Energy Conversion and Management 248 (2021) 114667

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Photovoltaic mono and bifacial module/string electrical model parameters

identification and validation based on a new differential evolution bee
colony optimizer
Oussama Hachana a, *, Belkacem Aoufi b, Giuseppe Marco Tina c, Mohamed Amine Sid d
a
Department of Drilling and Rig Mechanics, Faculty of Hydrocarbons, Renewable Energies, and Earth, and Universe Sciences, University of Ouargla, 30000, Algeria
b
Department of Electronics and Communications, Faculty of New Information Technologies and Communication, University of Ouargla, 30000 Algeria
c
Department of Electric, Electronic and Computer Engineering (DIEEI), University of Catania, 95125, Italy
d
Mechatronics Laboratory (LMETR) E1764200 Optics and Precision Mechanics Institute, Setif1 University, Setif 19000, Algeria

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

Keyword: Well estimating the electrical model parameters of the photovoltaic (PV) module/string serves to develop an
Hybrid meta-heuristic algorithm accurate simulator and a fault diagnosis tool. Several based evolutionary techniques were proposed to identify
Bifacial PV module the unknown circuit equivalent PV generator (PVG) parameters. Whereas most of them have not been examined
Lambert W function
to various real operating conditions of solar irradiance and PV cells temperature. That requires larger search
Online PV parameters extraction
Partial shading
range than the adopted one in the literature. Enlarging the search range imposes more computational time and
high exploration and exploitation features. Hence, a novel hybrid differential evolution and artificial bee colony
intelligence (nDEBCO) approach is proposed. In terms of convergence quality, CPU execution time, number of
function evaluations (NFE), and error standard deviation (StD). The newly developed approach permits to
accurately identify the PV module/string unknown parameters with suitable implementation complexity. Mono-
facial CLS 220P PV string has been utilized employing an adequate experimental setup with online imple­
mentation. 1080 I-V curves have been measured and estimated, where the overall RMSE ± StD is below 0.02 ±
1e− 16. The nDEBCO outperforms the present-day published works, for common case studies in the literature
with two based root mean square error (RMSE) objective functions namely Lambert W function (LWF) and
classic. It yields 7.73006268e− 4 of RMSE, 7.8785e− 18 of StD, and 2150 NFE under ODM with LWF for RTC
France PV cell. Bifacial PV module has been evaluated and the electrical parameters have been extracted within
less than 1.36 s of CPU run time and not>8.0299631e− 3 ± 6.9096e− 16 of RMSE ± StD for front and rear faces.
Additionally, the parameter identification procedure has been well validated to simulate the real partial shading
scenarios of the studied PV string with a RMSE less than 0.045 and 0.397% of power maximum point absolute
error.

different operating conditions, it is necessary to select a significant

synoptic model and a robust method to identify the unknown electrical
1. Introduction
parameters. Several models in the literature express the physical
behavior of the PV modules to emulate the estimated I-V curves [3]. The
Renewable electricity is the world’s fastest-growing source of
most common and used of which is the one diode-based circuit model
commercialized energy. It is expected that by 2050, photovoltaic and
(ODM). It is verified that this model provides an accurate prediction in
wind power generation will be the main part of several developed
outdoor measurements of the annual energy yield, especially for mono-
countries’ energy resources [1]. The task of replacing fossil fuels can be
c-Si and multi-c-Si technologies [4]. The double diode model (DDM) and
met by solar energy technology, not only because of their prices
the triple diode model (TDM) are still prevalent, by taking into account
continue to decrease while the cost of fossil fuels continues to increase.
the recombination effects of the charge carriers, however they require
But also because of the advantage of limitless resources, lightweight, no
more computational time [5].
emissions, preservation, and long-term maintenance required [2].
Due to the transcendental aspect of the PV module’s characteristic
To examine and evaluate the PV generator (PVG) behavior under

* Corresponding author.
E-mail address: [email protected] (O. Hachana).

https://doi.org/10.1016/j.enconman.2021.114667
Received 2 June 2021; Accepted 17 August 2021
Available online 13 October 2021
0196-8904/© 2021 Elsevier Ltd. All rights reserved.
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Nomenclature NP number of population

StD standard deviation
Variables NFE number of function evaluations
Iph photocurrent, A BBO biogeography-based optimization
Isat, Isat1, Isat2, Isat3 diode’s currents, A nDEBCO new differential evolution bee colony optimization
Rs series resistance, Ω ABC-DE artificial bee colony with differential evolution
Rsh shunt resistance, Ω MPSO modified particle swarm optimization
n, n1, n2, n3 diode’s ideality factors PV solar photovoltaic
q electron charge PSO particle swarm optimization
k Boltzmann constant BFA bacterial Foraging Algorithm
Ns number of solar cells in series GWO grey wolf optimizer
Np number of solar cells in parallel TLBO teaching-leaning based optimization
Im output current, A ICA imperialist competitive algorithm
Vm output voltage, V CSO cat swarm optimization
D problem dimension MVO multi-verse optimizer
Vth thermal voltage, V ELBA enhanced lévy flight bat optimization algorithm
T PV cells temperature, Kelvin SCA complex evolution algorithm
G solar irradiance, W/m2 ACO ant colony optimization (ACO)
BRO battle royal optimization (BRO)
Abbreviations MuFy mutated firefly
ODM one diode model OF objective function
DDM double diode model MADE memetic adaptive differential evolution
TDM three diode model CPU central processing unit
RMSE root mean square error DPDE permutation differential evolution algorithm
IAE individual absolute error EABOA enhanced adaptive butterfly optimization algorithm
AE absolute error
LWF Lambert W function

Table 1
PV cell/module/string case study conditions in this work.
PV generator Solar irradiance PV cell temperature Number of
[W/m2] [Kelvin] series PV cell

RTC France silicone 1000 306.15 1

PV cell
Photowatt PWP-201 1000 318.15 36
PV module
STP6-120/36 PV 1000 328.15 36
module
CLS 220P PV 701 319.82 60
module
Fig. 2. Circuit based double diode model.
CLS 220P PV string 829 325.69 180
Rear CEA INES PV 198 296.65 72
module
Front CEA INES PV 200 296.65 72
module

Fig. 1. Circuit based one diode model. Fig. 3. Circuit based one diode model of bifacial PV module [57].

function, different strategies have been presented to estimate the PVG workable solution [7]. Numerical techniques are based on iterative al­
parameters. The calculated I-V characteristics and data generated by the gorithms such as Levenberg-Marquardt, Newton-Raphson, recursive or
manufacturer concerning Standard Test Conditions (STC) are usually least-squares methods. That to approximate the resistance values and
used by analytical techniques [6]. It is based on a pure mathematical the ideality factor where usually the currents (the photo-generated and
resolution to determine the mean target parameters. Although it heavily the diode saturation currents) are analytically determined [8]. However,
depends on the measured values at the maximum power points; if steps these techniques rely entirely on the initial parameter values that could
are not accurate, the theoretical model applied cannot converge to a give an insignificant solution. Nonetheless, nowadays these techniques

2
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Fig. 4. Flowchart of the nDEBCO algorithm.

Table 2
Search range boundaries of the PV model unknown parameters.
Parameter RTC France PV cell PV module/string Bifacial PV module

Lower Higher Lower Higher Lower Higher

limit limit limit limit limit limit

Iph (A) 0 1 0 10 0 10
Isat1 & Isat2 0 1e− 4 0 1e− 4 0 1e− 4
(A)
Rs (Ω) 0 2 0 5 0 5
Rsh (Ω) 0 100 0 5000 0 5000
n1 & n2 0 2 0 2 0 2

could generate feasible results by applying the Lambert W-function

(LWF) based RMSE objective function [9].
Evolutionary computation and nature-inspired optimization tech­
niques have arisen due to their competence to deal with multi-model
Fig. 5. NFE evolution of PWP-201 PV module by using LWF with ODM.
and nonlinear problems. Also, they cope well with differentiability
and convexity. Several algorithms for the PV parameter identification (WOAs), Shuffled Complex evolution Algorithms (SCAs), Backtracking
problem have been proposed so far. As it is stated by [10], the parameter Search Algorithms (BSAs), (Grey Wolf Optimizers) GWOs, JAYAs, and
identification based on meta-heuristic techniques could be classified Hybrid algorithms.
according to the principal algorithm’s strategy. They are categorized GAs have acceptable accuracy but they are easy to fall into local
into several families namely: Genetic Algorithms (GAs), Differential optimum and their parameters highly affects the solution quality. DEs
Evolutions (DEs), Particle Swarm Optimizations (PSOs), Teaching- are easy to code with fast convergence and good accuracy under various
Leaning Based Optimizers (TLBOs), Whale Optimization Algorithms weather conditions. The variants give better results in terms of CPU run

3
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

algorithm (SSA) [30], Farmland Fertility Optimizer (FFO) [31], Sto­

chastic Fractal Search (SFS) [32].
Hybrid and improved algorithms have complex structure though
they have better performance and the CPU run time is greatly reduced
[10]. These techniques could achieve accurate findings compared to the
original versions due to the exploitation of the best features of both
hybridized techniques [33], such as hybrid Artificial Bee Colony Dif­
ferential Evolution (ABC-DE) [3], hybrid Grey Wolf Optimizer and
Cuckoo Search (GWOCS) [21], Performance-Guided JAYA (PGJAYA)
[17], Improved Teaching-Learning-Based Optimization (ITLBO) [12],
hybrid Bee Pollinator Flower Pollination Algorithm (BPFPA) [34],
Memetic Adaptive Differential Evolution (MADE) [35], Guaranteed
Convergence Particle Swarm Optimization (GCPSO) [36], Multiswarm
Spiral Leader Particle Swarm Optimization (M− SLPSO) [37], Improved
Opposition-Based Whale Optimization Algorithm (OBWOA) [13],
Directional Permutation Differential Evolution algorithm (DPDE) [38],
Enhanced Adaptive Butterfly Optimization Algorithm (EABOA) [39],
Fig. 6. Convergence curves of several types of objective functions. Teaching-Learning based Artificial Bee Colony (TLABC) [40], Back­
tracking Search Algorithm with Reusing Differential Vectors (BSARDVs)
time. However, their parameters need to be well set and they have lack
[15], Hybrid Firefly And Pattern Search (HFAPS) [41], Multiple
of exploitation ability. PSOs have good balance between the exploration
Learning Backtracking Search Algorithm (MLBSA) [42], Improved Ant
and exploitation phases, easy to code and they have good memory
Lion Optimizer (IALO) [43], and Improved JAYA (IJAYA) [44], Differ­
considering the best position of each particle and the swarm’s best po­
ential Evolution with Reinforcement Learning (RLDE) [45], Modified
sition. Meanwhile, they are easy to fall into local optimum with slow
Rao-1 [46], Interval Branch and Bound Global Optimization Algorithm
CPU run time. Indeed, the parameter setting is of crucial importance on
(IBBGOA) [47], Nonhomogeneous Cuckoo Search (NoCuSa) [48], Niche
the exploration phase, these parameters could differ from problem to
Particle Swarm Optimization in Parallel Computing (NPSOPC) [49],
another [11]. TLBOs are efficient and easy to code but they are slow in
Similarity-Guided evolutionary multi-task with the Differential
terms of computational time [12]. WOAs have simple structure and good
search ability; meanwhile they present premature and slow convergence
[13]. SCAs present interesting accuracy in terms of mean standard de­
viation, however they have complicated structure and the performance
is highly affected by the parameter setting [14]. BSAs have simple
structure and efficient results, but they are slow and the performance
depends on the random selection of historical population [15]. GWOs
have interesting accuracy under various weather conditions though the
convergence quality slows down in the later period [16]. JAYAs are
simple and effective; meanwhile they have poor quality of final solution
with large required NFE [17].
To name some meta-heuristic techniques in their original version
used to solve the PVG parameter extraction problem, they are PSO [18],
Firefly Algorithm [19], Bacterial Foraging Algorithm (BFA) [20], GWO
[21], JAYA [17], TLBO [22], Imperialist Competitive Algorithm (ICA)
[23], Cat Swarm Optimization (CSO) [24], Multi-Verse Optimizer
(MVO) [25], Moth Flame Optimization (MFO) [26], sunflower optimi­
zation (SFO) algorithm [27], Enhanced Lévy flight Bat Optimization Fig. 7. Measured and simulated I-V and P-V curves by one diode model
algorithm (ELBA) [28], Slime Mould Algorithm (SMA) [29], Salp Swarm with nDEBCO.

Table 3
Statistical analysis by nDEBCO versus the recent published algorithms in case of RTC France PV cell with ODM.
Technique RMSE SAE MAE SSE

Best (e− 4) NFE CPU time Best (e− 2) NFE CPU time Best (e− 4) NFE CPU time Best (e− 5) NFE CPU Time

nDEBCO 7.73006268 2150 0.755 s 1.65964131 8350 2.722 s 6.383235814 7300 2.379 s 1.55360059 2530 0.824 s
EPSO 8.0621 – 13.67 s 1.76324 – – 7.2259 – – 1.6900 – –
C-HCLPSO 7.73006 – – 1.01671 – – 6.78171 – – 1.55360 – –
FC-EPSO1 7.7301 – 11.51 s 1.7633 – – 6.7820 – – 1.5536 – –
FC-EPSO2 7.7339 – 11.36 s 1.7647 – – 6.7873 – – 1.5552 – –
FC-EPSO3 7.7309 – 11.26 s 1.7626 – – 6.7791 – – 1.5540 – –

Table 4
The extracted parameters of RTC France PV cell for four objective function types.
Parameters RMSE SAE MAE SSE

Iph (A) 0.76078796660164 0.76084152911751 0.76084152911751 0.76078796651084

Isat (µA) 0.31068463313903 0.32434418532084 0.32434418532102 0.31068458747221
Rs (Ω) 0.03654694478152 0.03641898622189 0.03641898622188 0.03654694546319
Rsh (Ω) 52.8897913461548 56.3228145405500 56.3228145405513 52.8897886124127
N 1.47726490925928 1.48134765885540 1.48134765885546 1.47726489456173
OF 7.73006268994108e− 4 1.65964131187409e− 2 6.38323581490068e− 4 1.55360059895107e− 5

4
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 5
Comparison results for RTC France PV cell parameters by using ODM with classic and LWF RMSE.
Classic RMSE

Algorithm/Author Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 4)

NPSOPC 0.7608 0.3325 0.03639 53.7583 1.4814 9.8856

GWOCS 0.76077 0.32192 0.03639 53.6320 1.4808 9.8607
CWOA 0.76077 0.3239 0.03636 53.7987 1.4812 9.8602
IJAYA 0.7608 0.3228 0.0364 53.7595 1.4811 9.8603
RLDE 0.7608 0.3231 0.0364 53.7185 1.4812 9.8602
MADE 0.76077553 0.323 0.03637709 53.7185236 1.48118359 9.860219
MLBSA 0.76077553 0.323 0.0363771 53.7185461 1.4811835 9.860219
CPMPSO 0.76077553 0.323 0.03637712 53.7183835 1.48118309 9.860219
SGDE 0.76078 0.32302 0.036377 53.71853 1.481184 9.8602187789
MRao-1 0.760776 0.323021 0.036377 53.718522 1.481135 9.860219
nDEBCO 0.760775530434227 0.323020774222916 0.0363770931030067 53.7185206085772 1.48118068229836 9.86021877891545

LWF RMSE
Laudani et al. 0.76078840 0.3102482 0.03655304 52.859056 1.4769641 7.7301
Cárdenas et al. 0.760788 0.3106847 0.036547 52.890468 1.4771051 7.7301
Toledo et al. 0.76078797 0.31068485 0.036546942 52.889804 1.4771052 7.730062726
Panchal et al. 0.7607880 0.3106884 0.0365469 52.8900296 1.4771063 7.730062726
HISA 0.760787966 0.310684592 0.036546945 52.889794264 1.477267785 7.73006
COA 0.76076929153 0.3083945801266 0.03655460766 52.82666150326 1.47654776591 7.75470161606
FC-EPSO 0.76079 0.31131 0.03654 52.944 1.4773 7.7301
LCROA 0.76079 0.31068 0.03655 52.88979 1.51690 7.730063
ImSMA 0.76079 0.31068 0.03655 52.89000 1.51770 7.73006
SMA 0.76079 0.31114 0.03654 52.95400 1.51780 7.73011
FFO 0.7608 0.3223 0.0364 53.8989 1.5215 7.75184
SFS 0.7609 0.3167 0.03648 53.2805 1.47918 7.931
nDEBCO 0.760787966601641 0.310684633139038 0.0365469447815297 52.8897913461548 1.47726490925928 7.73006268994108

Evolution (SGDE) [50]. Each one of the abovementioned techniques can getting more precise results. The adopted strategies are sealed in such a
reach an RMSE less than 2e− 2 for the well-known STP6-120/36 PV way to exploit the important features of each one within a competitive
module parameter identification problem. execution time. This is done, by well adjusting the algorithm setting
Nevertheless they differ heavily in reaching the target values when parameters and testing several designed algorithm flowcharts.
applied in several runs which could be indicated by the standard devi­ To validate the proposed algorithm, by using the one, two or even
ation if the mean RMSE value is very close to the best one. Indeed, three diode models, we have compared several meta-heuristic tech­
introducing the LWF requires more computational time and several niques; DE [51], ELBA [28], Modified Particle Swarm Optimization
number of function evaluations (NFE) especially for the double diode (MPSO) [3], Ant Colony Optimization (ACO) [52], Battle Royale Opti­
model, and some of these techniques are not able to reach better results. mization (BRO) [53], ABC-DE [3], Mutated Firefly (MuFy) [54], SCA
The authors of [11] have proposed FC-EPSO to enhance the accurately of [14] with the proposed nDEBCO.
the standard PSO, where ODM, DDM, and TDM have been used. The StD The abovementioned extraction techniques are validated with the
have been enhanced from 4.3109e− 04 to 3.3351e− 10, and the CPU run well-known measured I-V characteristics of the RTC France silicon PV
time is minimized by 2.4 s for RTC France PV cell under ODM. However, cell, the Photowatt PWP-201 PV module [55], and the STP6-120/36 PV
the convergence speed still very slow even if the adopted search range is module with 24 data points [14]. Then, they have been tested with
relatively narrow, and the mean standard deviation could be further measured I-V characteristics of the CLS 220P PV module and the CLS
improved compared to the present-day techniques. An improved SCA 220P PV string.
algorithm [14] has been presented for parameter extraction. This algo­ Bifacial floating PV modules technology is noticed to be the future
rithm is accurate but the CPU run time is slower because of the complex PV power plants as they augment the PVG performance by 25–30% or
structure. Trapping into the local minima is among the limitations of more [56]. Bifacial floating PV module I-V curves are exploited in a way
GWOCS [21]. ITLBO [12] accuracy is poor mainly in DDM for RTC to predict its electrical behavior; the extraction technique is used to find
France PV cell case study, and its CPU run time is significantly low. the unknown parameters of the based equivalent circuit model of the
TLABC [40] is not able to reach the best parameters for RTC France PV front and the back faces. The I-V curves of the bifacial floating class A
cell by using DDM. solar simulator by CEA INES are used. The aforementioned techniques
To get acceptable performances, it is required to develop more so­ were chosen to assess the nDEBCO validity as the bifacial PV module I-V
phisticated optimization technique of which hybrid meta-heuristic curves are more complicated than the I-V curves of the mono-facial
approach worth to enhance reliability, convergence speed, and accuracy technology under normal operating conditions. While, the solar irradi­
with well parameter setting. ance, the cells temperature and the number of series cells are specified in
In this paper, a novel hybrid meta-heuristic technique is proposed. It Table 1.
is based on strategies from three techniques namely: mutation strategy We assume that the cells which make up the PV module work under
of DE technique, probability evaluation and selection strategy of the the same environmental conditions. They are used for an application
artificial bee colony optimization (ABCO), and elitism principle of with separate experimental data from the CLS 220P PV string taken
biogeography-based optimization (BBO). The DE mutation strategy al­ successively for several days under different cells temperatures and ir­
lows improving information exchange with an interesting execution radiances measurements.
time which permits to enlarge the exploration features. ABCO proba­ After that, given the RMSE error, CPU execution time, and error
bility evaluation is a suitable strategy to ensure a good exploitation standard deviation, the best technique is taken as the adopted extraction
procedure [3]. In addition, its selection strategy is an interesting technique for online implementation using MATLAB code and LabVIEW
candidate to ameliorate the exploitation phase. To strengthen the al­ environment techniques to connect with the setup tools.
gorithm accuracy, translated by error standard deviation over many The unknown parameters interval should be large as possible and by
independent runs, the elitism concept of BBO is chosen which allows consequence; the chosen extraction technique could be able to deal with

5
 Table 6
Comparison results for RTC France PV cell parameters by using DDM with classic and LWF RMSE.
O. Hachana et al.

Classic RMSE

Algorithm Iph (A) Isat1 (µA) Isat2 (µA) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 4)

IJAYA 0.7601 0.0050 0.7509 0.0376 77.8519 1.2186 1.6247 9.8293

MADE 0.7608 0.2260 0.7493 0.0367 55.4854 1.4510 2.0000 9.8248
RLDE 0.7608 0.2260 0.7492 0.0367 55.4847 2.0000 1.4510 9.8248
MOWOA 0.760813 0.192684 0.999587 0.036861 55.933515 1.438003 1.983721 9.8387
SGDE 0.7607 9 0.2807 0.2500 0.0365 54.3667 1.4697 1.9323 9.8441
GWOCS 0.76076 0.5377 0.2486 0.0367 54.7331 2.0000 1.4588 9.8334
nDEBCO 0.760781079116673 0.749359716165149 0.225974180450110 0.0367404324014885 55.4854590511075 1.45101386446906 1.99999999999548 9.82484806504419

LWF RMSE

Algorithm Iph (A) Isat1 (µA) Isat2 (µA) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 4)

nDEBCO 0.761045683830275 0.031474968250 4.17893816629867 0.054840273330078 61.63910706195 1.316628885044 2.160089785420 6.33514460929039

ABC-DE 0.761042208515429 0.030713411880 4.2996632949734 0.054549220125503 61.73482923285 1.317854223093 2.168515387255 6.33538572354928
SCA 0.761013629176119 0.036271333541 4.72959065289964 0.053144178461470 61.96314429663 1.325715795932 2.201963785277 6.34287151984161
ELBA 0.761061424482193 0.031507713805 3.97207516377626 0.054826669757691 60.91954037302 1.318515316912 2.150472357918 6.33761442787326
MuFy 0.760787966734130 0.310684578616 0 0.036546945602764 52.88978758242 1.477264891672 3.970938898277 7.73006268994139
DE 0.760787966677688 0.310684609013 0 0.036546945144197 52.88978878968 1.746813973271 1.477264901539 7.730062689941377

6
Table 7
Comparison results for RTC France PV cell parameters by using TDM.
Classic RMSE Algorithm nDEBCO ABC-DE SCA ELBA MuFy DE

Iph (A) 0.760781079259053 0.760781079331713 0.760781078770616 0.760781081050807 0.760775529999677 0.760780339190694

Isat1 (A) 7.41287387178974e− 07 2.25973253521854e− 07 2.19756175054848e− 07 1.83338704962265e− 07 3.23020804114048e− 07 2.26154865905291e− 07
Isat2 (A) 8.07256220206361e− 09 7.49367870990994e− 07 6.21812419874780e− 09 2.25974602951125e− 07 0 7.48136129752433e− 07
Isat3 (A) 2.25974130894646e− 07 1.32634345577720e− 18 7.49358706852886e− 07 5.66010766586337e− 07 0 0
Rs (Ω) 0.0367404328741872 0.0367404362841885 0.0367404318143466 0.0367404408756642 0.0363770927761787 0.0367388793402091
Rsh (Ω) 55.4854556221509 55.4854789347404 55.4854579788953 55.4853829011982 53.7185245344415 55.4884300184036
n1 1.45101384492652 1.45101352366312 1.45101386819996 1.45101397469357 1.60210419429916 1.45108282621806
n2 1.99999999998498 1.46602710059845 1.45101533145303 1.99999999883499 1.48118069164705 2
n3 1.99999999999964 1.89408265674928 1.99999999999642 1.99999999831625 1.52012832506688 1.03814942168618
RMSE (e− 4) 9.82484806504297 9.82484806525273 9.82484806504424 9.82484806531513 9.86021877891571 9.82485038632230

LWF RMSE Algorithm nDEBCO ABC-DE SCA ELBA MuFy DE

Iph (A) 0.761146983732072 0.761148527766600 0.761148528304812 0.760804633505066 0.760787966568495 0.760909561233695
Isat1 (A) 1.46921473128381e− 8 1.44989792414279e− 8 1.63838742263814e− 8 3.13693478373768e− 32 0 1.8791833653627e− 7
Isat2 (A) 2.08105542196259e− 5 2.07622762715348e− 5 2.07622762715348e− 5 2.79342187969988e− 7 3.10684606823635e− 7 0.0001
Isat3 (A) 5.16441593001239e− 8 5.57821451712741e− 8 6.51940502813171e− 8 2.72072746277753e− e− 10 0 0
Rs (Ω) 0.0601753254311937 0.0603225404833754 0.06040293747234 0.0373268386485767 0.0365469452381889 0.0384721564177176
Rsh (Ω) 70.6750170761099 70.6690696213400 70.68303847263856 51.8256289295450 52.8897917826888 69.8336675147512
n1 1.26390870252020 1.26325821227559 1.263437469340053 0.351293241235929 1.99995270634807 1.43300509139303
n2 1.70601988767403 1.71292354386802 1.718393746284950 1.46688941069400 1.47726490070849 4.20452417167569
n3 2.73239203588216 2.73350276686879 2.733903094726459 1.86936620950309 1.47688857070316 1.83377270397397
RMSE (e− 4) 5.72894461132072 5.728945790713155 5.728945908827361 7.51489028419689 7.73006268994100 6.92266595146841
Energy Conversion and Management 248 (2021) 114667
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 8
Statistical results of the RTC France PV cell.
Model type RMSE Algorithm Best RMSE (e− 4) Mean RMSE (e− 4) Worst RMSE (e− 4) CPU Time (s) StD NFE

Using ODM Classic nDEBCO 9.86021877891545 9.86021877891556 9.86021877891567 0.134 1.8824e− 17 2050
ABC-DE 9.86021877891548 9.86021878134452 9.86021878683947 0.627 3.2994e− 14 3450
SCA 9.86021877891545 9.86021877891592 9.86021877891598 1.028 2.0558e− 17 3600
ELBA 9.86021877891550 9.86021877891667 9.86021877891682 1.107 5.0917e− 17 4800
DE 9.86021877891568 9.86021877891599 9.86021877891697 2.437 6.5873e− 17 19,000
MuFy 9.86021877891558 9.86021877891623 9.86021877891671 1.491 5.3370e− 17 13,100
MPSO 9.86021877891589 9.96388746001928 12.2390210948347 2.433 9.6938e− 4 39,000
MTLBO 9.860219 9.860219 – 1.62 1.9092e− 17 5000
Lambert W

Function LCROA – 141.491079 – 4.397 2.1851e− 2 –

nDEBCO 7.73006268994108 7.73006268994109 7.73006268994111 0.755 7.8785e− 18 2150
ABC-DE 7.73006268994112 7.73006268994176 7.73006268994239 1.530 6.6242e− 17 6500
SCA 7.73006268994120 7.73006268994125 7.73006268994165 1.510 9.8839e− 18 5800
ELBA 7.73006268994118 7.73006268994164 7.73006268994196 1.895 2.4543e− 17 6900
DE 7.73006268994132 7.73006268994152 7.73006268994245 5.019 1.3230e− 17 17,300
MuFy 7.73006268994121 7.73006268994137 7.73006268994202 4.318 0.2390e− 17 13,000
FC-EPSO3 7.7309 7.7309 – 11.26 3.3351e− 10 –

Using DDM Classic nDEBCO 9.82484806504419 9.824854355990927 9.82488298394872 0.789 2.4087e− 9 16,950
ABC-DE 9.82484811534282 9.827212147492618 9.82932989180029 1.549 1.2543e− 6 31,800
SCA 9.82484806608714 9.830315092549432 9.84292230918473 2.781 1.1989e− 6 32,500
ELBA 9.82484807329834 9.852394850104589 9.87883720203781 2.932 4.9762e− 6 40,200
DE 9.82484812392873 9.839359023795630 9.85203094883748 5.281 2.7234e− 6 > 50,000
MuFy 9.82484812230948 9.835234489130954 9.88929382039339 5.172 5.2457e− 6 > 50,000
MTLBO 9.8248 9,825,026 – – 3.3000e− 9 –
LWF nDEBCO 6.33514460929039 6.33514739847374 6.33514928377463 30.130 4.2093e− 9 64,550
ABC-DE 6.33540392874382 6.33613039847395 6.33692309283743 > 45 1.2543e− 5 > 100,000
SCA 6.33544460929039 6.33573984736541 6.33620938747270 > 45 2.9837e− 6 > 100,000
ELBA 6.33545488372354 6.33623930019238 6.34345938488311 > 55 7.2983e− 5 > 100,000
DE 7.73006268994139 7.73006329938723 7.73006438847329 > 55 3.3098e− 9 > 100,000
MuFy 7.73006268994137 7.73006312983742 7.73006390307424 > 55 2.0293e− 9 > 100,000

Using TDM Classic nDEBCO 9.82484806504297 9.82602710835253 9.86021878344563 2.287 6.3492e− 07 36,900
ABC-DE 9.82484806525273 9.83453366269928 9.86110818714423 3.440 1.5338e− 06 55,500
SCA 9.82484806504424 9.82838631396033 9.86021878186181 3.842 1.0611e− 06 56,300
ELBA 9.82484806531513 9.87001012209412 10.1344367823047 4.307 1.5943e− 06 94,360
MuFy 9.86021877891571 1240.04970195604 5439.17378675707 > 10 0.1808 >100000
DE 9.82485038632230 9.84771029131780 9.86021877891615 8.457 1.5772e− 06 >100000
LWF nDEBCO 5.72894461132065 5.72894461132928 5.72894461134239 622 4.3982e− 16 466,000
ABC-DE 5.72894579071338 5.72913092847539 5.72980397642345 > 900 3.0283e− 9 >1000000
SCA 5.72894590298223 5.72900299381345 5.72950297738083 > 900 1.3872e− 9 >1000000
ELBA 7.51489028419645 7.79938742540867 7.93783702745102 > 900 4.2492e− 4 >1000000
MuFy 7.73006268994156 7.93982738119325 8.74998374903866 > 900 2.6029e− 3 >1000000
DE 6.92266595146854 7.23883729201094 7.90308477362354 > 900 9.3894e− 4 >1000000

that relatively larger search space within an interesting CPU run time then described respectively by Eqs. (1) and (2). They are characterized
and error standard deviation allowing it to be exploitable for fault by some unknown parameters. Iph (photo-generated current) illustrates
diagnosis purposes [5]. Aside from developing an enhanced meta-heu­ the charge carrier generation in the semiconductor layer of the PV cell
ristic algorithm, the contribution of this paper materializes in the pos­ caused by the incident radiation, Rsh (shunt resistance) expresses the
sibility of finding an accurate RMSE value, interesting CPU execution losses due to the high-current path through the semiconductor
time, and feasible error standard deviation by using either LWF or classic throughout the mechanical defects and the leakage current to the
RMSE as objective functions with that larger search range. This is suit­ ground, Rs (series resistance) shows the internal losses due to current
able for online application, what indicates taking into consideration all flow and the connection between cells.
possible variation of solar irradiance and cells temperature during the Since the presence of the diode, there are two unknown parameters,
whole time of measurement consecutively. Isat (the reverse-bias saturation current) and n (ideality factor). In the
The paper is organized as follows: the second section presents the PV case of DDM, each diode has two unknown parameters; Isat1 and Isat2
module based on the equivalent circuit and the relationship between it (reverse-bias saturation current for the first diode and the second one,
and the PV string model for mono and bifacial technologies. The third consecutively) and n1 and n2 (the ideality factors of each diode). It is
section describes the newly proposed extraction technique, its optimi­ possible to generalize this aspect to three diode model by adding two
zation strategies, and the problem formulation. Fourthly, using real unknown parameters in comparison to DDM owing to the third diode
measured data, the experimental results are displayed, validated, and (Isat3 and n3).
discussed for the mono and bifacial PV modules/string. The sixth While, Vthm (thermal module voltage = Ns ∙k∙T/q), q (electron
paragraph concludes this work. charge = 1.602e-19C), k (Boltzman’s constant = 1.381e− 23 J/K) and Ns
(solar cell number in series) represent the known parameters, where T
2. PV generator modeling (cells temperature (Kelvin)) is measured variable for each I-V curve.
According to the voltage and the current output values, this circuit can
2.1. Mono-facial PV model be redefined for a PV string formed by parallel and/or series association
of divers PV modules by using Eq. (3). Where, Im and Vm are the
The equivalent one diode circuit and double diode models of PVG are measured current and voltage respectively. This is described by using
the mean models used in the literature, depicted in Figs. 1 and 2, and ODM, while the parameters are related to the constitutive modules

7
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Fig. 8. NFE evolution of RTC France PV cell by using classic RMSE for : (a) ODM, (b) DDM, (c) TDM, and by using LWF RMSE for : (d) ODM, (e) DDM, and (f) TDM.

parameters by using Eq. (4). Knowing that Np is the number of PV ( )

Vb − Ib ⋅Rsb
modules in parallel and Nsm is the number of series PV modules or sub-
Vb − Ib ⋅Rsb
Ib = Iphb − Isatb ⋅ e nb ⋅Vthm − 1 − (6)
strings. Rshb
( )
Vm − Im ⋅Rs where the unknown parameters of the front side to be identified are Iphf,
(1)
Vm − Im ⋅Rs
Im = Iph − Isat ⋅ e n⋅Vthm − 1 −
Rsh Isatf, Rsf, Rshf, and nf related to photo-generated current, saturation diode
current, series resistance, shunt resistance, and ideality factor consecu­
( ) ( )
Vm − Im ⋅Rs tively. Also, Iphb,
(2)
Vm − Im ⋅Rs Vm − Im ⋅Rs
Im = Iph − Isat1 ⋅ e n1 ⋅Vthm − 1 − Isat2 ⋅ e n2 ⋅Vthm − 1 −
Rsh Isatb, Rsb, Rshb, and nb related to photo-generated current, saturation
( ) diode current, series resistance, shunt resistance, and ideality factor of
Vms − Ims ⋅Rss the backside respectively.
(3)
Vms − Ims ⋅Rss
Ims = Iphs − Isats ⋅ e ns ⋅Vthm − 1 −
Rshs
⎧ 3. Identification technique and problem formulation
⎪
⎪
⎪ Iphs = Np ⋅Iph
⎪
⎪
⎪
⎪ The proposed algorithm is inspired from DE, ABCO, and BBO algo­
⎪
⎪
⎪
⎪ Isats = Np ⋅Isat rithms simulated to the bees’ behavior and their search strategy to reach
⎪
⎨ / the best honey source. Exchange among bees is done by the search
Rss = Nsm
Np s
⋅R (4)
⎪
⎪ strategy while each source is considered as possible solution (position).
⎪ /
⎪
⎪
⎪ Rshs = Nsm N ⋅Rsh After the first navigation, the bees share their reports according to the
⎪
⎪ p
⎪
⎪
⎪ quality of the discovered honey sources. They are sorted into three main
⎪ ns = Nsm ⋅n
⎩ troops, namely leaders, onlookers and scouts. The leaders are in the first
part position of the swarm; they lead the onlookers to exploit the best
Hence, the five unknown parameters to be identified are Iphs, Isats, Rss, honey source found. At the same time, the scout bees travel the rest of
Rshs, and ns related to the photo-generated current, the saturation diode the search field and try to find other profitable honey sources. After the
current, the series resistance, the shunt resistance, and the ideality factor swarm’s initialization, all the bees commence discovering the search
of the PV string consecutively. Where, Ims and Vms are the string space by using the search strategy.
measured current and voltage respectively. However for the ideality
factor ns, it is possible to consider it equal to n, that is limited in the range 3.1. Search strategy
of [0, 2], and Nsm will be integrated in Vthm = Nsm ∙Ns ∙k∙T/q.
Inspired from the mutation strategy of DE algorithm, Mi, expresses
2.2. Bifacial PV model the search strategy of each bee based on the reports of the other bees as it
is defined by Eq. (7).
Fig. 3 presents the based equivalent circuit of the bifacial PV module Mi = Bbest + Q(Br1 − Br2 ) + Q(Br3 − Br4 ) (7)
[57]. It is based on the one diode model and the both faces are connected
in parallel. While, the equations defining the current front and back while, r1, r2, r3, and r4 are different bee indices other than i (the
sides are given by Eq. (5) and Eq. (6), respectively. current bee’s index). Br1 , Br2 , Br3 , and Br4 are respectively their corre­
( Vf − If ⋅Rsf ) sponding honey sources (positions). Q is a random number between − 1
If = Iphf − Isatf ⋅ e nf ⋅Vthm − 1 −
Vf − If ⋅Rsf
(5) and 1. Bbest is the bee with the best source already found by the swarm.
Rshf

8
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 9
Statistical results of several PV modules.
Module/model types RMSE Algorithm Best RMSE (e− 2) Mean RMSE (e− 2) Worst RMSE (e− 2) CPU Time StD NFE
(s)

PWP-201 by using ODM Classic nDEBCO 0.242507486809511 0.242507486809518 0.242507486809522 0.274 1.0833e− 17 5300
ABC-DE 0.242507486809516 0.242507486809535 0.242507486809591 0.750 2.8764e− 16 9200
SCA 0.242507486809514 0.242507486809531 0.242507486809536 1.443 1.3763e− 17 12,750
ELBA 0.242507486809515 0.242507486809539 0.242507486809583 1.087 4.6503e− 16 13,000
DE 0.242507486809562 0.242507486809582 0.242507486809631 2.033 6.5029e− 16 27,000
MuFy 0.242507486809517 0.242507486809568 0.242507486809588 1.802 1.4528e− 17 21,700
MPSO 0.242634399384209 0.242894235980323 0.242932766253621 3.228 4.2358e− 4 42,800
MTLBO 0.2425075 0.2425075 – – 1.3070e− 17 5000
CPMPSO 0.2425075 0.2425075 – – 1.5159e− 17 –
MRao-1 0.2425075 0.2425075 – – 1.7161e− 17 –
ILTBO 0.24251 0.24251 – – 1.2700e− 17 –
Lambert LCROA – 9.42307945 – 4.267 1.5463e− 1 –
W nDEBCO 0.204652439739525 0.204652439739529 0.204652439739538 1.716 1.0348e− 17 5200
Function ABC-DE 0.204652439739532 0.204652439739552 0.204652439739589 2.558 6.7760e− 16 7500
SCA 0.204652439739525 0.204652439739532 0.204652439739551 6.288 1.4013e− 17 12,400
ELBA 0.204652439739531 0.204652439739549 0.204652439739569 3.718 2.6770e− 16 10,200
DE 0.204652439739589 0.204652439739640 0.204652439739679 10.622 1.3456e− 16 26,200
MuFy 0.204652439739578 0.204652439739615 0.204652439739662 12.129 1.0928e− 16 23,400

PWP-201 by using DDM Classic nDEBCO 0.242507486809507 0.242507486809523 0.242507486809582 0.575 2.8012e− 15 10,300
ABC-DE 0.242507486809507 0.242507486815992 0.242507486962939 0.970 2.1727e− 13 16,750
SCA 0.242507486809507 0.242507486809564 0.242507486809598 1.738 4.7244e− 17 14,210
ELBA 0.242507486809546 0.242705836761593 0.248137828397084 1.230 1.0095e− 5 22,400
MuFy 0.242507486809522 0.242507500305549 0.242507889798794 2.221 7.2328e− 10 27,780
DE 0.242507491118383 0.242866880528773 0.248619907045809 4.091 1.3554e− 5 45,200
LWF nDEBCO 1.028263633033673 1.043915598918393 1.091305149723098 45.129 1.6395e− 5 142,000
ABC-DE 1.028263633067232 1.074998734832302 1.239237483293384 51.182 7.2347e− 3 175,000
SCA 1.099498755453134 1.136100307815288 1.211118943425483 61.947 2.1659e− 5 244,000
ELBA 1.037939101438120 1.228094264771434 1.713394603271214 54.027 1.4847e− 4 208,800
MuFy 1.057824317044443 1.401285919830565 2.387952824146039 78.436 4.5240e− 4 275,500
DE 1.35119211305484 1.533867496616 2.052960640839 84.361 2.3727e− 4 282,500

STP6-120/36 by using Classic nDEBCO 1.66006031250859 1.66006031250859 1.66006031250859 0.214 6.4414e− 17 4950
ODM ABC-DE 1.66006031250859 1.66006031250870 1.66006031250892 0.309 3.1931e− 15 6300
SCA 1.66006031250859 1.66006031250859 1.66006031250859 1.329 7.9469e− 17 11,800
ELBA 1.66006031250859 1.66006031250866 1.66006031250878 0.630 4.8727e− 16 10,700
DE 1.66006031250871 1.66006031250891 1.66006031250898 2.132 6.2398e− 15 20,000
MuFy 1.66006031250859 1.66006031250869 1.66006031250890 1.378 7.9821e− 16 16,500
MPSO 1.66006039283745 1.66487273811237 1.66729983726001 2.660 6.3438e− 5 23,400
MTLBO 1.6600603 1.6600603 – – 8.0041e− 17 –
ITLBO 1.6600603 1.6600603 – – 8.6520e− 17 –
LWF LCROA – 36.0323788 – 4.237 5.6326e− 1 –
nDEBCO 1.42510635576861 1.42510635576861 1.42510635576861 1.782 2.3700e− 17 5400
ABC-DE 1.42510635576861 1.42510635576882 1.42510635576898 1.917 8.1784e− 15 5600
SCA 1.42510635576861 1.42510635576863 1.42510635576869 6.104 6.2572e− 17 12,400
ELBA 1.42510635576861 1.42510635576867 1.42510635576871 6.080 3.3820e− 16 15,200
MuFy 1.42510635576862 1.42510635576879 1.42510635576892 10.461 6.8193e− 15 20,700
DE 1.42510635576862 1.42510635576890 1.42510635576902 11.871 9.3349e− 15 24,100

STP6-120/36 by using Classic nDEBCO 1.66006031250850 1.6600603125086 1.6600603125086 0.276 4.8727e− 15 6420
DDM ABC-DE 1.66006031250850 1.6600603384023 1.6600603729382 0.625 3.0832e− 13 14,450
SCA 1.66006031250856 1.6600603125086 1.6600603125086 1.866 1.7724e− 16 15,550
ELBA 1.66006031250858 1.6614293108071 1.6978360213680 1.027 6.7733e− 5 22,800
MuFy 1.66006031250854 1.6609528606409 1.6734483945355 2.113 3.3396e− 5 25,300
DE 1.66006031250904 2.1035646564448 14.964226354352 3.870 0.0239 39,450
LWF nDEBCO 1.04305570761706 1.04305677949432 1.04306942285583 61.650 3.3272e− 8 124,000
ABC-DE 1.04305570772810 1.04320023875422 1.04363984772645 83.293 2.0928e− 5 183,500
SCA 1.04309020913122 1.04427667760338 1.04656653149524 155.167 1.2048e− 5 230,500
ELBA 1.04305570761753 1.10727282670909 1.15494724178863 101.982 2.3051e− 2 188,600
MuFy 1.04319782594629 1.12521436024071 1.42510635576855 137.150 0.0015 216,300
DE 1.06501738711573 1.25102452393798 1.42510635576862 >168 0.0014 >300000

CLS 220P by using ODM Classic nDEBCO 1.13133173011542 1.13133173011545 1.13133173011548 0.315 8.1583e− 17 12,200
ABC-DE 1.13133173011542 1.13133173011553 1.13133173011694 0.657 1.8491e− 15 14,950
SCA 1.13133173011542 1.13133173011548 1.13133173011588 3.357 1.9596e− 16 28,550
ELBA 1.13133173011542 1.13133173011549 1.13133173011625 2.258 3.3489e− 16 39,000
DE 1.13133173018232 1.13133173021913 1.13133173023386 5.103 4.7290e− 15 46,900
MuFy 1.13133173055945 1.13133173063874 1.13133173068930 3.757 7.3992e− 15 43,500
MPSO 1.13133173011613 1.13134330284533 1.13135349039284 3.535 2.3439e− 5 41,000
Lambert nDEBCO 0.64661280315236 0.64661280315236 0.64661280315236 0.675 5.5897e− 17 5500
W ABC-DE 0.64661280315236 0.64661280315240 0.64661280315243 1.231 1.4449e− 15 9700
Function SCA 0.64661280315236 0.64661280315236 0.64661280315236 4.627 7.5235e− 17 17,800
ELBA 0.64661280315238 0.64661280315238 0.64661280315240 2.558 1.0388e− 16 16,400
MuFy 0.64661280315238 0.64661280315282 0.64661280315312 4.257 4.7288e− 15 35,400
DE 0.64661280315238 0.64661280315267 0.64661280315328 5.603 5.3728e− 15 36,200

(continued on next page)

9
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 9 (continued )
Module/model types RMSE Algorithm Best RMSE (e− 2) Mean RMSE (e− 2) Worst RMSE (e− 2) CPU Time StD NFE
(s)

CLS 220P by using DDM Classic nDEBCO 0.433929841502380 0.433929841502987 0.4339298415072 17.321 1.0786e− 13 698,300
ABC-DE 0.456637429010392 0.456637428920381 0.4566360347728 40.294 3.1219e− 9 725,900
SCA 0.530966285331865 0.530966448244763 0.5309421511910 >200 2.9160e− 6 >1000000
ELBA 0.832153987299546 0.823698555880109 1.4570437094113 >200 9.9039e− 3 >1000000
MuFy 1.131331730115324 1.552423631384923 1.7024236313803 >200 7.6698e− 3 >1000000
DE 1.131331730115283 1.505812725678943 1.6367338973088 >200 3.0301e− 3 >1000000
LWF nDEBCO 0.360745466153802 0.41859471044329 0.6466128031756 5.7634 1.4075e− 4 13,000
ABC-DE 0.646612803152360 0.64661280314983 0.6466128031610 9.2261 6.9827e− 13 13,300
SCA 0.646612803152361 0.64661280315244 0.6466128031524 14.543 9.3947e− 15 28,400
ELBA 0.646612803154574 0.67564426805065 0.7029930056878 15.839 9.3947e− 4 36,300
MuFy 0.646612803152405 0.67532190386836 0.9337038103120 18.390 8.6085e− 8 40,200
DE 0.646612803152362 0.64661280315249 0.6466128031524 30.641 7.0298e− 17 54,100

Fig. 9. NFE evolution of : PWP-201 PV module by using classic RMSE for (a) ODM, (b) DDM, and by using LWF RMS for (c) ODM, (d) DDM, and STP6-120/36 PV
module by using classic RMSE for (e) ODM, (f) DDM, and by using LWF RMS for (g) ODM, (h) DDM.

3.2. Evaluation procedure the probability factor of the ith bee, OF(Bi) is the source value quality
found by the ith bee (objective function) and NP is the swarm size.
The evaluation procedure, defined by Eq. (8), is done whenever the
OF(Bi )
search strategy is settled for each bee. It allows the swarm to get quick Pi = ∑NP (9)
adaptive evolution. It should be noticed that the Mi is also compared to i OF(Bi )

the global best bee found yet; it will take its position if Mi is better.
{ 3.4. Crossover strategy
Mi if OF(Mi ) < OF(Bi )
Bi = (8)
Bi Otherwise According to the probability factor, the leader bees share their in­
formation with the worst scout bees to encourage them finding a more
accurate search strategy. This is done by selecting three random di­
3.3. Probability evaluation mensions (according to the search space size) as given by Eq. (10).

Inspired from the ABCO technique [58], this factor permits the
enhancement of the swarm’s exploitation, by using Eq. (9). While Pi is

10
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Fig. 10. NFE evolution of : CLS 220P PV module by using classic RMSE for (a) ODM, (b) DDM, (c) DDM with large search range, and by using LWF RMSE for (d)
ODM, (e) DDM, and CLS 220P PV string by using classic RMSE for (f) ODM, (g) DDM, and by using LWF RMSE for (h) ODM, (i) DDM.

Table 10
Statistical results of the CLS 220P PV string.
PV string model type RMSE Algorithm Best RMSE (e− 2) Mean RMSE (e− 2) Worst RMSE (e− 2) CPU Time (s) StD NFE

CLS 220P by using ODM Classic nDEBCO 1.03473847669449 1.03473847669449 1.03473847669450 0.373 5.6808e− 17 10,800
ABC-DE 1.03473847669451 1.03473847669482 1.03473847669833 0.576 1.1142e− 14 11,400
SCA 1.03473847669457 1.03473847669463 1.03473847669475 2.169 8.7958e− 17 18,500
ELBA 1.03473847669456 1.03473847669467 1.03473847669513 0.656 1.2230e− 16 14,800
DE 1.03473847669454 1.03473847669497 1.03473847669639 4.395 3.9817e− 14 36,900
MuFy 1.03473862358872 1.03473847669502 1.03473847669653 3.496 2.1388e− 14 29,000
MPSO 1.03473847669449 1.03492023492120 1.03513394853022 5.142 4.9203e− 5 43,900
Lambert W nDEBCO 0.887346444842353 0.887346444842356 0.887346444842410 2.092 1.0125e− 17 7000
Function ABC-DE 0.887346444842358 0.887346444842549 0.887346444843733 3.147 3.0251e− 15 9400
SCA 0.887346444842378 0.887346444842433 0.887346444842621 5.039 3.7756e− 17 11,900
ELBA 0.887346444842384 0.887346444842449 0.887346444842740 6.711 8.1992e− 17 15,300
MuFy 0.887346444842386 0.887346444842637 0.887346444846899 11.783 8.9037e− 15 27,800
DE 0.887346444842392 0.887346444842595 0.887346444845783 12.648 4.7883e− 15 32,300

CLS 220P by using DDM Classic nDEBCO 1.034738476694482 1.034738476695039 1.034738476698808 0.534 1.0388e− 14 12,150
ABC-DE 1.034738476694513 1.034738476696198 1.034738476712094 0.643 4.9721e− 12 13,400
SCA 1.034738476694579 1.034738476694643 1.034738476694682 2.069 1.9500e− 16 19,900
ELBA 1.034738476694560 1.240635297364634 6.592834713489783 1.449 0.0100 30,250
MuFy 1.034738476694543 1.059075597728809 1.762542063597998 3.236 0.0013 31,100
DE 1.034738623588725 1.082621531402162 2.433867425887874 5.444 0.0025 50,000
LWF nDEBCO 0.887156674403631 0.887320938714529 0.888446444842443 4.868 3.9122e− 6 22,300
ABC-DE 0.887156674403631 0.938397362741432 1.089372647362142 6.239 9.4536e− 3 25,450
SCA 0.887156674403631 0.887213605535272 0.887346444842493 7.484 8.6964e− 7 38,000
ELBA 0.887156674403631 0.904528244422783 1.059923522396182 8.141 5.1799e− 4 43,200
MuFy 0.887156674403554 0.887389513710709 0.889346444842422 8.793 8.6964e− 6 57,400
DE 0.887156674440304 1.349480720340228 5.510017554526906 31.165 0.0139 > 100,000

11
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 11
Statistical results of the Bifacial CEA INES PV module.
PV face and model types RMSE Algorithm Best RMSE (e− 2) Mean RMSE (e− 2) Worst RMSE (e− 2) CPU Time (s) StD NFE

Back face by using ODM Classic nDEBCO 0.768726123835313 0.768726123835313 0.768726123835315 1.351 1.1006e− 17 27,400
ABC-DE 0.768726123835323 0.768726123835401 0.768726123835510 1.786 3.0887e− 15 31,900
SCA 0.768849059538067 0.768849059538238 0.768849059538526 7.281 4.4426e− 15 > 50,000
ELBA 0.768726123835336 0.768726123835413 0.768726123835563 2.929 7.3823e− 15 46,100
Lambert W nDEBCO 0.620305304806038 0.620305304806038 0.620305304806038 11.024 8.4939e− 18 19,800
Function ABC-DE 0.620305304806038 0.620305304806038 0.620305304806124 21.290 1.8280e− 16 25,000
SCA 0.620309420283874 0.620309420283982 0.620309420284398 35.580 4.5936e− 16 49,000
ELBA 0.620309837642930 0.620309837642979 0.620309837643289 36.221 6.1238e− 16 49,500

Back face by using DDM Classic nDEBCO 0.768726123835202 0.768726123825334 0.768726123829643 1.247 6.9096e− 16 21,650
ABC-DE 0.768726123835297 0.768726459393938 0.768984739502393 1.935 3.1082e− 12 33,600
SCA 0.768726123835308 0.768726273575701 0.768730323455276 4.957 7.5386e− 9 40,600
ELBA 0.768726123835312 0.952999873645374 3.721920162187199 4.271 0.0069 > 50,000
MuFy 0.769544700898595 1.188400726301233 2.567097287451603 8.923 0.0064 > 50,000
DE 0.783926001533706 1.106937163180834 2.567097287451621 11.72 0.0066 > 50,000
LWF nDEBCO 0.330099082719454 0.330099149066920 0.330099339579255 163.702 1.1005e− 9 259,100
ABC-DE 0.367235182736691 0.367236271399134 0.367530328473234 > 225 4.9027e− 6 > 300,000
SCA 0.376345798977186 0.433714741036501 0.610305304806072 > 225 7.8700e− 4 > 300,000
ELBA 0.6203053047429 0.702216741162482 0.866269041852402 > 225 0.0012 > 300,000
MuFy 0.620305304806027 0.620364032104595 0.620598941298449 > 225 1.1745e− 6 > 300,000
DE 0.620305304806030 1.796513040651938 6.501343984035382 > 225 0.0235 > 300,000

Front face by using ODM Classic nDEBCO 0.802996313821798 0.802996313821798 0.802996313821798 0.883 9.2120e− 18 15,800
ABC-DE 0.802996313821935 0.802996313821984 0.802996313822109 1.260 3.3517e− 15 20,100
SCA 0.802996313821801 0.802996313821845 0.802996313821893 5.421 1.7625e− 16 44,200
ELBA 0.802996313821840 0.802996313821898 0.802996313821939 2.461 8.3342e− 16 35,600
Lambert W nDEBCO 0.739515498392151 0.739515498392151 0.739515498392151 10.480 8.1949e− 18 15,900
Function ABC-DE 0.739515498392151 0.739515498392227 0.739515498392316 12.831 3.9086e− 15 19,000
SCA 0.739515498392082 0.739515498392123 0.739515498392293 29.896 1.0356e− 16 36,500
ELBA 0.739515498392105 0.739515498392132 0.739515498392185 16.767 7.9209e− 17 32,000

Front face by using DDM Classic nDEBCO 0.802996313821784 0.802996313821864 0.802996313821894 0.938 4.4310e− 16 16,600
ABC-DE 0.802996313821789 0.802996347857929 0.803013894738332 1.408 3.8301e− 12 24,700
SCA 0.802996313821795 0.802996313821824 0.802996313821841 7.408 4.2280e− 15 > 50,000
ELBA 0.802996313821800 0.804500403443446 0.829671150135034 2.940 5.2312e− 5 > 50,000
MuFy 0.802996313821796 1.198816896982812 2.496938274676262 9.171 0.0072 > 50,000
DE 0.802996313821798 0.916036227015101 2.496786141874805 10.723 0.0042 > 50,000
LWF nDEBCO 0.607522826652045 0.607522826652236 0.607522826652523 169.174 2.0355e− 15 214,500
ABC-DE 0.607522826652061 0.607522826686938 0.607522826769045 179.218 9.2918e− 13 241,700
SCA 0.614405644411450 0.690238483907934 0.739515498392134 > 360 6.0364e− 4 > 300,000
ELBA 0.636239122518336 0.677322715047656 0.734019773227902 > 360 4.0354e− 4 > 300,000
MuFy 0.739515498392078 1.846189827486429 6.272874933767382 > 450 0.0221 > 300,000
DE 0.657285121109731 0.723069422935621 0.739515498392164 > 450 3.2892e− 4 > 300,000

{
BsNP+1− if i⩽δ {
Bsi = i
(10) Bv =
BNP− v if v > Kf
(11)
Bsi Otherwise Bv Otherwise

while δ is the maximum number of selected leader bees according to the while Kf is the keeping factor, calculated from Kr which is the
selection factor Sf . For instance, if Sf is 3 and NP is 20, δ will be 7 keeping rate. If Kr is 0.9, which means 90% of the population will be
(rounded value of 6.6667). Where s is the dimension index chosen kept. Assuming that NP is 20, so Kf is 18 allowing to get 90% of the
randomly, meanwhile s1 , s2, and s3 have different values from the population size. Where, v is the bee’s index after the sorting from 1 to
dimension which is 5 in the case of ODM. It should be noticed that when NP. We can introduce another factor namely Newf which represents the
DDM (7 dimension) or TDM (9 dimension) are used, we have utilized 10% new added population from the previous one. In the example
four different dimension indices (s1, s2, s3, and s4). mentioned above, Newf will be of 2.
To develop the proposed nDEBCO technique, the search strategy of
3.5. Elitism strategy each been, defined by Mi vector, is evaluated. This is executed in one
independent loop and it is considered to be the first strategy in the
Given the BBO technique [59], the new bee’s generation is refined nDEBCO technique. By the end of the first loop, the evaluation of the
according to a keeping factor which allows eliminating the worst posi­ probability factors then the crossover strategy will be executed to
tion bees and allocates new enhanced positions by using Eq. (11) after enhance the exploitation and the exploration phases. After that, the
sorting the whole current population. search strategy will be applied for the second time before the elitism
strategy in one other loop like in the first time.

12
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Fig. 11. NFE evolution of Bifacial CEA INES PV module : front face by using classic RMSE for (a) ODM, (b) DDM, and by using LWF RMS for (c) ODM, (d) DDM, and
back face by using classic RMSE for (e) ODM, (f) DDM, and by using LWF RMS for (g) ODM, (h) DDM.

by
Algorithm 1 nDEBCO

Input : NP, NFE_max, VarMax, VarMin, nVar, KeepRate, Sf Ob=@(x)RMSE_model(x);

Output : Gbestv, Gbestp
1 Set NFE = 0, Sum_p = 0, nKeep = round(KeepRate*NP), nNew = NP-nKeep;
2 Generate randomly the initial generation Ƥop;
3 Evaluate the objective functions : Eval = Ob(Ƥop);
4 [Gbestv, indexmin] = min(Eval);
5 Gbestp = Ƥop(indexmin);
6 NFE = NP;
7 Record = zeros(round(NFE_max),1);
8 Record(1:NFE) = Gbestv;
7 while NFE < NEF_max do
9 for i = 1 to NP do
10 Randomly select r1, r2, r3 and r4 ∈ {1,NP}, while r1 ∕
= r2 ∕
= r3 ∕
= r4 ∕
= i;
11 Generate the vectors Br1 , Br2 , Br3 , Br4 ∈ Ƥop;
12 Randomly select Q ∈ [-1, 1];
13 for k = 1 to nVar do
14 M1(k) = Gbestp(k) + Q*(Br1(k)-Br2(k)) + Q*(Br3(k)-Br4(k));
15 end for
Fig. 12. Measured and simulated I-V curves of bifacial module by ODM with
16 NFE = NFE + 1;
nDEBCO technique.
17 valM1 = Ob(M1);
18 if valM1 <= Eval(i) % Comparison of M1 to the ith bee
The pseudo-code of the nDEBCO technique is presented in Algorithm 19 Eval(i) = valM1;
20 Pop(:,i) = M1;
1. The relative flowchart is illustrated in Fig. 4. While, RMSE_model
21 end if
means the chosen objective function based on LWF or classic one. 22 if valM1 <= Gbestv % Comparison of M1 to the best bee
23 Gbestp = M1;
24 Gbestv = valM1;
3.6. Objective function 25 end if
26 Record(NFE) = Gbestv; % Recoding data based on NFE
Several objective functions could be chosen, from MAE, R2, MAPE, 27 if NFE >= NFE_max % Stopping criteria
ME to RMSE. According to several tests [3] and based on the literature 28 break;

[21], RMSE is the most suitable fitness function of this problem, defined (continued on next page)

13
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

(continued ) ( ( ) )
Vm − Im ⋅Rs
(13)
Vm − Im ⋅Rs
Algorithm 1 nDEBCO OF = Im − Iph − Isat ⋅ e n⋅Vthm − 1 −
Rsh
29 end if
30 Sum_p = Sum_p + Eval(i); It is denominated the classic RMSE objective function; it could sim­
31 end for ply be adapted to DDM or even TDM. Eq. (14) defines the explicit one,
32 for i = 1 to NP do
which is obtained by the introduction of the LWF to remove the
33 pi(i) = Eval(i)/Sum_p;
34 end for measured current value from the RMSE equation [60].
35 [pi_sort, index_pi] = sort(pi); [ ( ) ]
36 for j = 1 to round(NP/Sf) do
Rsh Iph + Isat1 + Isat2 − Vm n1 ⋅Vthm
OF = Im − − [W(β1 ) + W(β2 )] (14)
37 indf = index_pi (j); Rsh + Rs Rs
38 indl = index_pi ((NP - j) + 1);
39 Randomly select s1, s2, and s3 ∈ {1,nVar}, and s1 ∕
= s2 ∕
= s3; while, W is the principal LWF branch, and β is given by Eqs. (15) and
40 Ƥop(s1, indl(1)) = Pop(s1,indf(1)); (16).
41 Ƥop(s2, indl(1)) = Pop(s2,indf(1)); This objective function is described based on DDM which obviously
42 Ƥop(s3, indl(1)) = Pop(s3,indf(1));
could be fitted to ODM or even TDM [9].
43 end for
44 for i = 1 to NP do { ( )}
Isat1 Rs Rsh Rsh Iph Rs + Isat1 Rs + Vm
45 Randomly select r1, r2, r3 and r4 ∈ {1,NP}, and r1 ∕
= r2 ∕
= r3 ∕
= r4 ∕
= i; β1 = exp (15)
46 Generate the vectors Br1 , Br2 , Br3 , Br4 ∈ Ƥop; n1 Vthm (Rsh + Rs ) n1 Vthm (Rsh + Rs )
47 Randomly select Q ∈ [-1, 1];
{ ( )}
48 for k = 1 to nVar do Isat2 Rs Rsh Rsh Iph Rs + Isat2 Rs + Vm
49 M2(k) = Gbestp(k) + Q*(Br1(k)-Br2(k)) + Q*(Br3(k)-Br4(k)); β2 = exp (16)
n2 Vthm (Rsh + Rs ) n2 Vthm (Rsh + Rs )
50 end for
51 valM2 = Ob(M2);
52 if valM2 <= Eval(i) % Comparison of M2 to the ith bee 4. Parameter identification algorithm evaluation
53 Eval(i) = valM2;
54 Pop(:,i) = M2; The nDEBCO parameters identification effectiveness is perceived in
55 end if
56 if valM2 <= Gbestv % Comparison of M2 to the best bee
this section on five various mono-facial PV generator issues: RTC France
57 Gbestp = M2; PV cell, PWP-201 PV module, STP6 120/36 PV module, CLS 220P PV
58 Gbestv = valM2; module, and CLS 220P PV string. Afterwards, it is checked on real-time
59 end if parameters identification for CLS 220P PV string by using an adequate
60 Record(NFE) = gbestv; % Recoding data based on NFE
experimental setup. CEA INES bifacial PV module has been used to vali­
61 if NFE >= NFE_max % Stopping criteria
62 break; date the feasibility of identifying the electrical bifacial PV module, front
63 end if and rear, parameters utilizing ODM and DDM. The tests on the bifacial PV
64 end for module were carried out as follows: for each PV module two I-V curves
65 [SortedPop, SortOrder] = sort(Pop); were made; one on the front by covering the back and one on the back by
66 Pop = Pop(SortOrder);
67 Pop = [Pop(1:nKeep) Pop(1:nNew)];
covering the front. The test has been indoor using a class A solar simulator.
68 end while Subsequently, a comparison is made with robust present-day PV
69 Return Gbestv and Gbestp as the best solution. parameters identification techniques and other techniques developed
using MATLAB code. The search range of the unknown parameters for
the solar cell/module/string is presented in Table 2, it is relatively larger
than the boundaries given within the other recent references [12,61].
Eq. (12). Usually, the error is the difference between the measured and It is worthy noticing that the limited search range helps the meta-heu­
the estimated currents. ristic technique to reach the global optimum. Meanwhile, a larger one is
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ feasible to verify the superiority of one technique compared with the others
√
√ 1 ∑ hmax
OF = √ f (θ, Is , Vm )2 (12) mainly in the problem of parameters extraction where variety of commer­
hmax h=1 cial PV modules with different series connected PV cells are used in practice.
These modules are exposed to irradiances higher than 1000 W/m2, espe­
while Is is the simulated current and θ contains the unknown param­ cially in the desert, which enlarge the photo-generated current value.
eters, hmax is the maximum number of I-V data and h is the I-V data index. Furthermore, the use of PV string/array is more practical for power gen­
In reality, there are two mean objective functions, the implicit and eration and the parameter extraction of the whole PV array requires higher
the explicit one. Eq. (13) defines the implicit RMSE which is based values of Rs and Rsh, and even n if the total number of series PV cells is not
directly on the error by using the five parameters of ODM as follows. included in the thermal voltage (Vthm) formula [61].

Fig. 13. Measured and simulated I-V and P-V curves of CLS 220P PV string
under several conditions. Fig. 14. I–V curve measurement and extraction setup.
14
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

For instance, DE technique needs 26,200 NFE to reach the best RMSE indicates that nDEBCO technique outperforms the recently published
by using ODM with LWF when the search range is large for PWP-201 PV competing techniques mainly FC-EPSO3 [11], and C-HCLPSO [62].
module as it is described in Table 2. However, it requires less than While, - means not available. Table 4 shows the five extracted param­
19,000 NFE if it is limited as it is listed in [61]. Where, the ideality factor eters by means of several objective functions, it shows that the based-
accordingly is in the range of [1, 1.39]. Meanwhile, nDEBCO reaches the absolute error forms (MAE and SAE) have similar parameters, such as
target value by 2100 NFE for using narrow search range instead of 5080 the Rsh value (=56.32281454055 Ω). At the same time, the squared-
NFE by using larger search range as it is depicted in Fig. 5. based error forms have approximately similar parameters like the
Hence, it is convenient to choose such interval to cover all possible value of Iph (=0.760787966 A). Where OF is the objective function value.
parameters variation according to irradiance fluctuation, and that could For using three diode model for PV cell parameter extraction, the
evolve more the proposed techniques applicability and permit to assess ideality factor (n3) of the third diode is maximized to 2, and its reverse-
the algorithm fitness evolution and the CPU execution time. bias saturation current (Isat3) higher value is set to 1e− 2.
In Table 2, the PV Module/string column is dedicated to the PWP- Fig. 7 illustrates a good fitting between the measured and the
201, STP6-120/36, CLS 220P PV modules, and CLS 220P PV string. simulated I-V and P-V curves by means of ODM with classic RMSE of the
They share the same search interval for ODM and DDM, and it is the case proposed nDEBCO for RTC France PV cell case study.
for all manufactured PV modules and PV strings by taking into consid­ Table A1 of Appendix presents the measured and the simulated
eration the constitutive number of series and parallel PV cells. current–voltage data, and the individual absolute error (IAE) of the CLS
For instance, the STP6-120/36 contains 36 PV cells in series which 220P PV module and the CLS 220P PV string by using LWF for ODM and
means that Ns = 36 will be produced by the ideality factor parameter DDM. While, the CLS 220P PV module is under 701 W/m2 at 319.82
with the range of [0, 2], or included in Vthm equation. For the PV studied Kelvin, and the CLS 220P PV string is under 829 W/m2 at 325.69 Kelvin
PV string, there are three PV modules connected in series, each module of cells temperature. By using the PV module with ODM, the IAEs are in
contains 60 series connected PV cells, that yields Ns = 180. the interval of [3.3e− 5, 0.0132] where the mean value is of 0.0054. For
The RMSE metric is the most used discrimination tool between the the PV string the IAEs are in the range of [0.44e− 3, 0.02046] and the
competing techniques for parameters extraction problem in the litera­ mean value is of 0.0077.
ture as it is mentioned in [10]. First, it is adopted to compare the In almost all the works presented in the literature, it is not possible to
nDEBCO findings with present-day results. Then because of the good check their RMSE results as they exhibit only a few extracted parameter
accuracy given by means of the RMSE objective function. The other used numbers after the comma, if we try to verify that relative RMSE we will
metrics as objective function are: Summation Absolute Error (SAE), Sum find another one else. For example, for the application of PWP-201 PV
Squared Error (SSE), and Mean Absolute Error (MAE) [62]. module, if we select the parameters of the reference [61], it is mentioned
Fig. 6 shows the convergence evolution of the nDEBCO technique that the minimum RMSE is 2.01664954966e− 3, however if we check it by
with the abovementioned objective function types. This is by using using classic RMSE we will find an RMSE equal to 2.541008098595e− 3,
nDEBCO technique for RTC France PV cell under ODM with LWF. The then if we use LWF it will be 2.057844235894e− 3.
best convergence speed is given by RMSE, it requires 2150 NFE and Furthermore, in [61] and [11] it is mentioned that the optimum by
0.775 s of CPU run time. The second objective function is SSE as it re­ using classic RMSE for RTC France PV cell under ODM application is
quires 2530 NFE and the CPU run time is of 0.824 s. The absolute 7.730063e− 3, and 7.7301e− 3 respectively, but if we check it we will get
objective function forms need more computational time, MAE is with a higher value due to the limited parameters resolution. Indeed, the
7300 NFE (2.379 s) and SAE is with 8350 NFE (2.722 s). Table 3 presented extracted parameters are close to those by utilizing LWF.

Table 12
Sample of the lookup table (formed by 1080 I-V curves) extracted from CLS 220P PV string at several operating conditions of solar irradiance and cells temperature.
G (W/m2) T (K) Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 2) StD (e− 15) CPU run time (s)

965.79 322 8.57450 1.59660 1.71487 1432.04 1.29660 0.98595 0.128392 1.95
961.82 322 8.62869 1.65817 1.70157 473.850 1.29763 1.20136 0.230198 2.16
957.18 323.11 8.59383 2.13568 1.67325 540.952 1.30824 1.46038 0.338129 2.27
952.36 322.86 8.52321 1.60657 1.72408 563.339 1.28486 1.16416 0.194389 2.31
947.55 322.85 8.49591 0.86026 1.75031 435.658 1.23891 1.53370 0.393817 2.17
945.36 323.40 8.47621 1.49849 1.73312 514.607 1.27494 1.37685 0.123892 1.89
918.72 324.24 8.26700 1.29248 1.78024 432.375 1.25735 1.07025 0.291823 1.96
908.02 323.77 8.16386 1.38424 1.73666 466.347 1.26451 1.32812 0.188392 2.44
889.50 323.22 8.00628 1.12742 1.74777 485.215 1.25376 1.17525 0.203918 2.18
879.21 323.34 7.91596 1.27120 1.73102 484.240 1.26495 1.18125 0.382732 2.02
867.72 323.49 7.81825 1.10307 1.74996 503.644 1.25182 1.25386 0.321738 1.98
863.83 322.01 7.70197 0.53142 1.77036 809.801 1.21604 1.09742 0.328482 2.49
855.52 323.28 7.73169 1.00572 1.75290 426.043 1.24673 1.01612 0.423487 2.35
849.63 324.34 7.60754 1.31916 1.71773 880.283 1.25861 1.07828 0.473772 2.16
848.74 325.16 7.60593 1.31595 1.72972 832.559 1.25256 1.07453 0.238242 2.22
848.54 324.39 7.59851 1.38953 1.71481 852.023 1.26478 0.87779 0.327482 2.37
847.11 325.17 7.59878 1.29786 1.72905 750.751 1.25081 0.93056 0.346732 2.10
846.84 325.13 7.60139 1.48461 1.71386 742.948 1.26091 0.88576 0.423716 2.44
846.59 324.39 7.58464 1.17689 1.71932 778.892 1.25006 0.83702 0.384728 2.09
846.58 325.38 7.59474 1.48118 1.73449 793.821 1.25844 0.85735 0.073492 1.93
210.77 307.20 1.34092 1.6235 1.7238 1216.78 1.20806 0.115919 0.090384 1.22
206.29 307.05 1.29188 1.2176 1.7129 1341.56 1.21181 0.105159 0.138849 1.42
203.81 308.27 1.29157 1.2246 1.7324 769.850 1.21005 0.089416 0.129372 2.45
202.60 306.95 1.25145 0.7769 1.7109 1456.27 1.21759 0.111370 0.348293 1.84
199.02 306.68 1.21685 1.1656 1.7783 1396.48 1.22551 0.126751 0.453849 1.92
194.90 306.53 1.17713 1.3465 1.7238 1292.79 1.21998 0.129752 0.073284 2.06
194.36 306.10 1.25490 1.3419 1.7827 580.424 1.24226 0.314148 0.080230 2.43
190.43 306.47 1.13146 0.9283 1.7928 1227.91 1.20696 0.141636 0.134399 2.23
186.49 306.25 1.08719 1.2324 1.7028 1309.50 1.19490 0.146564 0.388272 2.03
180.60 301.23 1.06255 1.4337 1.7031 1149.37 1.20894 0.076454 0.239883 1.81

15
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Fig. 15. PV string simulation model by using MATLAB/Simulink/SPS.

The nDEBCO performance is compared with the following state of 5. Results analysis and discussion
the art techniques: NPSOPC [49] GWOCS [21], CWOA [63], IJAYA [44],
MADE [35], MLBSA [42], SGDE [50], CPMPSO [64], HISA [65], COA The proposed algorithm is compared with several robust present-day
[66], FC-EPSO [11], LCROA [61], MOWOA [67], WOA [68], ITLBO state-of-the-art algorithms to resolve the parameter extraction problems.
[12], MTLBO [22], ImSMA and SMA [29], FFO [31], MRao-1 [47], RLDE We notice that all codes were performed using MATLAB2019a
[45], SFS [32] and MRFO [69]. These techniques are chosen owing to platform under Windows 7–64 bits environment on a device using Intel
the fact that they are enhanced or hybrid present-day and their (R) Core(TM) i7-3632QM CPU @2.20 GHz and 6 GB of RAM. The
mentioned performances are competitive. maximum number of function evaluations is limited according to the
In addition, the references [70–73] are compared with the developed case study. The standard maximum NFE is 50,000 whereas after several
technique since they have used classic iterative-based algorithms such as tests if the techniques require more NFE to get smaller RMSE values, the
the nonlinear least-squares (NLLS) and two-step linear least-squares maximum NFE will be enlarged accordingly to reach 1,000,000 if
(TSLLS) for LWF objective function to find the target parameters. necessary permitting to well assess the competitive techniques even if it
is a particular situation owing to the application of TDM with LWF
which requires more computational efforts.
As mentioned above, the well-known I-V data are exploited to verify

Fig. 16. PV sub-string simulation model by using Matlab/Simulink/SPS.

16
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

To verify the feasibility of the proposed algorithm, three diode model

has been used which is known to be more computational than the ODM
and DDM. In addition, RTC France PV cell is the simplest case study in
comparison to the other PV modules and PV string in terms of compu­
tational time. It assesses the exploitability of the findings meanly in
terms of the required CPU run time.
Table 7 presents the results by using TDM. For classic RMSE, the
nDEBCO best value is 9.50372616309287e− 4 which is similar to the
others. However, by using LWF, nDEBCO gives the best value compared
to the recent published works. It finds 5.72894461132072e− 4 which is
better than MRFO (7.5936e− 4) by 32.5% [69], and FC-EPSO3
(7.4375e− 4) by 29.8% [11]. From the results shown, it is apparent
that ABC-DE and SCA are strong competitive techniques to nDEBCO. In
the case of using LWF for TDM, the best value given by ABC-DE and SCA
are higher than nDEBCO best RMSE by 2e− 5% and 2.2e− 5% consecu­
Fig. 17. Measured (colored) and simulated (black) I-V curves of several tively. Knowing that ELBA is higher than nDEBCO best RMSE by 31.1%.
shading scenarios of CLS 220P PV string. Hence, by looking only at the minimum RMSE found, the overall best
ranking of the proposed techniques for RTC France is: nDEBCO in rank 1,
the robustness of the developed algorithm with other extraction tech­
ABC-DE in rank 2, SCA in rank 3, and ELBA in rank 4.
niques drawn from the literature within both options of RMSE. We should
The minimum RMSE given by nDEBCO is similar or slightly better
notice that the algorithms are in the same conditions, where the maximum
than the others for the PWP-201 PV module using classic RMSE as it
number of evaluations and population size are similar for each case study.
listed in Table A2 of Appendix. However, it is better than LCROA
(2.09537667e− 03) by>2.3% in the case of using LWF for ODM [61].
Using DDM, nDEBCO shows similar results than SCA and ABC-DE for
5.1. Solutions accuracy and comparison
classic RMSE. But it is better than ELBA by 0.9% and SCA by 9.6% under
LWF as it is listed in Table A3.
Table 5 illustrates the comparison results of different parameter
The best RSME of nDEBCO by using ODM, for STP6-120/36 PV
extraction techniques and the RMSE values of the RTC France PV cell for
module, is improved compared to MLBSA by 2.9% [42], and it is similar
ODM. Where, the best results are shown in bold font.
or slightly better than the other competing techniques as it is illustrated
By using classic RMSE, nDEBCO is similar or better than SGDE but it
in Table A4. However, for LWF it gives 1.42510635576861e− 2 which is
outperforms the other compared techniques. Meanwhile, by using LWF,
better than SFS minimum value (0.0159) by>11.5%. Even if the authors
nDEBCO is better than the others. It outperforms the recent published
of [32] have used only 22 data points which should give less RMSE than
SFS technique [32] by an RMSE enhancement of 2.59%.
using 24 data points adopted in this work [71].
Double diode model results of RTC France PV cell are illustrated in
For DDM application, nDEBCO gives similar results than SCA, ABC-
Table 6. Indubitably, nDEBCO is superior to the recently published SGDE
DE, ELBA, DE, and MuFy in terms of RMSE minimum value by using
[50], where nDEBCO reaches 9.82484806504419e− 4 but SGDE gives
LWF or classic one as it is listed in Table A5.Table A6 shows the CLS
9.8441e-4 by using classic RMSE. Whereas nDEBCO outperforms the
220P PV module results which are almost similar when using ODM.
recent published works by using LWF. As its best value
However, as it is listed in Table A7, nDEBCO under classic RMSE and
(6.33514460929039e− 4) is better than LCROA best value (7.490069e− 4)
DDM reaches 4.33929841502380e− 3 which is better than ABC-DE
by>18% [61], and MRFO best value (7.6046e− 4) by>20% [69].

Fig. 18. Shading patterns graphical description.

17
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

Table 13 CPU run time required is 0.134 with 2050 of NFE, these results outper­
Algorithms parameter setting. form ABC-DE ones (0.627 of CPU run time and 3450 of NFE) and the
Algorithm Parameter others. By using LWF, the difference between the nDEBCO best and mean
values is 9.7578e− 19 with StD of 7.8785e− 18. It is a good result
nDEBCO NP = 20; Sf = 3.2 (Selection factor); Kr = 0.9 (Keeping rate).
ABC-DE NP = 20; Sf = 2 (Selection factor). compared to LCROA (2.1851e− 2) [61], and FC-EPSO3 (3.3351e− 10)
ELBA NP = 20; ρi = 0 (Initial value of loudness); ρf = 1 (Final value of [11]. Concerning the convergence speed, nDEBCO is highly better than
loudness); ABC-DE and the others, it requires less than half the time needed by ABC-
ai = 0 (Initial value of pulse rate); af = 1 (Final value of pulse rate); DE with only 2150 NFE instead of 6500 NFE as shown in Fig. 8(d). In the
fri = 0 (Initial value of frequency); frf = 5 (Final value of frequency);
β = 1.7 (Lévy flight exponent).
same context, SCA is the most competitive algorithm in this application
DE NP = 20; βmin = 0.68 (Lower bound of scaling factor); regarding the StD, the worst RMSE, and the CPU run time. From this
βmax = 0.72 (Up bound of scaling factor); Cr = 0.75 (Crossover relatively simplest case study results, we can remark the improvement
probability). done by using nDEBCO compared to ABC-DE. If we admit a time delay of
SCA NP = 20; nPop_complex = 8 (Complex Size);
1.5 s to 2 s, using LWF is more convenient for this application.
n_complex = 5 (Number of Complexes); α = 4 (Number of Offsprings);
b = 4 (Maximum Number of Iterations); The error between the best and the mean is 6.2909e− 10 with StD of
MPSO NP = 20; Wmin = 0.4 (Lower inertia weight); Cmin = 1 (Lower learning 2.4087e− 9 in the case of using DDM with classic RMSE. This is better
factor); than MTLBO, where the best and the mean RMSE difference is
Wmax = 0.9 (Upper inertia weight); Cmax = 2 (Upper learning factor). 2.2600e− 8 with StD of 3.3000e− 9 [22]. Obviously, the best StD is the
ACO NP = 20; nSample = 400 (Sample Size); q = 0.5 (Pheromone quantity);
zeta = 0.5 (Deviation-Distance Ratio).
minimum one with the smallest difference between the mean and the
BRO NP = 20; MaxFault = 4; CrossType = 1. best RMSE. The nDEBCO convergence time is 0.789 s with 16,950 NFE,
MuFy NP = 20; γ = 1.2 (Light Absorption Coefficient); β0 = 1.75 (Attraction as it is depicted in Fig. 8(b). It is the best compared to the most
Coefficient Base Value); α = 1.95 (Mutation Coefficient); αdamp = 0.75 competitive ABC-DE where its CPU run time is 1.549 s with 31,800 NFE.
(Mutation Coefficient Damping Ratio); βmin = 0.69 (Lower Bound of
By using LWF, the problem is more computational. nDEBCO is the
Scaling Factor); β0 = 0.72 (Lower Bound of Scaling Factor);
δ = 0.75*(VarMax-VarMin); m = 2. unique technique which finds the minimum RMSE with less than
100,000 NFE, it needs 64,550 NFE with 30.13 s. But, all the others
(4.56637429010392e− 3) by 5.2%, SCA (5.30966285331865e− 3) require>100,000 NFE with higher than 45 s of CPU execution time, as it
by>22.3%, and ELBA (8.32153987299546e− 3) by 91.77%. Further­ is mentioned in Fig. 8(e). MuFy and DE standard deviations are slightly
more, by using LWF, nDEBCO gives 3.60745466153e− 3 that is better better than the StD of nDEBCO. However, as the standard deviation is
than ABC-DE (6.46612803152360e− 3) and SCA based on the mean value, by looking at the best and the mean values of
(6.46612803152361e− 3) by>79.2%. This case study is interesting to MuFy (7.73e− 4) and DE (7.73e− 4) techniques in comparison to
verify the accuracy of the competing techniques. For that, the measured nDEBCO (6.33e− 4). We cannot consider their standard deviation values
I-V data are provided in Table A1. In addition, it confirms that the competitive as they fall into local minima in all the 30 independent runs,
parameter extraction problem based on the measured I-V curve of the and their mean values are far from being the right ones. Hence, for this
same PVG is different from one operating condition to another. So, the application the competing algorithms are classified as follows: nDEBCO
developed or the chosen algorithm should be verified for various I-V is the first rank, SCA is the second, and ABC-DE is the third.
curves with different irradiances and cells temperatures. The nDEBCO gives the best results in the case of TDM with classic
The CLS 220P PV string results under ODM and DDM do not show RMSE, as its StD is 6.3492e− 7 with 2.287 s of CPU execution time. It is the
remarkable differences between the algorithms, as it is listed in unique algorithm which requires less than 50,000 NFE (36900) to reach
Tables A8 and A9. Since the six competing techniques give the same the best RMSE, as it is displayed in Fig. 8(c). SCA is a good technique, as its
RMSE for both objective functions. For the CEA INES Bifacial PV module maximum value is 9.86021878186181e− 4 which is the best among the
the minimum RMSE values illustrated in Tables A10–A12 do not show remaining techniques, and its mean value is close to the nDEBCO one.
significant segregation between the competing algorithms. However, in Further improvement of this technique will yield interesting results.
Table A13, it is apparent that nDEBCO outperform the other techniques LWF application for TDM needs much more computational features;
for LWF. From the abovementioned tables, nDEBCO has the accurate nDEBCO is the only one which requires less than 1,000,000 NFE
results compared to the recent published techniques in the literature and (466000) to find the best RMSE (0.5728944611320e− 4), as it can be
the developed ones. seen in Fig. 8(f). Where, the error between the best and the worst is
2.2000e− 15 with StD of 4.3982e− 16 and 622 s of computational time.
The nDEBCO is the best computational technique since all the others
5.2. Statistical results and robustness need more 900 s to converge into the global minima, while SCA is the
closest to nDEBCO with StD of 1.3872e− 9.
The statistical results are conducted with 30 independent runs where Using LWF for DDM or TDM could reduce significantly the RMSE
the population size and the maximum NFE are similar. The illustrated value but it needs 30.13 s and 622 s of CPU run time respectively by using
results are mainly the best, the mean and the worst RMSE values. Three nDEBCO which is not feasible for our perspectives. This case of applica­
other metrics are mentioned to assess the competing techniques: the CPU tion is presented especially to assess the skills of the proposed algorithms.
run time, the NFE required for reaching the best RMSE value, and the That confirms the improvement of nDEBCO compared to ABC-DE, and its
mean standard deviation. It should be indicated that the CPU run time and performance quality in comparison to the other techniques.
the NFE needed to find the best RMSE are calculated to reach 1e− 9 of In this case study, under ODM, the RMSE is reduced by 21.6% for
resolution. For instance, in the case of STP6-120/36 PV module by using using LWF instead of classic RMSE, and going to 35.52% for DDM, and
ODM, the time and the NFE have been calculated when reaching 39.72% for TDM. Nevertheless, it needs 0.62 s, 29.34 s, and 619.71 s
0.016600603. Knowing that, the time to achieve 50,000 NFE is of 2.15 s. more computational time respectively. While by using LWF, we can
Table 8 and presents the statistical results of the RTC France PV cell, reduce the RMSE best value by 18.09% for DDM instead of using ODM,
while StD is the RMSE standard deviation. Fig. 8 presents the convergence and it is minimized by 9.56% from DDM to TDM.
curves of this case study. For ODM, the worst RMSE of nDEBCO is the The RMSE value is reduced by 41.89% for using LWF with TDM instead
minimum compared to ABC-DE and the other competing algorithms. The of using classic RMSE with ODM. Meanwhile it requires 621.8 s additional
difference between its best and its mean values is 1.1059e− 17 with a CPU run time to find the target value. Accordingly the first enhancement, by
standard deviation of 1.8824e− 17. That is slightly better than the using LWF with ODM, is suitable in terms of error percentage improvement
recently published MTLBO (1.9092e− 17) [22]. In terms of rapidity, the (21.6%) and the additional time needed (0.62 s).

18
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

The nDEBCO presents the best CPU run time in classic RMSE by using PV string by using ODM with interesting standard deviations of
ODM for the PWP-201 PV module, as it is described in Table 9. The 5.6808e− 17 and 1.0125e− 17 relative to classic RSME and LWF
nDEBCO has been improved by 3900 NEF and 0.47 s of CPU run time consecutively. While, the RMSE enhancement from using classic RMSE
compared to ABC-DE as it is shown by Fig. 9(a). In comparison to the (1.034738476694e− 2) to LWF one (0.887346444842e− 2) is of 16.61%
other algorithms, nDEBCO gives the best StD for classic RMSE with NFE reduction of 3800, as it is displayed in Fig. 10 (f) and (g).
(1.0833e− 17) and for LWF (1.0348e− 17), while LCROA gives a StD of When using DDM, nDEBCO remains the best in terms of convergence
1.5463e− 1 [61]. speed as it is mentioned in Table 10. While, SCA is better than nDEBCO
For DDM, nDEBCO presents the best CPU run time (0.575 s) with by looking at the standard deviation in both objective functions. Where,
10,300 NFE for classic RMSE, as it is depicted in Fig. 9(c). For LWF, it StD of SCA is 1.9500e− 16 for classic RMSE compared to 1.0388e− 14 of
needs 45.129 s with 142,000 NFE as presented in Fig. 9(d). In terms of nDEBCO. And it is of 8.6964e− 7 in comparison to 3.9122e− 6 of
standard deviation, SCA gives the best result (4.7244e− 17) under classic nDEBCO for LWF. This is an interesting case study which requires an
RMSE compared to nDEBCO (2.8012e− 15) and ABC-DE (2.1727e− 13). enhancement of the developed nDEBCO to be competitive to SCA
However, nDEBCO behaves much better than the other techniques regarding the standard deviation. It should be noticed that with other I-V
including SCA when using LWF. It gives better StD than SCA with a curves of the studied PV string, nDEBCO behaves better than SCA,
difference in the maximum RMSE of 10.97%. In addition, SCA is not able however using this I-V curve will encourage making further improve­
to reach the best RMSE by using LWF. ment of the proposed algorithm.
As it is apparent from Table 9 and Fig. 9, in STP6-120/36 PV module The RMSE is minimized by 14.24% from using classic RMSE to LWF
case study, the nDEBCO performs well except for the StD value under for ODM, and by 14.26% for DDM. But it needs 1.719 s and 4.334 s more
classic RMSE where the StD of SCA is 1.7724e− 16 in comparison to computational time respectively. While by using LWF, the RMSE best
4.8727e− 15 of nDEBCO. However, it behaves better by looking at the value is reduced by 0.02% for DDM instead of using ODM where the
CPU run time (0.276 s) and the required NFE (6420) compared to SCA needed additional CPU run time is 2.776 s which is not quite interesting.
(1.866 s of CPU run time and 15,550 NFE), as depicted in Fig. 9(f). It CEA INES bifacial back and front PV module parameter extraction
confirms that by using the same PV module and even the same I-V curve requires relatively more computational skills compared to the mono-
with variant models (ODM, DDM, or TDM), the techniques could have facial technology. The bifacial module I-V curve convexity allows
different behaviors and the best is the one which get the overall best assessing the techniques performances. nDEBCO is the best technique
results. This is the raison to present several PV cell/module/string sta­ regarding the standard deviation, the maximum RMSE value, and the
tistical results with different models. convergence speed. The StD of nDEBCO is 1.1006e− 17 for the back face
By using ODM with LWF, ABC− DE behaves well and it is close to with classic RMSE, and it is less than 9.3e− 18 for the remaining case
nDEBCO by looking at CPU run time and the required NFE, as displayed studies of ODM for back and front faces (classic and LWF), as it is
in Fig. 9(g). However, in terms of standard deviation, the nDEBCO StD mentioned in Table 11 and shown by Fig. 11.
(2.37e− 17) is better than ABC-DE (8.1784e− 15) and the others. Where, Fig. 12 presents the measured and the simulated I-V curves of
For the CLS 220P PV module by using ODM, as it is mentioned in both faces by means of nDEBCO with ODM and classic RMSE. It can be
Table 9 and Fig. 10(a), all the competing techniques reach the best observed that the simulated I-V curves match well the measured ones by
RMSE with less than 41,500 NFE. Meanwhile, they differ from the StD an RMSE lower than 1e− 3.
where nDEBCO is the best with 1583e− 17 and 0.315 s of computational However, in terms of CPU run time, the difference between using
time. LWF and classic one is higher than 9.6 s for both faces. CPU run time
Almost the same under LWF, all techniques give less than 36,500 does not exceed 1.4 s when using classic RMSE and it is of 11.1 s when
NFE where nDEBCO is improved by 4200 NFE compared to ABC-DE, as it using LWF despite the larger search range adopted. Meanwhile, the
depicted in Fig. 10(d). Also in terms of StD, nDEBCO is the best other techniques are not able to reach the target value in less than 35 s
(5.5897e− 17) and SCA is the second ranked (7.5235e− 17). It should be except for ABC-DE. With these computation conditions, the use of classic
noticed that using LWF instead of classic RMSE has permitted to enhance RMSE is more feasible even if the RMSE difference from using classic to
the RMSE by 74.96% for ODM. While, the CPU run time difference is LWF is of 1.5e− 3 and 6.3481e− 4 for both faces respectively.
0.36 s and the NFE reduction is by 6700 which means that reduced run The rest of the techniques, namely MPSO, DE, BRO, and MuFy are
time is possible by using more efficient computer properties. not able to reach the target value for the back and front sides.
In this case study by using DDM with classic RMSE, larger search Using DDM, the StD of nDEBCO is less than 7e− 17 in both faces, but
range means an ideality factors in the interval of [0, 2]. In that situation, it is of 3.1082e− 12 and 3.8301e− 12 for back and front faces with ABC-
all the algorithms take long time in the exploration procedure to find the DE consecutively. Where, the NFE enhancement is of 11,900 for the back
best RMSE. However, only nDEBCO is able to reach the desired face, and of 8100 for the front one, as it can be observed from Fig. 11(b)
4.339e− 3. In this condition, nDEBCO required CPU run time is 17.32 s and (f).
with 698,300 NFE, as it is displayed in Fig. 10(c). SCA do not reach the nDEBCO is the unique technique to reach 0.330099082714e− 2 with
target value after>1,000,000 NFE and higher than 200 s. an interesting StD of 1.1005e− 9 compared to the other competing
Considering a search range of [1,2] for the ideality factors. It allows techniques by using LWF with DDM for the back face. For the Front face,
to get CPU run time of 0.735 s with 13,900 NFE as it is illustrated by nDEBCO and ABC-DE are competitive in terms of the best RMSE.
Fig. 10(b). However, nDEBCO is the best regarding the convergence speed and the
By using LWF, the CPU run time to reach 0.646612803152360e− 2 is StD (2.0355e− 15) compared to ABC-DE (9.2918e− 13), and SCA
of 5.763 s with 13,000 NFE, as it is shown by Fig. 10(e). But, the StD is of (6.0364e− 4), as it can be seen from Fig. 11(d) and (h).
1.4075e− 4 which is not adequate by looking at SCA’s StD (9.3947e− 15) Based on these results we can remark that the ODM with LWF is more
or even ABC− DE’s value (6.9827e− 13). This is due to the best value precise than DDM by using classic RMSE for mono-facial technology.
found during 30 runs, as nDEBCO has reached 0.360745466153802e− 2 Nevertheless, it needs relatively more computational time where the
five times and it is the unique to get it. difference varies from one PV module/string to another. For instance, in
This case study could be considered a thorny problem to be the case of PWP-201 PV module, the CPU run time difference is 1.15 s
enhanced. On the first hand, to improve the standard deviation while with 5200 NFE by using LWF with ODM, as it shown by Fig. 9(c). While,
finding the RMSE best value (0.3607–2) under LWF, and on the other the required time to achieve 50,000 NFE for DDM by using classic RMSE
hand to find less NFE and less CPU execution time when using classic is 2.15 s, and the corresponding time needed for ODM by using LWF is
RMSE. For that, the I-V curve data are provided in Table A1. In both nearly 16.5 s. For STP6-120/36 PV module, the time difference is of 1.5 s
RMSE objective functions, nDEBCO is the best technique for CLS 220P with 5200 NFE, as it is presented in Fig. 9(f). For the CLS 220P PV string,

19
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

the time difference is of 1.55 s with 7000 NFE, as it can be observed from 5.4. Experimental shading scenarios simulation
Fig. 10(d), for ODM by using LWF. In the worst case, for the studied PV
string, the required NFE is less than 9000 which means the CPU run time The PV string simulated I-V curves in these shading conditions are
is less than 3 s, and hence choosing LWF with ODM is feasible if developed by means of MATLAB/Simulink model based on the extracted
maximum delay of 2 s is supported. parameters. Each PV module is constituted by three sub-strings series
Furthermore, the ideality factor of the second diode by using DDM connected, based on the lookup table described in Table 12 and by using
goes usually to the maximum value, which is not significant for fault Eq. (4) to match the relationship between the PV string and the consti­
diagnosis purposes. tutive sub-strings which are nine in our case study (See Figs. 15 and 16)..
The studied PV string has been validated under shading fault conditions.
5.3. Real time experimental validation Fig. 17 illustrates some measured and simulated I-V curves under several
shading scenarios whose patterns are graphically described in Fig. 18.
In this section we utilize data relative to the CLS 220P PV string. It is Where, SS means shading scenario, and noticed here in after:
composed of three series connected PV modules, each one of them is
consisting of 60 series connected PV cells collected in form of 3 sub- • “SS1”: the PV string is exposed randomly to a partial shading pattern
strings of 20 PV cells for each one. It has been installed on the that resembles the railing which closes the building terrace of the
coverage of the DIEES laboratories at the University of Catania (Italy) laboratory after 5.00 pm, while T = 320 K and G = 825 W/m2
buildings. The I-V characteristics have been extracted by means of a (measured I-V curve on Cyan color);
program implemented in the LabVIEW environment that controls a DC • “SS2”: Similar shading condition to “SS1”, while T = 324 K and G =
electronic load to span the voltage in a specified range. The ambient 742 W/m2 (measured I-V curve on Red color);
irradiance and the temperature have been measured using a SPEKTRON • “SS3”: the vertical half of each module is totally shaded, while T =
320 irradiance sensor and Pt100 temperature sensor respectively. 317 K and G = 676 W/m2 (measured I-V curve on Magenta color);
These data have been acquired using a data acquisition board. The • “SS4”: two sub-strings of different PV modules are totally shaded,
proposed techniques have been tested and validated using the I-V while T = 320 K and G = 561 W/m2 (measured I-V curve on Blue
curves, temperature, and irradiance data measured on specific days. In color).
the first time, all data have been acquired with a time step of about 30 s.
The I-V characteristics are constituted of 100 measured voltage and It should be noticed that the simulation model is developed by means
current data. Then, they are reduced to 25 I-V data for the PV string and of ODM with LWF. It is based mainly on the absolute error (AE) between
21 for the PV module to get less computational time (15 s) with practical the measured and the estimated current short circuit (Isc) and voltage
and fully I-V curve shape. open circuit (Voc). From the lookup table we can select the adequate
Fig. 13 presents some measured and simulated I-V and P-V curves of parameters of each PV sub-string. Where, the minimum AE for each PV
CLS 220P PV string under several irradiance and cells temperature sub-string of Isc is 0.4%, and 0.8% for Voc otherwise we should proceed
conditions. Fig. 14 illustrates the online measurement and extraction at additional measurements close to the target condition to get less AE in
setup, where the PV string is partially shaded. Table 12 illustrates a normal operating conditions. While, the Voc of one PV sub-string is equal
sample of the lookup table (formed by 1080 I-V curves) extracted from to the Voc of the PV string by 9, however the Isc is similar as the sub-
the PV string at several operating conditions of irradiance and cells strings are series connected. Further model description is provided in
temperature. Even with incorporating the sensor errors, results reveal [5].
that RMSE is less than 0.02. The CPU run time is less than 2.5 s and the The illustrated shading simulation is done to demonstrate the ability
mean standard deviation is less than 1e− 16. of the adopted simulation procedure. It is based on the parameter
In our study for fault diagnosis purposes, it is worthy to select the extraction technique and the experimental knowledge of the PV string
LWF as an objective function by using ODM. As the required CPU run under several measurements of irradiances and PV cells temperatures.
time do not exceed 2.5 s in all the measured I-V data conditions and all For shading scenario “SS1”, the RMSE value is 0.0323 and the AE be­
the extracted parameters do not fall into the maximum or minimum tween the measured and the simulated Pmpp is of 0.269% (the measured
limits of the search range. Pmpp is 408.5 W and the simulated one is 407.4 W).
Since, it is important to select an accurate technique in terms of
rapidity and precision simultaneously to well exploit the extracted pa­ • For “SS2”, the RMSE value is 0.0338 and the Pmpp AE is 0.245% (the
rameters. The CPU execution time should be reduced as much as measured Pmpp is 326.4 W and the simulated one is 325.8 W);
possible to quickly assess the possible occurrence of faults in way to • For “SS3”, the RMSE is 0.0241 and the Pmpp AE is 0.315% (the
proceed at removing it or making the adequate decision. Using DDM and measured Pmpp is 253.4 W and the simulated one is 252.6 W);
TDM reduces significantly the RMSE value, but that requires more • For “SS4”, the RMSE is 0.0292 and the Pmpp AE is 0.397% (where the
computational time and more technique features. Indeed, introducing measured Pmpp is 125.8 W and the simulated one is 125.3 W);
Lambert W function (LWF) based RMSE objective function, to enhance
the RMSE value requires more execution time especially when using In addition, the use of the parameters extraction based on the LWF
DDM and even more with TDM. objective function reduces slightly the RMSE error in the case of shading
Spending much more time is not feasible for online application in pattern simulation. For instance, in the first shading scenario (SS1) the
real conditions as we should include the additional time of measurement RMSE value is 0.0323 based on the LWF parameters. But it is 0.0527 by
and recording of the I-V curve data (relatively>15 s). Without forgetting using classic RMSE based parameters estimation.
the rapid variations of the solar irradiance in few seconds which could The overall RMSE in such conditions is less than 0.045 which vali­
change the measured I-V curve real operating conditions. dates the adopted simulation procedure under normal and abnormal
Hence, the use of the identification procedure for fault diagnosis operating conditions. This is done in sight of online PVG fault diagnosis
based parameters evaluation purposes should take into account a based on the I-V curve shape and the electrical parameters residues [5].
particular factor between the supported time to make faulty decision by Therefore, these results demonstrate the robustness and the validity
including the extraction CPU run time and the required RMSE precision of the developed method here proposed. While the algorithm parameters
to carefully evaluate the parameters variation. setting are shown in Table 13, where NP is the population size and
NFEmax is the maximum number of functions evaluation.

20
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

6. Conclusion an effective candidate for the extraction of solar module/string param­

eters. It is able to be exploited for real PVG simulation under normal and
This work proposes a novel hybrid parameter extraction technique abnormal operating conditions for fault diagnosis purposes based on the
(nDEBCO) to model the PV generator under normal and shading con­ I-V curve shape and the estimated parameters.
ditions. In comparison point of view, the model-based parameter
extraction has been employed by means of ODM, DDM, and even TDM. CRediT authorship contribution statement
This is by using two different RMSE objective functions, namely Lambert
W-function and classic. The nDEBCO technique superiority has been Oussama Hachana: Conceptualization, Data curation, Formal
verified compared to the present-day techniques in the literature by analysis, Investigation, Methodology, Resources, Software, Supervision,
using well-known measured I-V curves. Using RTC France PV cell I-V Validation, Visualization, Writing – original draft. Belkacem Aoufi:
curve gives 5.72894e− 4 of RMSE and 4.3982e− 18 of StD for TDM with Conceptualization, Data curation, Methodology, Resources, Software,
LWF. Visualization, Writing - review & editing. Giuseppe Marco Tina:
Meanwhile, under ODM with LWF it yields 7.73006268e− 4 of Conceptualization, Methodology, Supervision, Validation, Visualiza­
RMSE, 7.8785e− 18 of StD, and 2150 NFE. For PWP-201 PV module tion, Writing - review & editing. Mohamed Amine Sid: Methodology,
application, it shows a RSME of 0.24250748e− 2, StD of 1.0833, with Visualization, Writing - review & editing.
5300 NFE. In addition to 1.42510635e− 2 of RMSE, 2.3700e− 17 of StD,
and 5400 NFE by using STP6-120/36 PV module even if the search range Declaration of Competing Interest
is larger than the adopted one by the literature. Afterwards, the nDEBCO
has been validated with the commercial CLS 220P PV module/string The authors declare that they have no known competing financial
under several operating conditions of solar irradiance and cells tem­ interests or personal relationships that could have appeared to influence
perature by means of an adequate online experimental setup. The 1080 the work reported in this paper.
measured I-V curves RMSE of the PV string are less than 0.02, the CPU
run time do not exceed 2.5 s, and the standard deviations are less than Acknowledgements
1e− 16. This is after concluding that the ODM with LWF is the most
appropriate model for mono-facial technology. This work was supported by the General Directorate for Scientific
The CEA INES bifacial PV module has been tested as well. Where, Research and Technological Development (DGRSDT), Algeria. Thanks to
using ODM with classic RMSE is more adequate as it gives less than: the team of the laboratory of power system at DIEEI Department of
0.80299631e− 2 of RMSE, 6.9096e− 16 of StD, 1.351 s of CPU run time University of Catania, Italy.
with 27,400 NFE. However, using LWF requires>9.6 s of additional time
even if the error difference is less than 1.5e− 3 for both faces. Addi­ Appendix A
tionally, the robustness of the proposed model parameters estimation is
verified for integrated shading scenarios to the PV string. While, the
RMSE is less than 0.045 for all the shading patterns with a power
maximum point absolute error smaller than 0.397%.
Therefore, compared to the recent competing techniques, nDEBCO is

Table A1
The measured and the estimated values of current and voltage, and IAE of CLS 220P PV module and PV string.
CLS 220P PV module CLS 220P PV string

Item Experimental data Estimated data by ODM Estimated data by DDM Item Experimental data Estimated data by ODM Estimated data by DDMs

Im Vm Isim IAE (e− 3) Isim IAE (e− 3) Im Vm Isim IAE (e− 3) Isim IAE (e− 3)

1 6.036875 0.578125 6.0390891 2.2141866 6.038349 1.4746142 1 7.333 0.653 7.322795 10.2044238 7.322903 10.096055
2 6.0375 1.390625 6.0376548 0.1548987 6.036995 0.5043835 2 7.322 7.178 7.311882 10.1170444 7.311949 10.050491
3 6.0309375 3.03125 6.0347572 3.8197141 6.03426 3.3233273 3 7.314 11.093 7.305324 8.67561682 7.305366 8.6339928
4 6.04125 4.671875 6.031855 9.3949381 6.031523 9.7263348 4 7.303 15.009 7.298746 4.25334551 7.298763 4.2364194
5 6.0325 6.3125 6.0289418 3.558174 6.02878 3.7198733 5 7.281 18.924 7.292137 11.1376307 7.29213 11.130282
6 6.0246875 7.96875 6.0259728 1.2853561 6.025993 1.3055116 6 7.275 22.839 7.285466 10.4664014 7.285435 10.435502
7 6.0159375 9.609375 6.0229629 7.0254295 6.023185 7.248107 7 7.268 26.755 7.278676 10.6760668 7.278622 10.622858
8 6.0184375 11.125 6.0200349 1.5974505 6.020484 2.0470598 8 7.261 30.67 7.271666 10.6660707 7.271592 10.592672
9 6.0128125 12.78125 6.0164415 3.6290244 6.017222 4.409541 9 7.256 34.585 7.264238 8.23863404 7.264148 8.1485724
10 6.011875 14.4375 6.0118413 0.033699 6.013099 1.2240315 10 7.249 38.5 7.256024 7.02446912 7.255923 6.9235016
11 6.00875 16.09375 6.0047299 4.0200911 6.006652 2.0978674 11 7.241 42.416 7.246326 5.32632758 7.246223 5.2235801
12 5.9990625 17.75 5.9913854 7.677077 5.993988 5.0740701 12 7.231 46.331 7.233846 2.84646605 7.233755 2.755803
13 5.968125 19.40625 5.9627528 5.3721843 5.965118 3.0060816 13 7.219 50.246 7.216153 2.8465997 7.216094 2.9054619
14 5.900625 20.953125 5.9038248 3.199825 5.903133 2.5087897 14 7.204 54.162 7.188753 15.2466114 7.188751 15.248624
15 5.755 22.609375 5.7647547 9.7547778 5.757007 2.0077052 15 7.164 58.077 7.143531 20.4680343 7.143611 20.388154
16 5.458125 24.296875 5.464417 6.2920598 5.45628 1.8441344 16 7.075 61.992 7.066254 8.74522385 7.066426 8.5732772
17 4.9078125 26 4.8992594 8.5530659 4.907794 0.018027 17 6.933 65.908 6.933442 0.44253309 6.933674 0.6749382
18 4.0009375 27.734375 3.9920218 8.9156548 4.002614 1.6771964 18 6.7 69.823 6.710385 10.3859559 6.710581 10.581044
19 2.8003125 29.421875 2.8051367 4.8242323 2.797455 2.8568151 19 6.346 73.738 6.353546 7.54628373 6.353564 7.5648835
20 1.37875 31.09375 1.3919947 13.244736 1.381398 2.6481532 20 5.818 77.653 5.821229 3.22941763 5.821002 3.0027569
21 0.2571875 32.3125 0.2476707 9.5167977 0.256135 1.0520316 21 5.091 81.569 5.087922 3.07727156 5.087576 3.4234163
22 4.16 85.484 4.152629 7.37087522 4.152459 7.5400731
23 3.036 89.399 3.033141 2.85809036 3.033395 2.6041924
24 1.748 93.315 1.756082 8.08299455 1.756621 8.6218149
25 0.107 97.882 0.104793 2.20610885 0.104421 2.57804

21
 Table A2
Comparison results for PWP-201 PV module parameters by using ODM.
O. Hachana et al.

Classic RMSE

Algorithm/Author Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 3)

WOA 1.0294212 3.8525 1.1906302 1179.9442886 1.361962838 2.449051

SGDE 1.0305 3.4823 1.20127 981.9822 1.351188 2.4251
ITLBO 1.0305143 3.4823 1.2012710 981.9822603 1.351189858 2.425075
MRao-1 1.030514 3.4823 1.201271 981.9821 1.3511475 2.425075
MTLBO 1.0305143 3.4823 1.2012710 981.9823732 1.351189858 2.425075
RLDE 1.0305 3.4823 1.2013 981.9823 1.351188 2.4251
GWOCS 1.0305 3.4650 1.2019 982.7566 1.35065 2.4251
CPMPSO 1.030514 3.4823 1.201271 981.9823 1.35119 2.425075
nDEBCO 1.03051430258113 3.48226114219111 1.20127105198752 981.981568963186 1.35118716108452 2.42507486809511

LWF RMSE
Laudani et al. 1.0323759 2.5188848 1.2390187 745.6431 1.3174002 2.0465
Cárdenas et al. 1.0323377 2.517957 1.2390060 745.7122 1.3173635 2.0465
Toledo et al. 1.0323823 2.5129059 1.2392888 744.71302 1.3171591 2.0465347
Panchal et al. 1.0323827 2.5127089 1.2392956 744.6655703 1.3171511 2.0465347
COA 1.03152750757054 3.62496529235387 1.19791296762334 940.015655850509 1.40855995901487 2.94960692837003
LCROA 1.03229 2.149476 1.258598 719.87429 1.38888 2.095377
nDEBCO 1.03238235268979 2.51289308945248 1.23928990643412 744.712531838472 1.31730110057246 2.04652439739525

22
Table A3
Comparison results for PWP-201 PV module parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (µA) Isat2 (µA) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 3)

ABC-DE 1.03051429836255 2.6501471349068 8.321160368224 1.20127100308951 981.98238019182 1.3511872073853 1.35118727176975 2.42507486809507

nDEBCO 1.03051429836255 2.6501471349068 8.321160368224 1.20127100308951 981.98238019182 1.3511872073853 1.35118727176975 2.42507486809507
SCA 1.03051429838845 2.8621913128187 6.200716161000 1.20127100772148 981.98238019182 1.3511872559934 1.35118720666547 2.42507486809507
ELBA 1.03051429618887 3.4822632223767 1.825323418e− 10 1.20127100730644 981.98265146117 1.3511872240135 1.91969659850426 2.42507486809546
MuFy 1.03051430016597 3.4390904454032 0.431723572693 1.20127101263692 981.98213290010 1.3511867912422 1.35118721683113 2.42507486809522
DE 1.03051384238445 3.4822663098047 0 1.20127555922992 982.09073756045 1.3511871430517 1.96291769807921 2.42507491118383

LWF RMSE
nDEBCO 1.03747144869700 0.0434936190549 8.5924130274e− 4 2.73913352722640 512.21008946449 1.0581501530329 0.87951275699806 1.02826363303367
ABC-DE 1.0374893800293 0.0489133726491 8.7638874351e− 4 2.78376563432050 542.36284873241 1.0590340329987 0.90346673264821 1.02826363306723
SCA 1.03636599035353 0.1012354460044 2.8744459885e− 2 2.50536740618853 543.33526091570 1.1100273501118 1.04080280547623 1.09949875545313
ELBA 1.03753687403707 0.0540459214825 1.9158596586e− 3 2.68485485115715 510.82917112818 1.0710581440546 0.91157820982426 1.03793910143812
MuFy 1.03690454897501 0.0750503593693 6.7133678853e− 3 2.60561826202753 528.25181596150 1.0916180981172 0.96673341457084 1.05782431704444
DE 1.03418006417590 0.3413941678172 3.0536663538502 1.96528514663360 626.20563265399 1.1800165604172 1.40635355905566 1.35119211305484
Energy Conversion and Management 248 (2021) 114667
 O. Hachana et al.

Table A4
Comparison results for STP6-120/36 PV module parameters by using ODM.
Classic RMSE

Algorithm/Author Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 2)

MLBSA 7.4779086 6.3014 0.0040388 1500 1.3488426 1.7091055

IJAYA 7.4672 2.2536 0.0046 27.5925 1.2571 1.6731
MADE 7.4725 2.3350 0.0046 22.2199 1.2601 1.6601
RLDE 7.4725 2.3350 0.0046 22.2199 1.2601 1.6601
nDEBCO 7.47252991755498 2.33499580425185 0.165406839898594 799.916969236867 1.26010104060649 1.66006031250859

LWF RMSE
Toledo et al. 7.4752843 1.9308835 0.168918214 570.189132 1.2443189 1.42516
SFS 7.4757 3.01 0.1600 827.5815 1.2816 1.59
Panchal et al. 7.4752544 1.933973 0.1688919 546.2421092 1.2444467 1.42931
nDEBCO 7.47528406813505 1.93088757652929 0.168918188786233 570.197356671310 1.24445374463979 1.42510635576861

23
Table A5
Comparison results for STP6-120/36 PV module parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (A) Isat2 (A) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 2)

nDEBCO 7.4725299216758 2.3349951714e− 6 1.167224977408e− 18 0.16540684445391 799.916475336023 1.26010101791577 1.36392052044463 1.66006031250850

ABC-DE 7.4725299216758 2.3349951714e− 6 1.167224977408e− 18 0.16540684445391 799.916475336023 1.26010101791577 1.36392052044463 1.66006031250850
SCA 7.4725299247716 9.6110268810e− 7 1.373892419787e− 6 0.16540684523923 799.916363242153 1.26010099651971 1.26010102916048 1.66006031250856
ELBA 7.4725299145383 2.3349950303e− 6 2.044730428749e− 18 0.16540684596752 799.916904910334 1.56042395643484 1.26010101277496 1.66006031250858
MuFy 7.4725299176704 2.3349952013e− 6 0 0.16540684451041 799.916756098369 1.26010101891705 2 1.66006031250854
DE 7.4725299145657 2.3349939620e− 6 0 0.16540685285540 799.916226321289 1.25355443785422 1.26010097476864 1.66006031250904

LWF RMSE
nDEBCO 7.4959886958941 2.521106914e− 8 6.639449425037e− 7 0.39682287017287 215.830270566583 1.02330190887845 0.73074368689296 1.04305570761706
ABC-DE 7.4960286802837 2.653923742e− 8 8.907746359238e− 13 0.4333458228756 217.913748368923 1.02534657704230 0.74456758989071 1.04305570772810
SCA 7.4957661995176 2.659192205e− 8 8.232039822662e− 13 0.39454833322729 216.897490088556 1.02564988557814 0.73607754470894 1.04309020913122
ELBA 7.4959887051076 2.521109138e− 8 6.639482560240e− 13 0.39682276946639 215.830213892502 1.02330192368150 0.73074381495205 1.0430557076175
MuFy 7.4956862809168 1.006091922e− 12 2.808951162275e− 8 0.39226484844562 217.748155109734 0.74112187001662 1.02807970202239 1.04319782594629
DE 7.4900707998570 1.165930303e− 10 9.996558993601e− 8 0.33338721842964 251.766770724386 0.88712849427536 1.08687469289849 1.06501738711573
Energy Conversion and Management 248 (2021) 114667
 O. Hachana et al.

Table A6
Comparison results for CLS 220P PV module parameters by using ODM.
Classic RMSE

Algorithm Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 2)

nDEBCO 6.05229061843 0.04686549804853 0.7327597769154 457.8991470106 1.055368637312 1.131331730115

ABC-DE 6.05229061325 0.04686551333446 0.7327597706575 457.8992374976 1.055368655682 1.131331730115
DE 6.05229035972 0.04686580898898 0.7327595777737 457.9027039938 1.055369006781 1.131331730182
ELBA 6.05229061302 0.04686550483773 0.7327597755010 457.8992229778 1.055368645474 1.131331730115
SCA 6.05229061448 0.04686550131607 0.7327597765383 457.8991950161 1.055368641283 1.131331730115
MuFy 6.05229111322 0.04686422172143 0.7327604023350 457.8942741319 1.055367103189 1.131331730559
MPSO 6.05229060938 0.04686558525199 0.7327597318089 457.8994985441 1.055368741669 1.131331730116

LWF RMSE
nDEBCO 6.04774038461 0.086122202893 0.7150555983253 566.0599038235 1.090608310493 0.646612803152
ABC-DE 6.04774038461 0.086122202893 0.7150555983253 566.0599038235 1.090608310493 0.646612803152
DE 6.04774038466 0.0861222147018 0.7150555936356 566.0599224983 1.090608318596 0.646612803152
ELBA 6.04774038865 0.0861221933625 0.7150555999816 566.0598214528 1.090608303806 0.646612803152
SCA 6.04774038543 0.0861222044791 0.7150555981047 566.0599115334 1.090608311555 0.646612803152
MuFy 6.02698806612 0.1821320491661 0.6966149181162 2000 1.137296020736 0.933703810311

24
Table A7
Comparison results for CLS 220P PV module parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (A) Isat2 (A) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 3)

nDEBCO 6.04718448389620 4.485305129e− 8 1.116116773836e− 27 0.90433428177612 689.238712926145 1.08447772635152 0.31232632551232 4.33929841502380

ABC-DE 6.04718456949302 4.549320819e− 8 1.116116773836e− 27 0.90435266191998 647.583557565563 1.08432925386511 0.31289836514408 4.56637429010392
SCA 6.04322891115462 1.434056878e− 7 4.412259478148e− 21 0.86846220441571 831.200697350369 1.16068297774935 0.41192213652504 5.30966285331865
ELBA 6.04018650341331 1.694478478e− 6 4.803576362513e− 12 0.79405065758865 1109.70989716165 1.39707755224094 0.71942767970357 8.32153987299546
MuFy 6.05229061797400 0 4.686550202596e− 8 0.73275977632793 457.899163921103 1.98799752429577 1.05536864213318 11.3133173011532
DE 6.05229061639470 0 4.686550450650e− 8 0.73275977505138 457.899194765589 1.81425022514784 1.05536864509042 11.3133173011528

LWF RMSE
nDEBCO 6.05575184192986 1.02908633e− 15 9.491080953498e− 10 1.62931009452118 598.588611795364 0.65622088902742 1.00118070480448 3.607454661538
ABC-DE 6.04774038548066 2.17577742e− 20 8.612217198925e− 8 0.71505561065861 566.059893459364 1.99642359113270 1.09060828893191 6.46612803152360
SCA 6.04774038482085 1.22077563e− 19 8.612220288236e− 8 0.71505559969688 566.059923807192 1.09060831050385 1.97849265551034 6.46612803152361
ELBA 6.04774038381247 1.19307778e− 17 8.612219361938e− 8 0.71505560312280 566.060076772386 1.81355138844689 1.09060830381091 6.46612803154574
MuFy 6.04774038602233 0 8.612218885730e− 8 0.71505560385596 566.059885499121 1.09060830071202 1.32756024176715 6.46612803152362
DE 6.04774038376274 0 8.612220004960e− 8 0.71505560051646 566.059933348391 1.09060830845129 1.93520157860711 6.46612803152362
Energy Conversion and Management 248 (2021) 114667
 O. Hachana et al.

Table A8
Comparison results for CLS 220P PV string parameters by using ODM.
Classic RMSE

Algorithm Iph (A) Isat (µA) Rs (Ω) Rsh (Ω) n RMSE (e− 2)

nDEBCO 7.34644711607017 0.815472169750178 1.77967955421492 589.531798773296 1.21505514715082 1.034738476694

ABC-DE 7.34644711607017 0.815472169750178 1.77967955421492 589.531798773296 1.21505514715082 1.034738476694
DE 7.34644712527773 0.815471798068358 1.77967958817106 589.531700416406 1.21505511282384 1.034738476694
ELBA 7.34644712315134 0.815471894175819 1.77967958668144 589.531726044666 1.21505512181730 1.034738476694
SCA 7.34644712266607 0.815471806257314 1.77967959200452 589.531725280597 1.21505511359139 1.034738476694
MuFy 7.34644712685364 0.815471786813402 1.77967959403847 589.531695733892 1.21505511193419 1.034738476694
MPSO 7.34644712315134 0.815471894175819 1.77967958668144 589.531726044666 1.21505512181730 1.034738476694

LWF RMSE
nDEBCO 7.34571018141652 0.836698275455573 1.77824696575604 596.949272572821 1.21700219900005 0.887346444842353
ABC-DE 7.34571019652292 0.836697547973423 1.77824703968676 596.949119824550 1.21700213317687 0.887346444842358
DE 7.34571018556036 0.836698231655099 1.77824697055258 596.949240564897 1.21700219502876 0.887346444842392
ELBA 7.34571018606339 0.836698206894289 1.77824697012827 596.949246071618 1.21700219266198 0.887346444842384
SCA 7.34571018472942 0.836698230333011 1.77824696790735 596.949239374946 1.21700219486428 0.887346444842378
MuFy 7.34571018440691 0.836698281677503 1.77824696292549 596.949243932547 1.21700219951554 0.887346444842386

25
Table A9
Comparison results for CLS 220P PV string parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (A) Isat2 (A) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 2)

nDEBCO 7.34644712774605 5.051856194e− 19 8.1547169702842e− 7 1.77967960243530 589.531686690323 1.61092371690793 1.21505510351574 1.03473847669449

ABC-DE 7.34644712774605 5.051856194e− 19 8.1547169702842e− 7 1.77967960243530 589.531686690323 1.61092371690793 1.21505510351574 1.03473847669451
SCA 7.34644712414818 2.402996906e− 7 5.7517213088903e− 7 1.77967959130050 589.531726447263 1.21505519253037 1.21505508259319 1.03473847669457
ELBA 7.34644712796018 3.215910094e− 18 8.1547176114374e− 7 1.77967959393190 589.531694812180 1.42313625193420 1.21505510941169 1.03473847669456
MuFy 7.34644561914357 8.155035553e− 7 0 1.77968193357163 589.522192473692 1.21505826870100 2 1.03473862358872
DE 7.34644712391372 8.154717749e− 7 0 1.77967959525693 589.531712365058 1.21505511073702 1.16306516931750 1.03473847669454

LWF RMSE
nDEBCO 7.34601219705669 8.7750927346e− 9 8.0454484815835e− 7 1.78672864221128 594.639450494666 1.54200656885403 1.21444645030952 8.87156674403631
ABC-DE 7.34601219482259 8.5609273460e− 9 8.0454535263510e− 7 1.78672852987121 594.639460266314 1.54383475994734 1.21452217362343 8.87156674403631
SCA 7.34601219472257 8.7749152047e− 9 8.0454543558781e− 7 1.78672847813464 594.639462744727 1.21444649759572 1.54200640896852 8.87156674403631
ELBA 7.34601219699600 8.7748121085e− 9 8.0454493582227e− 7 1.78672858771765 594.639431534811 1.54200407041109 1.21444645479591 8.87156674403631
MuFy 7.34601219354462 8.7748592852e− 9 8.0454562840635e− 7 1.78672843741014 594.639472083091 1.54200608611630 1.21444651427991 8.87156674403554
DE 7.34601230041848 8.7770245222e− 9 8.0454613332985e− 7 1.78672907700173 594.639107845116 1.54201611671047 1.21444662555875 8.87156674440304
Energy Conversion and Management 248 (2021) 114667
 O. Hachana et al.

Table A10
Comparison results for Bifacial CEA INES back PV module parameters by using ODM.
Classic RMSE

Algorithm Iph (A) Isat (A) Rs (Ω) Rsh (Ω) n RMSE (e− 2)

nDEBCO 1.6707982857647 5.47162701191e− 16 1.48888310205531 849.785999340657 0.676592723128080 0.768726123835313

ABC-DE 1.6707982874498 5.47163033469e− 16 1.48888307173701 849.785946500774 0.676592734624707 0.768726123835323
DE 1.6683210244207 1.06673870211e− 14 1.37799609475657 979.800202414088 0.737982679907913 0.839592608358456
ELBA 1.6711411984726 5.28147074221e− 16 1.48982252241051 840.378749092543 0.675925152936797 0.768726123835336
SCA 1.6707982864831 5.47162450981e− 16 1.48888312100972 849.785978420057 0.676592714503034 0.768849059538067

LWF RMSE
nDEBCO 1.6723930915661 9.91114048653e− 18 1.65258009845092 792.172089975056 0.60846879226267 0.620305304806038
ABC-DE 1.6723930915661 9.91114048653e− 18 1.65258009845092 792.172089975056 0.60846879226267 0.620305304806038
ELBA 1.6723922718004 9.92981473538e− 18 1.65253065523988 792.360765768508 0.60849735994315 0.627782572018998
SCA 1.6713891573312 4.67555553663e− 17 1.59752379179015 820.012391733074 0.63315512332767 0.620309420283874

26
Table A11
Comparison results for Bifacial CEA INES front PV module parameters by using ODM.
Classic RMSE

Algorithm Iph (A) Isat (A) Rs (Ω) Rsh (Ω) n RMSE (e− 2)

nDEBCO 1.83314813297892 8.4134859792e− 13 1.24808024656751 3493.13812113599 0.851063240509298 0.802996313821798

ABC-DE 1.83314813475920 8.4134610602e− 13 1.24808043011618 3493.13658836514 0.851063152396848 0.802996313821935
DE 1.83252138554262 1.1168841512e− 12 1.23247346231574 3980.95117989403 0.859577932489928 0.804315390267749
ELBA 1.83314813526120 8.4134715736e− 13 1.24808033249718 3493.13623950430 0.851063189430492 0.802996313821840
SCA 1.83314813327779 8.4134854550e− 13 1.24808025746002 3493.13805495872 0.851063238701088 0.802996313821801

LWF RMSE
nDEBCO 1.83392636212983 2.1993546829e− 13 1.33231224039234 2956.87766073906 0.812993004811143 0.739515498392151
ABC-DE 1.83392636212983 2.1993546829e− 13 1.33231224039234 2956.87766073906 0.812993004811143 0.739515498392151
ELBA 1.83392635985305 2.1993644866e− 13 1.33231200768053 2956.87939215845 0.812993125995067 0.739515498392105
SCA 1.83392635775180 2.1993615210e− 13 1.33231208162720 2956.87981057109 0.812993089257240 0.739515498392082
Energy Conversion and Management 248 (2021) 114667
 O. Hachana et al.

Table A12
Comparison results for bifacial CEA INES front PV module parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (A) Isat2 (A) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 3)

nDEBCO 1.83314813213652 8.413487090e− 13 5.773102786747e− 24 1.24808024716779 3493.13871752965 0.85106324448802 1.14451873132616 8.02996313821784

ABC-DE 1.833239347545213 8.453943850e− 13 5.834736237484e− 24 1.27349348209123 3484.23820343718 0.86349348322134 1.14634398732001 8.02996313821789
SCA 1.83314813274708 8.362796364e− 13 5.068916271274e− 15 1.24808025632188 3493.13816333471 0.85106323937487 0.85106317045053 8.02996313821795
ELBA 1.83314813194634 8.413485615e− 13 2.005982086662e− 21 1.24808025704942 3493.13866308714 0.85106323924783 1.99487049578062 8.02996313821800
MuFy 1.83314813369122 8.413486677e− 13 0 1.24808025117209 3493.13777556547 0.85106324305504 0.51654541274792 8.02996313821796
DE 1.83314813252872 0 8.413487357810e− 13 1.24808024260036 3493.13862669400 1.70666699170594 0.85106324533211 8.02996313821798

LWF RMSE
nDEBCO 1.83745982111884 2.332794037e− 20 1.503400994256e− 32 3.23972902259297 2111.67722076268 0.55169709103932 0.36428190151363 6.07522826652045
ABC-DE 1.837577364292411 3.392773652e− 20 1.614388723641e− 32 3.24383766238264 2117.27633434345 0.56337472610002 0.36334834623450 6.07522826652061
SCA 1.83690623433433 3.316066785e− 26 1.134364045261e− 18 2.93680901262941 2202.82391995185 0.45375170834465 0.59939817023488 6.14405644411450
ELBA 1.83621956351245 4.005417786e− 20 7.689415594495e− 17 2.56771904738585 2330.21031476017 0.59477956918851 0.66230666590989 6.36239122518336
MuFy 1.83392635787096 2.199360919e− 13 0 1.33231208887357 2956.87966132134 0.81299308178431 1.99471557841035 7.39515498392078
DE 1.83555407339543 7.539132922e− 16 1.497652065894e− 16 2.29831130138730 2504.34106433892 0.70162773783676 0.72967003485977 6.57285121109731

27
Table A13
Comparison results for Bifacial CEA INES back PV module parameters by using DDM.
Classic RMSE

Algorithm Iph (A) Isat1 (A) Isat2 (A) Rs (Ω) Rsh (Ω) n1 n2 RMSE (e− 3)

nDEBCO 1.67079828697828 4.504943090e− 22 5.471622396297e− 16 1.48888313109465 849.785946427481 1.76878587468278 0.67659270721803 7.68726123835202

ABC-DE 1.6707982947283 4.525489384e− 22 6.471625839472e− 16 1.49888275423443 849.945083743625 1.76740023814355 0.6754746300921 7.68726123835297
SCA 1.67079828666613 5.324153099e− 16 1.474705833812e− 17 1.48888311755021 849.785968520443 0.67659271189432 0.67659270029634 7.68726123835308
ELBA 1.67079828969041 5.125808242e− 22 5.471621894788e− 16 1.48888312475316 849.785841366274 1.55521467971786 0.67659270549545 7.68726123835312
MuFy 1.67070768848444 7.418945603e− 16 0 1.47745266202749 857.384507973720 0.68241095760924 0.53591872475040 7.69544700898595
DE 1.67001493549105 0 2.112553342996e− 15 1.44168774847833 889.514453157291 2 0.70321223472200 7.83926001533706

LWF RMSE
nDEBCO 1.67895928719149 2.146696915e− 33 2.923240783500e− 41 3.65037876871826 717.214464722378 0.33076069163528 0.28604643663110 3.30099082719454
ABC-DE 1.67696822710646 8.427908163e− 30 6.474408133141e− 27 3.00000000000000 732.198199683700 0.39769878270632 0.40665494245408 3.67235182736691
SCA 1.67682397251411 1.075670876e− 25 9.460687257678e− 27 2.99992774740671 738.016649441648 0.42724758187924 0.44418413180499 3.76345798977186
ELBA 1.67239309310023 9.911120095e− 18 3.253221439947e− 19 1.65258017335301 792.172067042408 0.60846876090299 1.22399248009052 6.20305304742917
MuFy 1.67239309330957 0 9.911116062862e− 18 1.65258018464848 792.172049744468 2 0.60846875459367 6.20305304806027
DE 1.67239309310253 0 9.911113360717e− 18 1.65258020107207 792.172040733417 2 0.60846875050963 6.20305304806030
Energy Conversion and Management 248 (2021) 114667
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

References [29] El-Fergany AA. Parameter identification of PV model using improved slime mould
optimizer and Lambert W-function. Energy Rep 2021;7:875–87.
[30] Messaoud RB. Extraction of uncertain parameters of single and double diode model
[1] Fasihi M, Weiss R, Savolainen J, Breyer C. Global potential of green ammonia
of a photovoltaic panel using Salp Swarm algorithm. Measure 2020;154:107446.
based on hybrid PV-wind power plants. Appl Energy 2021;294:116170. https://
https://doi.org/10.1016/j.measurement.2019.107446.
doi.org/10.1016/j.apenergy.2020.116170.
[31] Agwa AM, El-Fergany AA, Maksoud HA. Electrical characterization of photovoltaic
[2] Mutezo G, Mulopo J. A review of Africa’s transition from fossil fuels to renewable
modules using farmland fertility optimizer. Energy Convers Manage 2020;217:
energy circular economy principles. Renew Sustain Energy Rev 2021;137:110609.
112990. https://doi.org/10.1016/j.enconman.2020.112990.
[3] Hachana O, Hemsas K E, Tina GM, Ventura C. Comparison of different
[32] Rezk H, Babu TS, Al-Dhaifallah M, Ziedan HA. A robust parameter estimation
metaheuristic algorithms for parameter identification of photovoltaic cell/module.
approach based on stochastic fractal search optimization algorithm applied to solar
J Renew Sustain Energy 2013;053122.
PV parameters. Energy Rep 2021;7:620–40.
[4] Wang M, Peng J, Luo Y, Shen Z, Yang H. Comparison of different simplistic
[33] Yang Bo, Wang J, Zhang X, Yu T, Yao W, Shu H, et al. Comprehensive overview of
prediction models for forecasting PV power output: assessment with experimental
meta-heuristic algorithm applications on PV cell parameter identification. Energy
measurements. Energy 2021;224:120162. https://doi.org/10.1016/j.
Convers Manage 2020;208:112595. https://doi.org/10.1016/j.
energy.2021.120162.
enconman.2020.112595.
[5] Hachana O, Tina GM, Hemsas KE. PV array fault Diagnostic Technique for BIPV
[34] Ram JP, Babu TS, Dragicevic T, Rajasekar N. A new hybrid bee pollinator flower
systems. Energy Build 2016;126:263–74.
pollination algorithm for solar PV parameter estimation. Energy Convers Manage
[6] Tina GM, Ventura C. Evaluation and Validation of an Electrical Model of
2017;135:463–76.
Photovoltaic Module Based on Manufacturer Measurement. In Hakansson A, Höjer
[35] Li S, Gong W, Yan X, Hu C, Bai D, Wang L. Parameter estimation of photovoltaic
M, Howlett R, Jain L (eds) Sustainability in Energy and Buildings. Smart
models with memetic adaptive differential evolution. Sol Energy 2019;190:
Innovation, Systems and Technologies, Springer, Heidelberg 2013:22.
465–74.
[7] Yousri D, Fathy A, Rezk H, Babu TS, Berber MR. A reliable approach for modeling
[36] Nunes HGG, Pombo JAN, Mariano SJPS, Calado MRA, Felippe de Souza JAM.
the photovoltaic system under partial shading conditions using three diode model
A new high performance method for determining the parameters of PV cells and
and hybrid marine predators-slime mould algorithm. Energy Convers Manage
modules based on guaranteed convergence particle swarm optimization. Appl
2021;243:114269. https://doi.org/10.1016/j.enconman.2021.114269.
Energy 2018;211:774–91.
[8] Abdulrazzaq AK, Bognár G, Plesz B. Evaluation of different methods for solar cells/
[37] Nunes HGG, Silva PNC, Pombo JAN, Mariano SJPS, Calado MRA. Multiswarm
modules parameters extraction. Sol Energy 2020;196:183–195.
spiral leader particle swarm optimisation algorithm for PV parameter
[9] Ridha HM. Parameters extraction of single and double diodes photovoltaic models
identification. Energy Convers Manage 2020;225:113388. https://doi.org/
using Marine Predators Algorithm and Lambert W function. Sol Energy 2020;209:
10.1016/j.enconman.2020.113388.
674–93.
[38] Gao S, Wang K, Tao S, Jin T, Dai H, Cheng J. A state-of-the-art differential
[10] Li S, Gong W, Gu Q. A comprehensive survey on meta-heuristic algorithms for
evolution algorithm for parameter estimation of solar photovoltaic models. Energy
parameter extraction of photovoltaic models. Renew Sustain Energy Rev 2021;141:
Convers Manage 2021;230:113784. https://doi.org/10.1016/j.
110828. https://doi.org/10.1016/j.rser.2021.110828.
enconman.2020.113784.
[11] Yousri D, Thanikanti SB, Allam D, Ramachandaramurthy VK, Eteiba MB. Fractional
[39] Long W, Wu T, Xu M, Tang M, Cai S. Parameters identification of photovoltaic
chaotic ensemble particle swarm optimizer for identifying the single, double, and
models by using an enhanced adaptive butterfly optimization algorithm. Energy
three diode photovoltaic models’ parameters. Energy 2020;195:116979. https://
2021;229:120750. https://doi.org/10.1016/j.energy.2021.120750.
doi.org/10.1016/j.energy.2020.116979.
[40] Chen Xu, Xu B, Mei C, Ding Y, Li K. Teaching-learning- based artificial bee colony
[12] Li S, Gong W, Yan X, Hu C, Bai D, Wang L, et al. Parameter extraction of
for solar photovoltaic parameter estimation. Appl Energy 2018;212:1578–88.
photovoltaic models using an improved teaching-learning-based optimization.
[41] Beigi AM, Maroosi A. Parameter identification for solar cells and module using a
Energy Convers Manage 2019;186:293–305.
hybrid firefly and pattern search algorithm. Sol Energy 2018;171:435–46.
[13] Abd Elaziz M, Oliva D. Parameter estimation of solar cells diode models by an
[42] Yu K, Liang JJ, Qu BY, Cheng Z, Wang H. Multiple learning backtracking search
improved opposition-based whale optimization algorithm. Energy Convers Manage
algorithm for estimating parameters of photovoltaic models. Appl Energy 2018;
2018;171:1843–59.
226:408–22.
[14] Gao X, Cui Y, Hu J, Xu G, Wang Z, Qu J, et al. Parameter extraction of solar cell
[43] Wu Z, Yu D, Kang X. Parameter identification of photovoltaic cell model based on
models using improved shuffled complex evolution algorithm. Energy Convers
improved ant lion optimizer. Energy Convers Manage 2017;151:107–15.
Manage 2018;157:460–79.
[44] Yu K, Liang JJ, Qu BY, Chen Xu, Wang H. Parameters identification of photovoltaic
[15] Zhang Y, Huang C, Jin Z. Backtracking search algorithm with reusing differential
models using an improved jaya optimization algorithm. Energy Convers Manage
vectors for parameter identification of photovoltaic models. Energy Convers
2017;150:742–53.
Manage 2020;223:113266. https://doi.org/10.1016/j.enconman.2020.113266.
[45] Hu Z, Gong W, Li S. Reinforcement learning-based differential evolution for
[16] Li S, Gu Q, Gong W, Ning B. An enhanced adaptive differential evolution algorithm
parameters extraction of photovoltaic models. Energy Rep 2021;7:916–28.
for parameter extraction of photovoltaic models. Energy Convers Manage 2020;
[46] Jian X, Zhu Y. Parameters identification of photovoltaic models using modified
205:112443. https://doi.org/10.1016/j.enconman.2019.112443.
Rao-1 optimization algorithm. Optik 2021;231:166439. https://doi.org/10.1016/j.
[17] Yu K, Qu B, Yue C, Ge S, Chen Xu, Liang J. A performance guided JAYA algorithm
ijleo.2021.166439.
for parameters identification of photovoltaic cell and module. Appl Energy 2019;
[47] Chenouard R, El-Sehiemy RA. An interval branch and bound global optimization
237:241–57.
algorithm for parameter estimation of three photovoltaic models. Energy Convers
[18] Bana S, Saini RP. Identification of unknown parameters of a single diode
Manage 2020;205:112400. https://doi.org/10.1016/j.enconman.2019.112400.
photovoltaic model using particle swarm optimization with binary constraints.
[48] Cheung NJ, Ding XM, Shen HB. A nonhomogeneous cuckoo search algorithm based
Renew Energy 2017;101:1299–310.
on quantum mechanism for real parameter optimization. IEEE Trans Cybern 2017;
[19] Franco RAP, Filho GL, Vieira FHT. Firefly algorithm applied to the estimation of
47:391–402.
the parameters of a photovoltaic panel model. Adv. Nat. Inspir Comput Appl
[49] Lin X, Wu Y. Parameters identification of photovoltaic models using niche-based
Springer 2019:101–34.
particle swarm optimization in parallel computing architecture. Energy 2020;196:
[20] Subudhi B, Pradhan R. Bacterial foraging optimization approach to parameter
117054. https://doi.org/10.1016/j.energy.2020.117054.
extraction of a photovoltaic module. IEEE Trans Sustain Energy 2018;9(1):381–9.
[50] Liang J, Qiao K, Yuan M, Yu K, Qu B, Ge S, et al. Evolutionary multi-task
[21] Long W, Cai S, Jiao J, Xu M, Wu T. A new hybrid algorithm based on grey wolf
optimization for parameters extraction of photovoltaic models. Energy Convers
optimizer and cuckoo search for parameter extraction of solar photovoltaic models.
Manage 2020;207:112509. https://doi.org/10.1016/j.enconman.2020.112509.
Energy Convers Manage 2020;203:112243. https://doi.org/10.1016/j.
[51] Storn R, Price K. Differential Evolution - A Simple and Efficient Heuristic for Global
enconman.2019.112243.
Optimization over Continuous Spaces. J Global Optimiz 1997;11:341–59.
[22] Abdel-Basset M, et al. An efficient teaching-learning-based optimization algorithm
[52] Dorigo M, Birattari M, Stutzle T. Ant colony optimization. IEEE Comp Intell
for parameters identification of photovoltaic models: analysis and validations.
Magazine 2006;1(4):28–39.
Energy Convers Manage 2021;227:113614.
[53] Rahkar Farshi T. Battle royale optimization algorithm. Neural Comp Appl 2021;33
[23] –>Fathy A, Rezk H. Parameter estimation of photovoltaic system using imperialist
(4):1139–57.
competitive algorithm. Renew Energy 2017;111:307–20.
[54] Aoufi B, Hachana O. A New Mutated-Firefly Algorithm for Parameters Extraction of
[24] Guo L, Meng Z, Sun Y, Wang L. Parameter identification and sensitivity analysis of
Solar Photovoltaic Cell Model. In Hatti M. (eds) Artificial Intelligence and
solar cell models with cat swarm optimization algorithm. Energy Convers Manage
Renewables Towards an Energy Transition, ICAIRES. Lecture Notes in Networks
2016;108:520–8.
and Systems, Springer 2021;174. (https://doi.org/10.1007/978-3-030-63846-7_
[25] Ali EE, El-Hameed MA, El-Fergany AA, El-Arini MM. Parameter extraction of
52).
photovoltaic generating units using multi-verse optimizer. Sustain Energy Tech
[55] Easwarakhanthan T, BOTTIN J, bouhouch I, Boutrit C. Nonlinear minimization
Assess 2016;17:68–76.
algorithm for determining the solar cell parameters with microcomputers. Sol
[26] Allam D, Yousri DA, Eteiba MB. Parameters extraction of the three-diode model for
Energy 1986;4(1):1–12.
the multi-crystalline solar cell/ module using moth-flame optimization algorithm.
[56] Tina GM, Bontempo Scavo F, Merlo L, Bizzarri F. Comparative analysis of
Energy Convers Manage 2016;123:535–48.
monofacial and bifacial photovoltaic modules for floating power plants. Appl
[27] Qais MH, Hasanien HM, Alghuwainem S. Identification of electrical parameters for
Energy 2021;281:116084. https://doi.org/10.1016/j.apenergy.2020.116084.
three-diode photovoltaic model using analytical and sunflower optimization
[57] Bhang BG, Lee W, Kim GG, Choi JH, Park SY, Ahn H-K. Power performance of
algorithm. Appl Energy 2019;250:109–17.
bifacial c-Si PV modules with different shading ratios. IEEE J Photovolt 2019;9(5):
[28] Deotti LMP, Pereira JLR, da Silva IC. Parameter extraction of photovoltaic models
1413–20.
using an enhanced Lévy flight bat algorithm. Energy Convers Manage 2020;221:
113114.

28
O. Hachana et al. Energy Conversion and Management 248 (2021) 114667

[58] Karaboga D, Gorkemli B, Ozturk C, Karaboga N. A comprehensive survey: artificial [66] Diab AAZ, Sultan HM, Do TD, Kamel OM, Mossa MA. Coyote optimization
bee colony (ABC) algorithm and applications. Artif Intell Rev 2014;42(1):21–57. algorithm for parameters estimation of various models of solar cells and PV
[59] Simon D. Biogeography-based optimization. IEEE Trans Evol Comput 2008;12(6): modules. IEEE Access 2020;8:111102–40.
702–13. [67] Wang J, Du P, Niu T, Yang W. A novel hybrid system based on a new proposed
[60] Ćalasan M, Abdel Aleem SHE, Zobaa AF. On the root mean square error (RMSE) algorithm—Multi-Objective whale optimization algorithm for wind speed
calculation for parameter estimation of photovoltaic models: a novel exact forecasting. Appl energy 2017;208:344–60.
analytical solution based on Lambert W function. Energy Convers Manage 2020; [68] Elazab OS, Hasanien HM, Elgendy MA, Abdeen AM. Whale optimisation algorithm
210:112716. https://doi.org/10.1016/j.enconman.2020.112716. for photovoltaic model identification. J Eng 2017;2017(13):1906–11.
[61] Lekouaghet B, Boukabou A, Boubakir C. Estimation of the photovoltaic cells/ [69] Houssein EH, Zaki GN, Diab AAZ, Younis EMG. An efficient Manta Ray Foraging
modules parameters using an improved Rao-based chaotic optimization technique. Optimization algorithm for parameter extraction of three-diode photovoltaic
Energy Convers Manage 2021;229:113722. https://doi.org/10.1016/j. model. Comput Electr Eng 2021;94:107304. https://doi.org/10.1016/j.
enconman.2020.113722. compeleceng.2021.107304.
[62] Yousri D, Allam D, Eteiba MB, Suganthan PN. Static and dynamic photovoltaic [70] Laudani A, Riganti Fulginei F, Salvini A. High performing extraction procedure for
models’ parameters identification using chaotic heterogeneous comprehensive the one-diode model of a photovoltaic panel from experimental i–v curves by using
learning particle swarm optimizer variants. Energy Convers Manage 2019;182: reduced forms. Sol Energy 2014;103:316–26.
546–63. [71] Cardenas AA, Carrasco M, Mancilla-David F, Street A, Cardenas R. Experimental
[63] Oliva D, Abd El Aziz M, Ella Hassanien A. Parameter estimation of photovoltaic parameter extraction in the single-diode photovoltaic model via a reduced-space
cells using an improved chaotic whale optimization algorithm. Appl Energy 2017; search. IEEE Trans Ind Electron 2017;64(2):1468–76.
200:141–54. [72] Toledo FJ, Blanes JM, Galiano V. Two-step linear least-squares method for
[64] Liang J, Ge S, Qu B, Yu K, Liu F, Yang H, et al. Classified perturbation mutation photovoltaic single-diode model parameters extraction. IEEE Trans Ind Electron
based particle swarm optimization algorithm for parameters extraction of 2018;65(8):6301–8.
photovoltaic models. Energy Convers Manage 2020;203:112138. https://doi.org/ [73] Panchal AK. A per-unit-single-diode-model parameter extraction algorithm: A
10.1016/j.enconman.2019.112138. high-quality solution without reduced-dimensions search. Sol Energy 2020;207:
[65] Kler D, Goswami Y, Rana KPS, Kumar V. A novel approach to parameter estimation 1070–7.
of photovoltaic systems using hybridized optimizer. Energy Convers Manage 2019;
187:486–511.

29