K., et al., 2020a. Hybrid-driven finite-time H∞ sampling synchronization global optimization. Inform. Sci. 492, 181–203.
control for coupling memory complex networks with stochastic cyber Yang, X.S., 2010a. Firefly algorithm, stochastic test functions and design
attacks. Neurocomputing 387, 241–254. optimisation. Int. J. Bio-Inspired Comput. 2 (2), 78–84.
Shi, K., et al., 2020b. Reliable asynchronous sampled-data filtering of T–S fuzzy Yang, X.S., 2010b. A new metaheuristic bat-inspired algorithm. In: Studies in
uncertain delayed neural networks with stochastic switched topologies. Computational Intelligence. pp. 65–74.
Fuzzy Sets and Systems 381, 1–25. Yang, L., Chen, H., 2019. Fault diagnosis of gearbox based on RBF-PF and particle
Shockley, W., Field, E.M., 1952. Electrons and holes in semiconductors. Phys. swarm optimization wavelet neural network. Neural Comput. Appl. 31 (9),
Today 5 (12), 18–19. 4463–4478.
Singh, V., Gu, N., Wang, X., 2011. A theoretical framework of a BIM-based Yang, X.S., Deb, S., 2009. Cuckoo search via Lévy flights. In: 2009 World Congress
multi-disciplinary collaboration platform. Autom. Constr. 20 (2), 134–144. on Nature and Biologically Inspired Computing, NABIC 2009 - Proceedings.
Song, S., et al., 2020. Dimension decided harris hawks optimization with Yang, S., Deng, B., Wang, J., Li, H., Lu, M., Che, Y., Loparo, K.A., 2019. Scalable digi-
Gaussian mutation: Balance analysis and diversity patterns. Knowl.-Based tal neuromorphic architecture for large-scale biophysically meaningful neural
Syst. 106425. [Link] network with multi-compartment neurons. IEEE Trans. Neural Networks
Soon, J.J., Low, K., 2012. Photovoltaic model identification using particle swarm Learn. Syst. 31 (1), 148–162.
optimization with inverse barrier constraint. IEEE Trans. Power Electron. 27 Yu, K., et al., 2017. Parameters identification of photovoltaic models using
(9), 3975–3983. an improved JAYA optimization algorithm. Energy Convers. Manage. 150,
Su, Z., et al., 2019. Flow field and noise characteristics of manifold in natural gas 742–753.
transportation station. Oil & Gas Sci. Technol.–Revue d’IFP Energies Nouv. 74, Zeng, H.-B., Liu, X.-G., Wang, W., 2019a. A generalized free-matrix-based integral
70. inequality for stability analysis of time-varying delay systems. Appl. Math.
Sun, L., Li, C., Zhang, C., Liang, T., Zhao, Z., 2019a. The strain transfer mechanism Comput. 354, 1–8.
of fiber bragg grating sensor for extra large strain monitoring. Sensors 19 Zeng, H.-B., et al., 2019b. New results on stability analysis of systems with time-
(8), 1851. varying delays using a generalized free-matrix-based inequality. J. Franklin
Sun, L., Li, C., Zhang, C., Su, Z., Chen, C., 2018. Early monitoring of rebar corrosion Inst. B 356 (13), 7312–7321.
evolution based on FBG sensor. Int. J. Struct. Stab. Dyn. 18 (08), 1840001. Zeng, B., et al., 2020. Co-optimized public parking lot allocation and incen-
Sun, P., et al., 2016. Research and application of parallel normal cloud mutation tive design for efficient PEV integration considering decision-dependent
shuffled frog leaping algorithm in cascade reservoirs optimal operation. uncertainties. IEEE Trans. Ind. Inf..
Water Resour. Manage. 30 (3), 1019–1035. Zhang, C., Alam, Z., Sun, L., Su, Z., Samali, B., 2019a. Fibre bragg grating sensor-
Sun, G., et al., 2019b. An adaptive differential evolution with combined strategy based damage response monitoring of an asymmetric reinforced concrete
for global numerical optimization. Soft Comput. 24 (9), 6277–6296. shear wall structure subjected to progressive seismic loads. Struct. Control
Tripathy, B., Dash, S., Padhy, S.K., 2015. Multiprocessor scheduling and neural Health Monitoring 26 (3), e2307.
network training methods using shuffled frog-leaping algorithm. Comput. Zhang, X., Ding, S., Sun, T., 2016. Multi-class LSTMSVM based on optimal directed
Ind. Eng. 80, 154–158. acyclic graph and shuffled frog leaping algorithm. Int. J. Mach. Learn. Cybern.
Wang, M., Chen, H., 2020. Chaotic multi-swarm whale optimizer boosted support 7 (2), 241–251.
vector machine for medical diagnosis. Appl. Soft Comput. 88, 105946. http: Zhang, Y., Huang, C., Jin, Z., 2020a. Backtracking search algorithm with reusing
//[Link]/10.1016/[Link].2019.105946. differential vectors for parameter identification of photovoltaic models.
Wang, B., Zhang, B.F., Liu, X.W., 2020. An image encryption approach on the Energy Convers. Manage. 223, 113266.
basis of a time delay chaotic system. Optik 225, 165737. Zhang, H., Qiu, Z., Cao, J., Abdel-Aty, M., Xiong, L., 2020b. Event-triggered syn-
Wang, M., et al., 2017. Toward an optimal kernel extreme learning machine using chronization for neutral-type semi-Markovian neural networks with partial
a chaotic moth-flame optimization strategy with applications in medical mode-dependent time-varying delays. IEEE Trans. Neural Networks Learn.
diagnoses. Neurocomputing 267, 69–84. Syst. 31 (11), 4437–4450. [Link]
Wang, M., et al., 2020a. Evaluation of constraint in photovoltaic models by Zhang, X., et al., 2019b. Robust low-rank tensor recovery with rectification and
exploiting an enhanced ant lion optimizer. Sol. Energy 211, 503–521. alignment. IEEE Trans. Pattern Anal. Mach. Intell. [Link]
Wang, X., et al., 2020b. Multi-population following behavior-driven fruit fly op- TPAMI.2019.2929043.
timization: A Markov chain convergence proof and comprehensive analysis. Zhang, K., et al., 2020c. A comprehensive assessment framework for quantifying
Knowledge-Based Syst. 210, 106437. climatic and anthropogenic contributions to streamflow changes: A case
Wang, S., et al., 2020c. Physically-based landslide prediction over a large region: study in a typical semi-arid north China basin. Environ. Model. Softw.
Scaling low-resolution hydrological model results for high-resolution slope 104704.
stability assessment. Environ. Model. Softw. 124, 104607. Zhang, H., et al., 2020d. Orthogonal Nelder–Mead moth flame method for
Wen, D., et al., 2017. Evaluating the consistency of current mainstream wearable parameters identification of photovoltaic modules. Energy Convers. Manage.
devices in health monitoring: a comparison under free-living conditions. J. 211, 112764.
Med. Internet Res. 19 (3), e68. Zhao, C., Li, J., 2020. Equilibrium selection under the Bayes-based strategy
Wolf, M., Rauschenbach, H., 1963. Series resistance effects on solar cell updating rules. Symmetry 12 (5), 739.
measurements. Adv. Energy Convers. 3 (2), 455–479. Zhao, Z., Xu, Q., Jia, M., 2016. Improved shuffled frog leaping algorithm-based BP
Wu, C., Wu, P., Wang, J., Jiang, R., Chen, M., Wang, X., 2020a. Critical review of neural network and its application in bearing early fault diagnosis. Neural
data-driven decision-making in bridge operation and maintenance. Struct. Comput. Appl. 27 (2), 375–385.
Infrastruct. Eng. 1–24. [Link] Zhao, X., et al., 2019. Chaos enhanced grey wolf optimization wrapped ELM for
Wu, T., et al., 2019. New stabilization results for semi-Markov chaotic systems diagnosis of paraquat-poisoned patients. Comput. Biol. Chem. 78, 481–490.
with fuzzy sampled-data control. Complexity 2019. Zhou, J., et al., 2015. A modified shuffled frog leaping algorithm for PAPR
Wu, T., et al., 2020b. New results on stabilization analysis for fuzzy semi-Markov reduction in OFDM systems. Ieee Trans. Broadcast. 61 (4), 698–709.
jump chaotic systems with state quantized sampled-data controller. Inform. Zhu, B., Su, B., Li, Y., 2018a. Input-output and structural decomposition analysis
Sci. 521, 231–250. of India’s carbon emissions and intensity, 2007/08–2013/14. Appl. Energy
Xie, J., et al., 2018. Evaluating the validity of current mainstream wearable 230, 1545–1556.
devices in fitness tracking under various physical activities: comparative Zhu, B., et al., 2018b. Hilbert spectra and empirical mode decomposition: A
study. JMIR mHealth uHealth 6 (4), e94. multiscale event analysis method to detect the impact of economic crises
on the European carbon market. Comput. Econ. 52 (1), 105–121.

246