Identifying The Parameters of Different Configurations of Photovoltaic Models Based On Recent Artificial Ecosystem-Based Optimization Approach

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Received: 21 April 2020 Revised: 4 June 2020 Accepted: 26 June 2020

DOI: 10.1002/er.5747

RESEARCH ARTICLE

Identifying the parameters of different configurations


of photovoltaic models based on recent artificial
ecosystem-based optimization approach

Dalia Yousri1 | Hegazy Rezk2,3 | Ahmed Fathy4,5

1
Electrical Engineering Department,
Faculty of Engineering, Fayoum
Summary
University, Faiyum, Egypt Identifying accurate and precise photovoltaic models' parameters is the pri-
2
College of Engineering at Wadi mary gate in providing a proper PV system design simulate its real behavior.
Addawaser, Prince Sattam Bin Abdulaziz
Therefore, this article proposed a new approach based on a recent meta-
University, Al-Kharj, Saudi Arabia
3
Electrical Engineering Department,
heuristic algorithm of artificial ecosystem-based optimization (AEO) to identify
Faculty of Engineering, Minia University, the optimal parameters of PV cell and module models. Various PV models are
Minya, Egypt considered in this work as single diode (SD), double diode (DD), and triple
4
Electrical Engineering Department, diode (TD)-based circuits. The analysis is performed on which are
Faculty of Engineering, Jouf University,
Al-Jouf, Saudi Arabia R.T.C. France silicon solar cell, FSM-25 PV module, and Canadian-Solar-
5
Electrical Power & Machine Department, (CS6P-240P) multi-crystalline solar panel with the aid of experimental data
Faculty of Engineering, Zagazig under different operating conditions. Moreover, Lambert form is employed to
University, Zagazig, Egypt
validate the constructed model. Furthermore, comparative analysis with Harris
Correspondence hawks optimizer (HHO), gray wolf optimizer (GWO), and salp swarm algo-
Dalia Yousri, Electrical Engineering rithm (SSA) is performed. Additionally, statistical analysis using the Wilcoxon
Department, Faculty of Engineering,
Fayoum University, Fayoum, Egypt.
signed rank test is implemented across the three series of experiments for all
Email: day01@fayoum.edu.eg employed optimizers. The obtained results confirmed the competence of the
proposed approach in identifying the PV cell and modules equivalent circuits'
parameters.

KEYWORDS
artificial ecosystem-based optimizer, double diode PV model, PV parameters estimation, single
diode PV model, three-diode model

1 | INTRODUCTION Many problems arise with operating the PV panels as their


characteristics are very sensitive to the irradiance and tem-
Recently, renewable energy sources (RESs) penetrated in perature. Therefore, many researchers face great challenges
many engineering applications as alternatives of fossil fuel in modeling the PV panel especially with the presence of
sources. Solar energy is one form of RESs in which the sun- parameters not defined in the module datasheet given by
light is converted to electrical energy via solar photovoltaic the manufacturer. It is important to establish an equivalent
(PV) panels. Different PV panels are available which are circuit of the PV panel converges to the practical one. Many
mono-crystalline, multi-crystalline, and amorphous type. studies have been reported in estimating the unknown
parameters of the PV panels.
Dalia Yousri, Hegazy Rezk, and Ahmed Fathy contributed equally to Several methods have been presented for extracting
this article. the unknown parameters of the PV models that can be

Int J Energy Res. 2020;1–21. wileyonlinelibrary.com/journal/er © 2020 John Wiley & Sons Ltd 1
2 YOUSRI ET AL.

categorized to (i) deterministic (ie, Analytical and estimation of PV cell equivalent circuit have been
Numerical methods), (ii) Soft computing algorithms. The reviewed.16 The author divided the approaches into three
deterministic methods include the analytical method that categories; analytical, meta-heuristic, and hybrid methodol-
was proposed by the authors in Reference 1. These ogies comprising both. Yu et al17 modified JAYA optimizer
methods utilized several points on the current-voltage to solve the parameter estimation process of different PV
curve (I-V), such as Vmp, and Imp, Isc, Voc, and inter- modules. The modification has been presented via incorpo-
section of axis of I-V curve.2 The efficiency of the analyti- rating self-adaptive weight to avoid stuck in local optima.
cal approach mainly relies upon the selection of points Abbassi et al18 solved the problem of evaluating the
on the I-V curve. On the other hand, the numerical unknown parameters of the PV panel circuit via salp swarm
extraction method was presented by the authors in Refer- algorithm (SSA) with the aid of experimental data. Xu
ence 3, where the parameters can be identified by et al19 introduced an approach comprising FPA and Nelder-
implementing curve fitting. The curve fitting algorithm Mead simplex method to evaluate the solar cells and PV
requires high amount of computations. Its accuracy relies panels equivalent circuits' parameters.
upon the proper selection of the fitting method. The huge Moth-flame optimizer has been proposed for modeling
number of parameters cause also increasing the complex- the multi-crystalline PV solar cell and module.20 Teaching-
ity of the algorithm and reducing its capability to esti- learning-based optimization (TLBO) has been combined
mate accurate values.4-6 These methods are more with artificial bee colony (ABC) and employed to solve the
dominant in the local search process since they have parameter estimation problem of solar PV modules.21
been executed via a gradient-based approach. Therefore, Nassar et al22 presented two approaches to identify the opti-
they can easily get absorbed in local optimal.7 mal parameters of PV modules, the first one is iterative
In the circumstance of the soft computing algorithms, method while the second one is Lambert W function analyt-
Alam et al8 presented flower pollination algorithm (FPA) as ical method. Ma et al23 investigated different meta-heuristic
a tool to identify the parameters of PV module modeled via approaches of genetic algorithm (GA), differential evolu-
single and double diodes. The analysis has been performed tion (DE), particle swarm optimization (PSO), bacteria for-
with the aid of experimental data of voltage–current charac- aging algorithm (BFA), ABC, and Cuckoo search (CS) in
teristic of the module. Fathy et al9 introduced imperialist estimating the electrical parameters of the PV module
competitive algorithm (ICA) for identifying the PV module model. A self-adaptive TLBO has been introduced by Yu
parameters considering maximum power point (MPP) as et al17 to obtain the parameters of PV circuit. Hasanien24
objective function. Yu etal10 adopted the performance- employed shuffled frog leaping algorithm (SFLA)to evalu-
guided JAYA (PGJAYA) algorithm to estimate the parame- ate the parameters of single diode-based circuit of PV mod-
ters of single diode (SD) and double diode (DD) models for ule with the aid of experimental data. Guo et al25 presented
PV cell and module. In Reference 7, the authors applied the cat swarm optimizer (CSO) to identify the unknown param-
multiple learning backtracking search algorithm (MLBSA) eters of solar cells modeled via single and double diodes. A
to determine the parameters of SD and DD models, the pro- generalized oppositional TLBO based approach has been
posed algorithm results have been validated against several presented to solve the solar cell single and double diodes
optimization algorithms in terms of the accuracy and com- based model parameter estimation problem.26 Jordehi27
putational time. Oliva et al11 incorporated chaotic maps in presented an improved PSO via incorporating enhanced
the whale optimizer to adapt its internal parameters auto- leader to improve its convergence to estimate the optimal
matically, the modified approach has been employed to esti- parameters of solar cells and PV modules equivalent cir-
mate the optimal parameters of solar cells. Chin et al12 cuits. Chen et al28 presented a methodology based on diver-
reviewed many approaches of estimating the optimal sification enhanced Harris hawks optimizer (HHO) to
parameters of PV cells showing the features, advantages, estimate the unknown parameters of PV module circuit.
and defects of each one. A simplified bird mating optimizer SDM, DDM, and the dynamic PV model have been pro-
(SBMO) has been employed to identify the electrical param- posed in Reference 29. Comparative analysis of several opti-
eters of the amorphous type panel operated under different mizers employed in estimating the solar cells' parameters
conditions.13 Chikh et al14 presented a maximum power has been presented in Reference 30. Chenouard et al31
point tracker approach for PV system with estimating its introduced interval branch and bound algorithm to esti-
model parameters via adaptive neuro-fuzzy inference sys- mate the parameters of three models (SD, DD, and TD) of
tem (ANFIS) and measuring the voltage and current of the PV module. The maximum absolute error has been
array. Ram et al15 introduced a hybrid approach comprising employed as objective function. Ridha et al32 determined
bee pollinator and FPA to extract the PV module circuit the parameter of the SDM-based module via electromagne-
parameters considering single and double diode models. tism like algorithm (IEM). Grasshopper optimization algo-
Many reported approaches employed in parameter rithm (GOA) has been employed to estimate the optimal
YOUSRI ET AL. 3

parameters of three-diode model of PV module under dif- than 20 well established algorithms to confirm its superi-
ferent operating conditions.33 Long et al34 introduced a ority based on several statistical measures. Moreover, the
modified whale optimization algorithm (WOA) named Lambert W function is implemented to validate the iden-
refraction-learning-based WOA as a tool available for esti- tified parameters by the proposed algorithm. Addition-
mating the PV models parameters. Yousri et al35 utilized ally, Wilcoxon signed rank test with significant difference
the fractional chaos maps to improve the ensemble particle 0.05 is performed to assess the consistency of the pro-
swarm optimizer (FC-EPSO) during extracting the parame- posed AEO technique in comparable with well-regarded
ters of single, double and three-diode models. algorithms such as (HHO), gray wolf optimizer (GWO),
As per the literature some drawbacks have been and salp swarm algorithm (SSA) techniques. Further-
detected, first, although many meta-heuristic approaches more, the proposed algorithm compared with 20 recent
have been employed in estimating the unknown parame- optimization algorithm in the literature to highlight the
ters of both solar cells and modules models, most of them reliability and applicability of the proposed AEO. The
have defects in converging to the global solution due to obtained results confirmed the competence of AEO in
stuck in local optima. Moreover, many researchers providing the best fitting with highest detection for the
employed experimental data available in some other maximum power point value with fastest convergence
works which lack of accuracy. Furthermore, some speed and highest consistency for the RMSE value. More-
approaches are complex in construction, requiring large over, the comparison with the reported literature reveal
controlling parameters, and consuming long time in that the AEO is the algorithm that applied for identifying
implementation. Most of the proposed techniques have the SDM, DDM, and TDM. Whereas most of the litera-
been implemented for SDM and DDM only without test- ture have been proposed for SDM and DDM besides most
ing them for the more detailed and complicated model as of the algorithms utilized huge number of iterations to
TDM. Furthermore, the output of the meta-heuristic is converge that is why the authors recommended AEO as
dependent on many considerations, the employed fitness most applicable optimizer in comparable with the
function is the main one of them. By reviewing the litera- literature.
ture, it can be recognized that some of the researchers
determined the estimated current on the basis of the
actual measurement values of the current and the 2 | DIFFERENT MODELS OF P V
extracted parameters that can be regarded as an approxi- MODULE
mate procedure, because from the mathematical perspec-
tive, this nonlinear equation should be solved In the literature, three different mathematical models;
numerically.35-37 Therefore, the authors motivated to single, double, and triple diodes are very common to
highlight this point and clarify the precise objective func- mimic the performance of the solar PV panel. The main
tion that is the root mean square error (RMSE) between advantage of single diode model (SDM) is its simplicity.
the measured and estimated current since the extracted However, the accuracy of SDM is less than double and
current was determined with the aid of the estimated three diodes models specifically under low solar irradi-
parameters, and the nonlinear equation of the PV models ance.38 This because the SDM declines the consideration
was resolved on the basis of the Newton–Raphson of recombination losses developed in the depletion
method. Since detecting which objective is most suitable region.35 The different models of solar PV panel are illus-
for the identification of more precise parameters is a key trated in Figure 1. The generated solar PV current based
issue for the provision of precise PV model; therefore the on the SDM can be estimated using the following
authors motivated to employ this approach while com- relations35,39:
puting the objective function.
In this article, a simple, less controlling parameters, I = I g −I d −I p ð1Þ
and consuming acceptable time approach named artifi-
  
cial ecosystem-based optimizer (AEO) has been proposed ðV + IRs Þ ðV + IRs Þ
for the first time to estimate the electrical parameters of I = I g − I o exp −1 I 0 − ð2Þ
aV t Rp
solar cell and PV modules models based on several series
of experiments to tackle the early mentioned drawbacks.
Measured experimental datasets under several environ- where Ig, Id, Ip, Rs, and Rp are the generated-light current,
mental condition of radiation and temperature values are diode current, shunt-leakage current, series, and shunt
considered to assess the quality of the algorithm with resistances. The symbol of a denotes the diode ideality
handling the parameters of SDM, DDM and TDM factor and Vt represents the thermal voltage which is
models. The proposed algorithm is compared with more defined as:
4 YOUSRI ET AL.

FIGURE 1 Different models of solar photovoltaic panel [Colour figure can be viewed at wileyonlinelibrary.com]

N s KT Equation (4), it can be see that DDM has seven parame-


Vt = ð3Þ
q ters which are Ig, Io1, Io2, a1, a2, Rs, and Rp.
In order to simulate the leakage current in the grain
where k is the constant of Boltzmann with value of boundaries of multi-crystalline silicon solar cells, third
1.35 * 10−23J/K, T is the cell temperature in Kelvin, q is diode must be inserted to DDM35 as illustrated in three-
the electron charge (1.6 × 10−19 C) and Ns is the number diode model (TDM) (Figure 1). To estimate the generated
of series cells. solar PV current based on TDM, Equation (4) can be
Based on Equation (2), it can be noted that there are modified as following;
main five parameters; Ig, Io, a, RS, and RP that requited to be
       
identified for accurate simulating the behavior of PV panel. qðV + IRs Þ qðV + IRs Þ
I = I g −I o1 exp −1 − I o2 exp −1
Whereas in (DDM), as illustrated in Figure 1, additional a1 kT a2 kT
   
diode has been added to SDM in order to enhance the accu- qðV + IRs Þ ðV + IRs Þ
− I o3 exp −1 −
racy of simulating the physical effects at the P-N junction. a3 kT Rp
The first diode simulates the diffusion current while the sec- ð5Þ
ond one simulates the recombination effects. This will
increase the number of the parameters to seven instead of where Io3 denotes the saturation current and a3 is the ide-
five as in SDM. The generated solar PV current based on the ality factor of the third diode. Therefore, based on Equa-
DDM can be estimated using the following relation.38 tion (4), it can be see that TDM has nine parameters.
These parameters are Ig, Io1, Io2, Io3, a1, a2, a3, Rs, and Rp.
   
qðV + IRs Þ
I = Ig −Io1 exp −1
a1 kT
    ð4Þ
qðV + IRs Þ ðV + IRs Þ 3 | PROB LEM FOR MULATION
−Io2 exp −1 −
a2 kT Rp
The process of parameter determination of different
models of PV panel is considered as nonlinear optimiza-
where Io1 and Io2 denote the diffusion and saturation cur- tion problem. It is accounted as vital process to provide
rents whereas a1 and a2 are the ideality factors of the first an accurate model of PV that converges to the real one.
and second diode respectively.Accordingly, based on The implementation of such optimization problem
YOUSRI ET AL. 5

requires identifying the variables to be designed and the where Imeas is the measured current, and Iest is the esti-
objective function to be minimized. The accuracy of the mated current. The estimated current is calculated with
obtained model parameters via the optimization tech- the aid of the extracted parameters and solving Equa-
niques are sensitive to the variation of the employed fit- tions (2), (4), and (5) using Newton-Raphson method as
ness function. In this work, the authors employed the root follows:
mean square error (RMSE) between the measured and the
estimated current for identifying the SDM, DDM, and ðdI Þ
I est = I est −  0  ð7Þ
TDM parameters as an objective function. The estimated dI
current can be computed based on the identified parame-
ters via employing Newton-Raphson function to solve the
nonlinear equations of the PV model as given below: where dI is the difference function of I. The symbol dI' is
• The objective function: the first derivative of dI with respect to I, their expres-
sions can be written as follows:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u M Regarding to SDM, the dI and dI' functions can be
u1 X
objðZ Þ = t ðI measi − I esti ðV measi , Z ÞÞ2 ð6Þ expressed as follows:
M i=1

FIGURE 2 Flowchart of AEO-based PV parameters estimation process [Colour figure can be viewed at wileyonlinelibrary.com]

TABLE 1 The lower and upper boundaries of SDM, DDM, and TDM parameters

RTC solar cell (SDM/DDDM) FSM-25 PV solar panel (SDM) CS6P-240P solar module (DDM/TDM)

Parameters LB UB Parameters LB UB Parameters LB UB


Rs(Ω) 0 0.5 Rs(Ω) 0 2 Rs(Ω) 0 2
Rp(Ω) 0 1000 Rp(Ω) 0 3000 Rp(Ω) 0 5000
Ipv(A) 0 2 Ipv(A) 0 15 Ipv(A) 0 9
Io1(μA) 0 2 Io(μA) 0 1 Io1(μA) 0 2
Io2(μA) 0 2 a 0.5 2.5 Io2(μA) 0 2
a1 1 2 Io3(μA) 0 2
a2 1 2 a1 1 2
a2 1 3
a3 1 5

Note: LB is lower bounds and UB is the upper boundaries.


6 YOUSRI ET AL.

  
ðV + IRs Þ ðV + IRs Þ

0.00015132

0.00019391
0.00022802
0.00082216
4.5711e-05

9.7749e-05
dI = I g −I o1 exp −1 − −I: ð8Þ

0.0019025

0.0011438
aV t Rp

AEMPP
  
0 Rs ðV + IRs Þ ðR s Þ
dI = −I o1 exp −1 − −1 ð9Þ
aV t aV t Rp

−0.00010545
−1.2902e-17

−2.9273e-16
2.8623e-17

2.8623e-17
1.4456e-05

1.2531e-05
3.0047e-05
DiffRMSE
Similarly, for DDM and TDM the Iest, dI, and dI' can
be calculated.

The estimated parameters R.T.C. France cell obtained via the proposed approach under different irradiances and temperatures for SDM and DDM

RMSElambert
0.00077301

0.00076068
0.0021607
0.0010452
0.0014255

0.0012491
0.0011226
0.0017887
4 | EVALUATING THE RESULTS

In order to investigate the results obtained via the pro-


posed meta-heuristic approach, Lambert forms are

0.00077301

0.00074623
employed to calculate the current of the constructed

0.0021607
0.0010452
0.0014255

0.0013545
0.0011101
0.0017587
RMSE
model (SDM and DDM), accordingly RMSE values are cal-
culated. The maximum difference value among the RMSE
based on Lambert and for Equation (6) refers to inaccuracy
in the estimated parameters. Therefore, the corresponding

0.76079
0.76143
0.76059
0.76183
0.76081
0.76035
0.75952
0.76119
Iph(A)
fitness function is considered as an approximation one.
Regarding to SDM and DDM, the Lambert forms have
been presented whereas for TDM, it faces complexity in

2.0341e-07
2.3956e-07
1.3523e-07

2.2658e-07
providing its empirical formula. Therefore, the authors

Io2(A)
employed Lambert equation for SDM and DDM as follows:





Lambert form for SDM Equation (2)
  3.1067e-07

2.4365e-07
5.2224e-07
8.9974e-07
9.1286e-07
5.0049e-07
3.8853e-07
Rp I g + I o1 −V a1 V t
I Lambert = − W ðδÞ, where ð10aÞ 9.65e-07
Io1(A)

Rs + Rp Rs

 !
I o1 Rp Rs Rp Rs I g + Rs I o1 + V
δ=   exp   ð10bÞ
52.8891
72.4432
51.3267
49.7338
55.5501
64.5645
75.7797
67.0463
,
Rp(Ω)

a1 V t Rs + Rp a1 V t R s + R p

Lambert form for DDM Equation (4)


0.036547
0.031311
0.038051

0.037239
0.034875
0.037154
0.036546
0.03414
Rs(Ω)

RpðIoh + Io1 + Io2 − V Þ a1 Vt


ILambert = −r W ðδ 1 Þ
Rs + Rp Rs
ð11aÞ
a2 Vt
1.4111
1.4498
1.4549

− ð1 −r Þ W ðδ2 Þ, where
1.467

Rs




a2

h
i
I o1 exp ðVa+1 VIRt s Þ −1
1.4771
1.6004
1.4532
1.5314
1.8662
1.8645
1.9797
1.6909

r= h
i h
i ð11bÞ
a1

I o1 exp ðVa+1 VIRt s Þ − 1 − I o2 exp ðVa+2 VIRt s Þ −1


Cond/ Mod/Alg

 !
I o1 Rs Rp Rp Rs I g + Rs I o1 =r + V
δ1 =   exp   ð11cÞ
ra1 V t Rs + Rp a1 V t R s + R p
GWO

GWO
HHO

HHO
AEO

AEO
SSA

SSA

!
Parameters


Rp Rs I g + Rs I o2 =ð1− r Þ + V
TABLE 2

I o2 Rs Rp
δ2 =   exp   ,
ð1 − r Þa2 V t Rs + Rp a2 V t Rs + Rp
DDM
SDM

ð11dÞ
YOUSRI ET AL. 7

F I G U R E 3 The algorithms approaches performances for RTC-France solar cell of SDM (A) (I-V) characteristic, (B) (P-V) characteristic,
(C) absolute error curve, (D) mean convergence curve, and (E) RMSE values across 30 independent runs [Colour figure can be viewed at
wileyonlinelibrary.com]
8 YOUSRI ET AL.

FIGURE 4 Legend on next page.


YOUSRI ET AL. 9

TABLE 3 Statistical analysis followed to assess the performance of the proposed AEO algorithm

Metric Formula Metric Formula


qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PM ffi
2
i = 1 ðI m − I e Þ
1 RMSE
RMSE M Weighted RMSE (ξ) I sc
PM PM jI m −I e j
Summation absolute error (SAE) 1 j Im − Ie j Mean absolute error (MAE) 1 M
PM 2 PM I m − I e
Sum squared error (SSE) 1 ðI m − I e Þ Mean bias error (MBE) 1 M

Note: Im, Ie are the measured and estimated currents.

where ILambert is the calculated current via Lambert form, U 1 ðt + 1Þ = ð1 −dÞU n ðt Þ + dU rand ðt Þ, where ð13aÞ
W is Lambert W function solution. Accordingly RMSE
can be written as follow:  
t
d = 1− rand1 ð13bÞ
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi T max
u M
u1 X
RMSELambert = t ðI measi − I Lamberti Þ2 ð12Þ U rand ðt Þ = rand2 ðÞðUB−LBÞ + LB, ð13cÞ
M i=1

where LB and UB are the lower and upper bounds of the


5 | ARTIFICIAL ECOSYSTEM- optimized variables. rand1 and rand2 are random values
BASED O PTIMIZATION ( AEO) drawn in the range of [0, 1] and d is a weight
OVERVIEW coefficient.
2. Consumption operation: the consumer with low
The transformation process of the foods among the living energy or the producer are the source of the food in this
creatures is the basic idea in the Artificial ecosystem- stage. Some of the consumers feed on another consumer
based optimization (AEO) where the foods transfer via only and other feed on consumers and producer. Each
three sequential operations which are (1) production, type of those consumer (herbivores, omnivores, and car-
(2) consumption, and (3) decomposition.40 Zhao et al40 nivores) has its own feature in updating its location with
modeled these sequential operation mathematically in respect to its food source as presented below:
form of AEO optimizer. Zhao et al40 considered the pro- a. The herbivore consumers feed on the producers
duction operation is the responsible stage on achieving the only therefore they update their location based on the
balance among the algorithm exploration and exploitation producers location only as below:
capabilities. For the exploration phase, the consumption
operator devoted for discovering the search space whereas U i ðt + 1Þ = U i ðt Þ + KðU i ðt Þ −U 1 ðt ÞÞ ð14Þ
the decomposition operator is excluded in the exploitation
phase.40 Accordingly, Zhao et al40 assumed existing one where producer U1 refers to the producer location and
producer and decomposer agent while the other agents are K is a consumption factor which calculated via levy flight
recognized as consumers. The consumers can be catego- as below:
ries for three classes based on their type of foods as herbi-
vores (creatures eat plants), omnivores (creatures eat 1u
K= , , uNormð0, 1Þ, vNormð0, 1Þ ð15Þ
plants and animals), and carnivores creatures which eat 2v
animals only. Based on this hierarchy Zhao et al40 used the
following equations to proposed the AEO algorithm: where Norm(0, 1) is a normal distribution that has zero
1. Production operation: the producer is the worst mean with unity variance.
agent in the populations therefore it updates its solution b. The carnivore consumers feed on another animal
based on the best one in the search space as well as ran- only therefore they update their location based on a ran-
domly selected individual in the problem search space to dom consumer with high level energy with an index (l).
generate a new randomly individual agent in this land- The mathematical formula of updating the carnivore
space as defined below location can be written as below:

F I G U R E 4 The algorithms approaches performances for RTC-France solar cell of DDM (A) (I-V) characteristic, (B) (P-V)
characteristic, (C) absolute error curve, (D) mean convergence curve, and (E) RMSE values across 30 independent runs [Colour figure can be
viewed at wileyonlinelibrary.com]
10 YOUSRI ET AL.

TABLE 4 Statistical analysis by AEO vs EPSO, HCLPSO variants and recent published algorithms in case of R.T.C France solar cell and
CPU time (s)

Statistical analysis

Models/
Techs RMSE ± STD ξ SAE MAE SSE MBE CPU time (s)

SDM AEO 7.7301e-04± 1.6402e-05 1.01673e-04 1.7632e-02 6.7818e-04 1.5536e-05 −5.9995e-09 40.5233

EPSO 8.0621e − 04 ± 4.3109e − 04 1.0604e − 03 1.8787e − 02 7.2259e − 04 1.6900e − 05 −1.0023e − 05 13.6707

PGJAYA 10
9.8602e − 04 ± 1.4485e − 09 1.2966e − 03 − 1.2969e − 03 − − 41

CWOA 43
9.98678e − 04 ± 6.33831e − 03 1.3132e − 03 − − − − −

PSO-WOA43 1.07101e − 03 ± 1.17001e − 03 1.4083e − 03 − − − − −

SATLBO 17
9.86022e − 04 ± 2.3002e − 06 1.2965e − 03 − − − − −

ELPSO 27
7.7301e − 04 ± 3.4508e − 07 1.0164e − 03 − − − − −

HFAPS 44
9.8602e − 04 ± − 1.2965e-03 − 7.5e − 03 − − −

MLBSA7 9.8602e − 04 ± 9.1461e − 12 1.2965e-03 − − − − 44

TVACPSO 47
7.7301e − 04 ± 5.5805e − 10 — − − − − −

CPSO 47
7.7301e − 04 ± 2.8344e − 5 — − − − − −

GSA 45
4.1020e − 03 ± 2.7197e − 03 — − − − − −

CSA46 9.86023e − 04 ± 8.570571e − 06 — − − − − −

ICSA 46
9.8602e − 04 ± 2.987589e − 12 — − − − − −

GA 51
19.08e − 03 ± − — − − − − −

CS 52
1.000e − 03 ± − — − 7.00e − 4 − − −

HCLPSO29 7.89580e − 04 ± 3.0090e − 04 1.03851e-03 1.83898e-02 7.07300e-04 1.62093e-05 6.48591e-07 204.5567

ISCE 6
9.86022e − 04 ± − 1.29688e-03 1.77041e-02 6.80926e-04 — —

CWOA 11
9.8600e − 04 ± − 1.29686e-03 2.15280e-02 8.2800e-04 — 7.2800e-08

BMO 11
9.8608e − 4 ± − 1.29696e-03 2.20480e-02 8.4800e-004 — 1.1302e-06

STLBO11 9.8602e − 04 ± − 1.29688e-03 2.15540e-02 8.2900e-04 — 4.0948e-7

EHA-NMS 26
9.8602e − 04 ± − 1.2965e-03 1.7704e-02 6.80923e-04 — —

NM-MPSO 53
9.8602e − 04 ± − 1.2965e-03 1.19548e-01 4.5980e-03 — —

PPSO 50
— — 1.77798e-02 6.8384e-04 — —

FPA8 7.7301e − 04 ± − 1.01672e-03 1.5971e-02 6.14269e-04 5.181638e-06 —

MPSO49 9.8602e − 04 ± − 1.2965e-03 — — — —

CPSO 48
2.6500e − 03 ± − 3.48547e-03 1.6800e-03 1.6800e-03 — 6.100e-05

DDM AEO 7.6068e − 04 ± 1.4844e − 04 1.0005e-03 1.7042e-02 6.5546e-04 1.4478e-05 4.5618e-06 10.4134

EPSO35 7.6312e − 04 ± 1.5424e − 04 1.0037e − 03 1.7439e − 02 6.7075e − 04 1.5141e − 05 −6.5272e − 06 15.4857

PGJAYA10 9.8263e − 04 ± 2.5375e − 06 1.4859e − 03 − − − − −

CWOA 43
1.130041e − 03 ± 2.153407e − 03 2.1959e − 03 − − − − −

PSO-WOA 43
1.669996e − 03 ± 9.551144e − 04 1.2923e − 03 − − − − −

SATLBO17 9.828037e − 04 ± 1.951533e − 05 9.7620e − 04 − − − − −

ELPSO 27
7.4240e − 04 ± 9.4291e − 06 1.2919e − 03 − − − − −

HFAPS 44
9.8248e − 04 ± − 1.2919e − 03 − 7.1e − 03 − − 39

MLBSA 7
9.8249e − 04 ± 1.3482e − 06 — − − − − −

GA45 5.91958e − 03 ± 1.818e − 03 — — — — — —

CSA 46
9.8292e − 04 ± 4.1755e − 06 — — — — — —

ICSA 46
9.8249e − 04 ± 2.8197e − 07 — — — — — —

HCLPSO 29
7.6680e − 04 ± 1.8624e − 04 1.0086e-03 1.7399e-02 6.6919e-04 1.5288e-05 −1.3767e-05 265.3884
YOUSRI ET AL. 11

TABLE 4 (Continued)

Statistical analysis

Models/
Techs RMSE ± STD ξ SAE MAE SSE MBE CPU time (s)

ISCE 6
9.8248e-04 1.2922e-03 1.7319e-02 — — —

CWOA11 9.8279e-04 1.2926e-03 2.1294e-02 8.1900e-04 — 1.23263e-07

BMO 11
9.8262e − 04 ± − 1.2924e-03 2.1554e-02 8.2900e-04 — 5.8807e-07

STLBO 11
9.8248e − 04 ± − 1.2922e-03 2.3348e-02 8.9800e-04 — 1.3684e-07

TVACPSO47 7.4365e − 04 ± − 9.7784e-04 — — — —

EHA-NMS26 9.8248e − 04 ± − 1.2923e-03 1.7319e-02 6.6612e-04 — -

NM-MPSO 53
9.8250e − 04 ± − 1.2919e-03 11.6610e-02 4.4850e-03 — -

PPSO50 — — 1.7267e-02 6.6415e-04 — -

FPA 8
7.8425e − 04 ± − 1.0312e-03 1.7298e-02 6.6531e-04 1.6873e-05 -

MPSO 49
9.8247e − 04 ± − 1.2919e-03 — — — -

Note: (−) For the not available data in the corresponded paper.

TABLE 5 The estimated parameters FSM-25 PV solar panel by proposed techniques under different irradiance and temperatures
for SDM

Parameters

Cond/Mod
/Alg a1 Rs(Ω) Rp(Ω) Io1(A) Iph(A) RMSE RMSElambert DiffRMSE AEMPP
SDM AEO 1.8846 0.58203 2968.2654 4.5348e − 05 1.3156 0.0065206 0.0065529 3.2297e − 05 0.073598
HHO 2.0309 0.39057 2853.1507 9.5957e − 05 1.3184 0.0073067 0.0073379 3.1203e − 05 0.15338
GWO 1.8686 0.59075 1902.1776 4.1488e − 05 1.3166 0.0065497 0.0065371 −1.2545e − 05 0.070386
SSA 1.8603 0.58814 1630.5361 3.9577e − 05 1.3163 0.0065999 0.006632 3.2036e − 05 0.067767

U i ðt + 1Þ = U i ðt Þ + KðU i ðt Þ −U l ðt ÞÞ, where ð16aÞ U i ðt + 1Þ = U i ðt Þ + DðeU n ðt Þ −hU i ðt ÞÞ, i = 1, …, N, where


ð18aÞ
l = randið½2i−1Þ,i = 3, …, N, ð16bÞ
D = 3u, uN ð0, 1Þ ð18bÞ
c. The Omnivore consumers feed on both the producer
and consumers therefore they update their location using the e = rand4 randið½1 2Þ− 1, ð18cÞ
producer position and randomly selected consumer with
index (l) and with higher level of energy as given below: h = 2rand4 −1, ð18dÞ

Ui ðt + 1Þ = Ui ðt Þ + Kðrand3 ðUi ðtÞ −U1 ðt ÞÞÞ where D is the decomposition factor, h and e are the
ð17aÞ weight parameters.
+ ð1 −rand3 ÞðUi ðtÞ −Ul ðt ÞÞ, where
The flowchart of the proposed AEO depicts on
l = randið½2i−1Þ,i = 3, …, N, ð17bÞ Figure 2 to illustrate the structure of the AEO-based PV
parameters estimation process. The algorithm starts with
generating an initial solutions for the PV model parame-
3. Decomposition process: it is the last process in the ters then the RMSE via Equation (6) has been computed.
ecosystem where the decomposer feed on the remains of The estimated current obtained via solving Equation (7)
the died agents. Zhao et al40 realized that in this process based on the identified parameters. Then the solutions
the algorithm performs its exploitation capability to con- have been modified based on the main structure of AEO
verge for the optimal solution as given below: algorithm until the termination criteria has been met.
12 YOUSRI ET AL.

F I G U R E 5 The algorithms approaches performances for FSM-25 PV solar panel (A) (I-V) characteristic, (B) (P-V) characteristic,
(C) absolute error curve, (D) mean convergence curve, and (E) RMSE values across 30 independent runs [Colour figure can be viewed at
wileyonlinelibrary.com]
YOUSRI ET AL.

TABLE 6 The estimated parameters CS6P-240P solar module by proposed techniques under different irradiance and temperatures for DDM and TDM

Parameters

Cond/Mod/Alg a1 a2 a3 Rs(Ω) Rp(Ω) Io1(A) Io2(A) Io3(A) Iph(A) RMSE RMSElambert DiffRMSE AEMPP
2
673.5 W/m , DDM AEO 1.1635 4.764 — 0.31127 315.075 1.6992e − 07 4.0582e − 07 — 6.9848 0.028487 0.028487 −1.3706e − 08 0.071084
45.92 ∘C HHO 1.239 1.2554 — 0.29856 2689.9386 4.6208e − 07 4.2034e − 08 — 6.9536 0.033909 0.03391 4.8672e − 07 0.47437
GWO 1.1153 4.0846 — 0.3253 258.382 7.9781e − 08 4.6593e − 07 — 6.989 0.029892 0.029892 −2.491e − 07 0.28051
SSA 1.2529 3.8288 — 0.28078 1554.168 5.9983e − 07 3.0496e − 07 — 6.9524 0.030669 0.030669 −1.5715e − 07 0.4184
TDM AEO 1.163 2.9876 3.5586 0.31149 313.7978 1.6866e − 07 1.8524e − 07 2.5812e − 07 6.9852 0.028487 — — 0.066118
HHO 1.1574 2.5144 2.5962 0.30363 1599.6091 1.5731e − 07 7.7785e − 08 4.9075e − 07 6.9364 0.037768 — — 0.79602
GWO 1.1815 2.4468 3.9519 0.30306 270.3277 2.2167e − 07 1.1371e − 07 4.2953e − 08 6.9895 0.029233 — — 0.47076
SSA 1.1918 2.6839 2.5052 0.30946 973.8534 2.5838e − 07 1.8568e − 07 7.7269e − 07 6.9563 0.030313 — — 0.011843
2
347.8W/m , DDM AEO 1.1033 3.0964 — 0.34327 663.0711 3.7329e − 08 1.1161e − 07 — 3.0402 0.015148 0.015148 −1.4601e − 08 0.086586
43.95 ∘C HHO 1.0118 1.3801 — 0.43027 2720.246 6.5074e − 09 1.3346e − 07 — 3.0246 0.018686 0.018886 0.00019965 0.34561
GWO 1.8293 1 — 0.42152 601.9616 9.0974e − 07 5.565e − 09 — 3.0396 0.016151 0.016201 4.9608e − 05 0.29257
SSA 1.299 2.8876 — 0.14576 3207.0614 5.9169e − 07 3.6055e − 07 — 3.0271 0.023728 0.023727 −1.0773e − 06 0.52374
TDM AEO 1.1069 2.6588 2.8927 0.34068 705.1794 3.9606e − 08 4.7689e − 12 2.1848e − 08 3.0391 0.015157 — — 0.076422
HHO 1.3456 2.5925 3.7072 0.16626 1961.0107 9.9991e − 07 9.9991e − 07 9.9991e − 07 3.0377 0.021712 — — 1.0549
GWO 1.1513 2.669 3.2013 0.30766 2114.7241 8.0489e − 08 2.6105e − 07 4.753e − 08 3.0304 0.015956 — — 0.032393
SSA 1.0846 2.3338 4.8289 0.38097 3864.5112 2.7351e − 08 8.1702e − 07 4.0798e − 07 3.0271 0.017038 — — 0.36461
109.2 W/m2, DDM AEO 1.0232 2.1834 — 0.68879 482.6963 3.5682e − 09 — 2.0935e − 08 0.99776 0.0036421 0.003642 −4.5151e − 08 0.034826
37.32 ∘C HHO 1.3644 1.7096 — 0.0019576 744.6646 4.575e − 07 — 1.124e − 07 0.99394 0.0057763 0.0057763 −1.9326e − 09 0.38843
GWO 1.2604 4.7914 — 0.20458 846.6237 1.4098e − 07 — 9.1601e − 08 0.99375 0.005341 0.005341 −6.3432e − 10 0.040251
SSA 1.2807 3.4991 — 0.43672 1878.1577 1.8234e − 07 — 4.575e − 07 0.99046 0.007656 0.0076561 1.4517e − 07 0.016281
TDM AEO 1.0487 2.5487 4.9253 0.60401 455.3895 5.6883e − 09 6.8265e − 07 2.5339e − 07 0.9981 0.003696 — — 0.079783
HHO 1.2563 2.025 2.0217 0.31942 763.8902 1.3211e − 07 3.6182e − 08 1.0936e − 07 0.99399 0.0055414 — — 0.20803
GWO 1.3458 2.773 2.232 0.0063652 1259.4814 3.8544e − 07 1.5497e − 07 1.0983e − 07 0.99087 0.0062726 — — 0.032709
SSA 1.3872 2.3969 3.0739 0.14379 1349.2457 5.9427e − 07 8.7248e − 07 2.8701e − 07 0.99162 0.0072735 — — 0.29747

Note: DiffRMSE is the difference among RMSElambert and RMSE of Equation (6).
13
14 YOUSRI ET AL.

FIGURE 6 Legend on next page.


YOUSRI ET AL. 15

6 | S I MULATIONS A ND ANALYSES function and estimated parameters. The large deviation


among the two RMSE values refers to uncertainly in the
To clarify the competence of the proposed approach for identified parameters. Furthermore, detecting the maxi-
determining the parameters of PV equivalent circuit of mum power point is an essential target in the PV model-
SDm, DDM, and TDM models, several series of experi- ing process, therefore the absolute error at the maximum
ments have been established as listed below. power point (AEMPP) has been computed in the Table 2.
1. Experiment series 1: A set of measured data of a The results reveal that, the AEO algorithm provides the
commercial R.T.C. France silicon solar cell at 1000W/m2 least values of the RMSE of 7.77301 × 10−4 and
and 33 C. The electric specifications of the cell are 7.4623 × 10−4 for SDM and DDM respectively while the
Isc = 0.7605 [A], Voc = 0.5727 [V], Imp = 0.6755 [A] and GWO occupied the second rank with RMSE of
Vmp = 0.4590 [V].41 The length of the dataset is 2.1607 × 10−3 and 1.1226 × 10−3. The identified parame-
26 samples. ters by AEO illustrate more accurate and precise fitting
2. Experiment series 2: A dataset was measured for based on the deviation from RMSELambert that reach for
FSM-25 PV module at 45 C, the length of the dataset is 10−17 for SDM and 10−5 for DDM with most efficient
36 samples. The electric specifications of the cell are detection for the MPP as the AEMPP has values of 10−5 for
Isc = 0.7605 [A], Voc = 0.5727 [V], Imp = 0.6755 [A] and both of the considered models.
Vmp = 0.4590 [V].42 To check on the efficiency of the fitting process,
3. Experiment series 3: Three measured datasets for current-voltage and power-voltage curves have been plot-
Canadian-Solar-(CS6P-240P) multi-crystalline solar panel ted in Figures 3A,B and 4A,B for both SDM and DDM.
at different solar radiations and temperatures that have As illustrated from the obtained curves the identified
profiles of 673.5, 347.8, and 109.2W/m2 with 45.92, 43.95, parameters provide an accurate fitting for the dataset
and 37.32 C respectively. The length of the data is while AEO exhibits a least deviation from the experimen-
128 samples. The electric specifications of the panel at tal dataset and followed by GWO as illustrated from Fig-
STC are Isc = 8.59 [A], Voc = 37 [V], Imp = 8.03 [A] and ures 3C and 4C of the absolute error curves across the
Vmp = 29.9 [V]. dataset points.
The used instruments in the measuring process are The speed of the algorithm in offering the optimal
I-V 400 PHOTOVOLTAIC PANEL ANALYZER with solutions and its consistency in providing those solutions
HT304N radiating sensor and PT300 temperature sensor. for several independent runs are essential issues that
The software of TOPVIEW was utilized to transfer the should be addressed. Therefore, the mean convergence
measured data to personal PC. curves and the RMSE values for 30 independent runs are
The employed algorithms are implemented for 30 runs plotted in Figures 3D, 4D and 3E, 4E for SDM and DDM
with 30 and 500 population size and number of iterations respectively. The drawn figures confirm that, the AEO
respectively. All the simulations and analyses are per- has capability to converge for the optimal solutions faster
formed and established on Core i7-6500U CPU laptop than the HHO, GWO, and SSA with highest consistence
with 16 GB of RAM and 2.5 GHz of speed. The lower and performance as depicted in Figures 3E and 4E for both
upper bounds of the models variables are listed in the studied models.
Table 1. From the previous discussion, the AEO proved its effi-
ciency in providing accurate parameters with catching
the least RMSE and AEMPP as well as better fitting for the
6.1 | Experimental series 1: dataset with the highest acceleration rate and reliability
R.T.C. France silicon solar cell in the results.
For further evaluation for the algorithm performance
The obtained SDM and DDM parameters by the proposed it has been compared with numerous recent optimizers in
algorithms are reported in Table 2 with the literature using several statistical measures as reported in
corresponding values of the RMSE based on Equation (6). Table 3. The considered algorithms included a wide range
To validate the accuracy and precision of the identified of the proposed algorithms across the previous decades
parameter, the RMSELambert has been computed as in that are fractional chaotic ensemble particle swarm opti-
Table 2 with using the current based on Lambert W mizer (EPSO),35 chaotic heterogeneous comprehensive

F I G U R E 6 The algorithms performances for CS6P-240P solar module of DDM (A) (I-V) characteristic, (B) (P-V) characteristic,
(C) absolute error curve, (D) mean convergence curve, and (E) RMSE values across 30 independent runs [Colour figure can be viewed at
wileyonlinelibrary.com]
16 YOUSRI ET AL.

FIGURE 7 Legend on next page.


YOUSRI ET AL. 17

TABLE 7 Wilcoxon signed rank test results AEO vs HHO, GWO or SSA

AEO VS HHO AEO VS GWO AEO VS SSA


Exp/model
R+ R− p-Value h0 R+ R− p-Value h0 R+ R− p-Value h0
2 
Exp series 1 1000 W/m , 33 C SDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
DDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
2 
Exp series 2 1000 W/m , 45 C SDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
2 
Exp series 3 673.5 W/m , 45.92 C DDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
TDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
347.8 W/m2, 43.95 C DDM 465 0 1.7344e-06 No 464 1 1.9209e-06 No 465 0 1.7344e-06 No
TDM 465 0 1.7344e-06 No 463 2 2.1266e-06 No 465 0 1.7344e-06 No
2 
109.2 W/m , 37.32 C DDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No
TDM 465 0 1.7344e-06 No 465 0 1.7344e-06 No 465 0 1.7344e-06 No

learning PSO (HCLPSO),29 performance-guided JAYA 0.0065206 and 0.0065371 respectively while SSA displays
(PGJAYA),10 improved whale optimization algorithm var- the least value of the AEMPP.
iants (CWOA) and (PSO-WOA),43 self-adaptive teaching- For meta-heuristic algorithms, one should investigate
learning-based optimization (STLBO),17 chaotic whale the consistency of the algorithm that divulges if this solu-
optimizer (CWOA),11 improved shuffled complex evolu- tion occurred by chance or the algorithm is reliable.
tion algorithm (ISCE),6 hybrid firefly and pattern search Therefore, the mean convergence curves and the
algorithms (HFAPS),44 multiple learning backtracking obtained RMSE by the considered approaches have been
search algorithm (MLBSA),7 gravitational search algo- plotted in Figure 5D, E. The figures reveal that, the AEO
rithm (GSA),45 improved cuckoo search algorithm (CSA) is the most reliable and stable algorithm in comparable
and (ICSA),46 time varying acceleration coefficients parti- with the HHO, GWO, and SSA, Accordingly AEO is the
cle swarm optimisation (TVACPSO),47 chaos PSO most applicable one.
(CPSO),48 modified PSO (MPSO),49 parallel PSO (PPSO),50
genetic algorithm (GA),51 and cuckoo search (CS).52 The
results of the Table 4 reveal the efficiency of the AEO in 6.3 | Experimental series 3: Canadian-
providing higher accurate results with shorter execution Solar-(CS6P-240P) multi-crystalline solar
time in comparing with PGJAYA, MLBSA and HCLPSO panel
across the SDM and DDM.
In this section, three measured datasets under different
levels of irradiation and temperature for CS6P-240P PV
6.2 | Experimental series 2: FSM-25 PV module have been utilized to assess the quality of the
solar panel proposed AEO while dealing with DDM and TDM. The
extracted parameters of the considered models and the
In the second series of experiments, the proposed algo- corresponding values of RMSE, RMSELambert, and differ-
rithms have been implemented for identifying the param- ence value among the two RMSE as well as the AEmpp
eter of SDM of FSM-25 PV module. The identified are reported in Table 6. The attached results are the evi-
parameters are listed with RMSE, RMSELambert, and devi- dence for the applicability of AEO as it provides the least
ation value among the obtained RMSE in the Table 5, RMSE values with less deviation from RMSELambert for
moreover, the AEMPP value has been calculated and tabu- the first two datasets and it shows a comparable perfor-
lated. Furthermore, the I-V, P-V characteristics, and AE mance with GWO for the third dataset. The drawn I-V,
curves are plotted in Figure 5A-C respectively. The P-V, and AE curves of Figures 6C and 7C illustrate that,
inserted results show that, AEO and GWO have compa- AEO provides better fitting for the measured data as
rable performance in providing the least RMSE value of AEO-AE curves expose the least values across the curve

F I G U R E 7 The algorithms performances for CS6P-240P solar module of TDM (A) (I-V) characteristic, (B) (P-V) characteristic,
(C) absolute error curve, (D) mean convergence curve, and (E) RMSE values across 30 independent runs [Colour figure can be viewed at
wileyonlinelibrary.com]
18 YOUSRI ET AL.

points for the three employed datasets especially in com- to identify the unknown parameters of the PV cell and mod-
parison with HHO, and SSA. The mean convergence cur- ule equivalent circuits. This is achieved via minimizing
ves and the box-plots of RMSE values with number of objective function of the root mean square error (RMSE)
runs manifest the robust and efficiency of AEO as it dis- between the measured and the estimated current that based
plays a highest acceleration speed and highest uniformity on the estimated parameters. The considered PV models
in the distribution of the obtained RMSE across 30 inde- are single diode (SDM), double diode (DDM), and triple
pendent runs for DDM and TDM. diode (TDM)-based circuit. The analysis is performed on
which are R.T.C. France silicon solar cell, FSM-25 PV mod-
ule, and Canadian-Solar-(CS6P-240P) multi-crystalline
7 | S T ATI S T I C AL AN A LY S I S solar panel with the aid of experimental data under differ-
ent operating conditions. The obtained results via the pro-
In this section, the AEO quality has been proven based posed AEO are compared to Harris hawks optimizer
on statistical analysis of Wilcoxon signed rank test across (HHO), gray wolf optimizer (GWO), salp swarm algorithm
the three series of experiments. Wilcoxon signed rank test (SSA) and with more than 20 algorithm from literature.
has been performed among the AEO and the other peers Lambert form is employed to validate the constructed PV
based on the following criteria54: model. For R.T.C. France silicon solar cell, the proposed
1. Store the RMSE values for 30 runs obtained via the AEO succeeded in achieving RMSE of 7.77301 × 10−4 and
employed techniques (AEO vs HHO, AEO vs GWO and 7.4623 × 10−4 for SDM and DDM, respectively. Addition-
AEO vs SSA). ally, the proposed AEO outperformed others by achieving
2. Calculate R+ which denotes the ranks sum for runs RMSE of 0.0065206 for FSM-25 PV module. Regarding to
in which AEO exposes the best performance (minimum Canadian-Solar-(CS6P-240P) multi-crystalline solar panel,
RMSE) in comparison with the other competitor (HHO, three experimental datasets are considered, the minimum
GWO or SSA). RMSE is obtained via the proposed AEO of 0.0036421 and
3. Calculate R− which denotes the ranks sum for runs 0.003696 for DDM and TDM, respectively. Finally, Statisti-
in which the peers (HHO, GWO or SSA) give minimum cal analysis using Wilcoxon signed rank test is implemented
RMSE values in comparison with AEO. across the three series of experiments for all employed opti-
4. Calculate P-value which detects the significance mizers. The obtained results confirmed the competence of
difference of the results in a statistical hypothesis test. the proposed approach in determining the optimal parame-
The smaller the P-value is, the more evidence against ters of PV cell and modules equivalent circuits.
null hypothesis. For future work, first, the authors will study several
Null hypothesis is valid when H0 = “Yes,” with P-value PV models to achieve the highest representation for the
is greater than .05, which denotes similar statistical perfor- physical and electrical repose of numerous PV solar mod-
mance of the two handled algorithms. On the other hand, ules types. Moreover, formulating the PV optimization
there are significant difference between the algorithms if problem as a multi-objective optimization problem is
P-value is less than .05. In such case, H0 = “No,” accounted for as the second future plan.
The computed R+, R−, P-value, and H0 are listed in
Table 7 for the considered series of experiments and studied NOMENCLATURE
models. The results show that AEO has a significant differ-
ence from the GWO, HHO, and SSA in favor of AEO. AEO Acronyms
has the highest ranks values R+ where it provides the least ABC artificial bee colony
RMSE in comparison with GWO, HHO, and SSA as well as AEO artificial ecosystem-based optimizer
the obtained P-value confirm the superiority of AEO (P- ANFIS adaptive neuro-fuzzy inference system
value <.05). Consequently, based on this pairwise compari- BFA bacteria foraging algorithm
sons the AEO approved its efficiency and reliability. CHCLPSO chaotic heterogeneous comprehensive
learning particle swarm optimizer
CS cuckoo search
8 | C ON C L U S I ON S CSO cat swarm optimizer
DDM double diode model
Construction of the PV module equivalent circuit simulat- DE differential evolution
ing the practical operation is a big challenging especially FC-EPSO fractional chaotic ensemble particle swarm
with the lack of some parameters given by the manufac- optimizer
turer. This article proposed a recent meta-heuristic FPA flower pollination algorithm
approach of artificial ecosystem-based optimization (AEO) GA genetic algorithm
YOUSRI ET AL. 19

GOA grasshopper optimization algorithm 2. Ortiz-Conde A, Sánchez FJG, Muci J. New method to extract
GWO gray wolf optimizer the model parameters of solar cells from the explicit analytic
HHO Harris Hawks optimizer solutions of their illuminated I–V characteristics. Solar Energy
Mater Solar Cells. 2006;90(3):352-361.
I-V current-voltage curve
3. Wolf P, Benda V. Identification of PV solar cells and modules
ICA imperialist competitive algorithm parameters by combining statistical and analytical methods.
IEM electromagnetism like algorithm Solar Energy. 2013;93:151-157.
MLBSA multiple learning backtracking search 4. Hejri M, Mokhtari H, Azizian MR, Ghandhari M, Soder L. On
algorithm the parameter extraction of a five-parameter double-diode
MPP maximum power point model of photovoltaic cells and modules. IEEE J Photovoltaics.
P-V power-voltage curve 2014;4(3):915-923.
5. Laudani A, Fulginei FR, Salvini A. High performing extraction
PGJAYA performance-guided JAYA
procedure for the one-diode model of a photovoltaic panel from
PSO particle swarm optimization
experimental I–V curves by using reduced forms. Solar Energy.
PV photovoltaic 2014;103:316-326.
RESs renewable energy sources 6. Gao X, Cui Y, Hu J, et al. Parameter extraction of solar cell
RMSE root mean square error models using improved shuffled complex evolution algorithm.
SBMO simplified bird mating optimizer Energy Conver Manage. 2018;157:460-479.
SDM single diode model 7. Yu K, Liang J, Qu B, Cheng Z, Wang H. Multiple learning
SFLA shuffled frog leaping algorithm backtracking search algorithm for estimating parameters of
photovoltaic models. Appl Energy. 2018;226:408-422.
SSA salp swarm algorithm
8. Alam D, Yousri D, Eteiba M. Flower pollination algorithm
TDM triple diode model based solar PV parameter estimation. Energy Convers Manage.
TLBO teaching-learning-based optimization 2015;101:410-422.
WOA whale optimization algorithm 9. Fathy A, Rezk H. Parameter estimation of photovoltaic system
using imperialist competitive algorithm. Renew Energy. 2017;
Constants 111:307-320.
k Boltzmann constant 1.35 * 10−23J/K 10. Yu K, Qu B, Yue C, Ge S, Chen X, Liang J. A performance-
q charge of electron (1.6 × 10−19 C) guided JAYA algorithm for parameters identification of photo-
voltaic cell and module. Appl Energy. 2019;237:241-257.
11. Oliva D, El Aziz MA, Hassanien AE. Parameter estimation of
Variables photovoltaic cells using an improved chaotic whale optimiza-
Ig generated-light current tion algorithm. Appl Energy. 2017;200:141-154.
Id diode current 12. Chin VJ, Salam Z, Ishaque K. Cell modelling and model
ILambert calculated current via Lambert form parameters estimation techniques for photovoltaic simulator
Imp current at maximum power point application: a review. Appl Energy. 2015;154:500-519.
Io ,Io1, Io2, Io3 saturation currents 13. Askarzadeh A, dos Santos Coelho L. Determination of photo-
voltaic modules parameters at different operating conditions
Ip shunt-leakage current
using a novel bird mating optimizer approach. Energy Convers
Isc short circuit current
Manage. 2015;89:608-614.
Ns number of cells in series 14. Chikh A, Chandra A. An optimal maximum power point track-
Rp shunt resistance ing algorithm for PV systems with climatic parameters estima-
Rs series resistance tion. IEEE Trans Sustain Energy. 2015;6(2):644-652.
Vmp voltage at maximum power point 15. Ram JP, Babu TS, Dragicevic T, Rajasekar N. A new hybrid bee
Voc open circuit voltage pollinator flower pollination algorithm for solar PV parameter
Vt thermal voltage estimation. Energy Convers Manage. 2017;135:463-476.
16. Jordehi AR. Parameter estimation of solar photovoltaic
a,a1,a2,a3 ideality factors
(PV) cells: a review. Renew Sustain Energy Rev. 2016;61:354-371.
T temperature of a PV cell in Kelvin 17. Yu K, Chen X, Wang X, Wang Z. Parameters identification
of photovoltaic models using self-adaptive teaching-learning-
ORCID based optimization. Energy Convers Manage. 2017;145:
Dalia Yousri https://orcid.org/0000-0001-6551-2371 233-246.
Hegazy Rezk https://orcid.org/0000-0001-9254-2744 18. Abbassi R, Abbassi A, Heidari AA, Mirjalili S. An efficient salp
Ahmed Fathy https://orcid.org/0000-0002-8340-4056 swarm-inspired algorithm for parameters identification of pho-
tovoltaic cell models. Energy Convers Manage. 2019;179:
362-372.
R EF E RE N C E S
19. Xu S, Wang Y. Parameter estimation of photovoltaic modules
1. Silva EA, Bradaschia F, Cavalcanti MC, Nascimento AJ. Parame- using a hybrid flower pollination algorithm. Energy Convers
ter estimation method to improve the accuracy of photovoltaic Manage. 2017;144:53-68.
electrical model. IEEE J Photovoltaics. 2015;6(1):278-285.
20 YOUSRI ET AL.

20. Allam D, Yousri D, Eteiba M. Parameters extraction of the optimizer for identifying the single, double, and three diode
three diode model for the multi-crystalline solar cell/module photovoltaic models' parameters. Energy. 2020;195:116979.
using moth-flame optimization algorithm. Energy Convers 36. 
Calasan M, Aleem SHA, Zobaa AF. On the root mean square
Manage. 2016;123:535-548. error (RMSE) calculation for parameter estimation of photovol-
21. Chen X, Xu B, Mei C, Ding Y, Li K. Teaching–learning-based taic models: a novel exact analytical solution based on Lambert
artificial bee colony for solar photovoltaic parameter estima- W function. Energy Convers Manag. 2020;210:112716.
tion. Appl Energy. 2018;212:1578-1588. 37. Chin VJ, Salam Z. Coyote optimization algorithm for the
22. Nassar-Eddine I, Obbadi A, Errami Y, Agunaou M, et al. parameter extraction of photovoltaic cells. Solar Energy. 2019;
Parameter estimation of photovoltaic modules using iterative 194:656-670.
method and the Lambert W function: a comparative study. 38. Nunes H, Pombo J, Bento P, Mariano S, Calado M. Collabora-
Energy Convers Manage. 2016;119:37-48. tive swarm intelligence to estimate PV parameters. Energy Con-
23. Ma J, Bi Z, Ting TO, Hao S, Hao W. Comparative performance vers Manage. 2019;185:866-890.
on photovoltaic model parameter identification via bio-inspired 39. Ibrahim MN, Rezk H, Al-Dahifallah M, Sergeant P. Hybrid
algorithms. Solar Energy. 2016;132:606-616. photovoltaic-thermoelectric generator powered synchronous
24. Hasanien HM. Shuffled frog leaping algorithm for photovoltaic reluctance motor for pumping applications. IEEE Access. 2019;
model identification. IEEE Trans Sustain Energy. 2015;6(2): 7:146979-146988.
509-515. 40. Zhao W, Wang L, Zhang Z. Artificial ecosystem-based optimi-
25. Guo L, Meng Z, Sun Y, Wang L. Parameter identification zation: a novel nature-inspired meta-heuristic algorithm. Neu-
and sensitivity analysis of solar cell models with cat swarm ral Comput Appl. 2020;32:9425-9383.
optimization algorithm. Energy Convers Manage. 2016;108: 41. Easwarakhanthan T, Bottin J, Bouhouch I, Boutrit C.
520-528. Nonlinear minimization algorithm for determining the solar
26. Chen X, Yu K, Du W, Zhao W, Liu G. Parameters identification cell parameters with microcomputers. Int J Solar Energy. 1986;
of solar cell models using generalized oppositional teaching 4(1):1-12.
learning based optimization. Energy. 2016;99:170-180. 42. Rezk H, Tyukhov I, Al-Dhaifallah M, Tikhonov A. Perfor-
27. Jordehi AR. Enhanced leader particle swarm optimisation mance of data acquisition system for monitoring PV system
(ELPSO): an efficient algorithm for parameter estimation of parameters. Measurement. 2017;104:204-211.
photovoltaic (PV) cells and modules. Solar Energy. 2018;159: 43. Xiong G, Zhang J, Shi D, He Y. Parameter extraction of solar
78-87. photovoltaic models using an improved whale optimization
28. Chen H, Jiao S, Wang M, Heidari AA, Zhao X. Parameters algorithm. Energy Convers Manage. 2018;174:388-405.
identification of photovoltaic cells and modules using 44. Beigi AM, Maroosi A. Parameter identification for solar cells
diversification-enriched Harris hawks optimization with cha- and module using a hybrid firefly and pattern search algo-
otic drifts. J Clean Prod. 2020;244:118778. rithms. Solar Energy. 2018;171:435-446.
29. Yousri D, Allam D, Eteiba M, Suganthan PN. Static and 45. Jordehi AR. Gravitational search algorithm with linearly
dynamic photovoltaic models' parameters identification using decreasing gravitational constant for parameter estimation of
chaotic heterogeneous comprehensive learning particle swarm photovoltaic cells. IEEE Congress Evol Comput. 2017;2017:
optimizer variants. Energy Convers Manage. 2019;182:546-563. 37-42.
30. Cuevas E, Gálvez J, Avalos O. Comparison of solar cells param- 46. Kang T, Yao J, Jin M, Yang S, Duong T. A novel improved
eters estimation using several optimization algorithms. Recent cuckoo search algorithm for parameter estimation of photovol-
Metaheuristics Algorithms for Parameter Identification. Cham, taic (PV) models. Energies. 2018;11(5):1060.
Switzerland: Springer; 2020:51-95. 47. Jordehi AR. Time varying acceleration coefficients particle
31. Chenouard R, El-Sehiemy RA. An interval branch and bound swarm optimisation (TVACPSO): a new optimisation algorithm
global optimization algorithm for parameter estimation of for estimating parameters of PV cells and modules. Energy Con-
three photovoltaic models. Energy Convers Manag. 2020;205: vers Manage. 2016;129:262-274.
112400. 48. Wei H, Cong J, Lingyun X, Deyun S. Extracting solar cell
32. Ridha HM, Gomes C, Hizam H. Estimation of photovoltaic model parameters based on chaos particle swarm algorithm.
module model's parameters using an improved Paper presented at: 2011 International Conference on IEEE
electromagnetic-like algorithm. Neural Comput Appl. 2020; Electric Information and Control Engineering (ICEICE); 2011:
1-16. 398-402.
33. Elazab OS, Hasanien HM, Alsaidan I, Abdelaziz AY, 49. Hachana O, Hemsas K, Tina G, Ventura C. Comparison of dif-
Muyeen S. Parameter estimation of three diode photovoltaic ferent metaheuristic algorithms for parameter identification of
model using grasshopper optimization algorithm. Energies. photovoltaic cell/module. J Renew Sustain Eenergy. 2013;5(5):
2020;13(2):497. 053122.
34. Long W, Wu T, Jiao J, Tang M, Xu M. Refraction-learning- 50. Ma J, Man KL, Guan SU, Ting T, Wong PW. Parameter estima-
based whale optimization algorithm for high-dimensional tion of photovoltaic model via parallel particle swarm optimi-
problems and parameter estimation of PV model. Eng Appl zation algorithm. Int J Energy Res. 2016;40(3):343-352.
Artif Intel. 2020;89:103457. 51. Jervase JA, Bourdoucen H, Al-Lawati A. Solar cell parameter
35. Yousri D, Thanikanti SB, Allam D, Ramachandaramurthy VK, extraction using genetic algorithms. Meas Sci Technol. 2001;12
Eteiba M. Fractional chaotic ensemble particle swarm (11):1922-1925.
YOUSRI ET AL. 21

52. Ma J, Ting T, Man KL, Zhang N, Guan SU, Wong PW. Parame-
ter estimation of photovoltaic models via cuckoo search. J Appl How to cite this article: Yousri D, Rezk H,
Math. 2013;2013:1-8. Fathy A. Identifying the parameters of different
53. Hamid NFA, Rahim NA, Selvaraj J. Solar cell parameters iden-
configurations of photovoltaic models based on
tification using hybrid Nelder-Mead and modified particle
recent artificial ecosystem-based optimization
swarm optimization. J Renew Sustain Energy. 2016;8(1):015502.
54. Yousri D, Allam D, Eteiba M. Chaotic whale optimizer variants approach. Int J Energy Res. 2020;1–21. https://doi.
for parameters estimation of the chaotic behavior in permanent org/10.1002/er.5747
magnet synchronous motor. Appl Soft Comput. 2019;74:
479-503.

You might also like