A combination of Newton-Raphson method and heuristics algorithms for parameter estimation in photovoltaic modules.

Parameters extraction is instrumental to standard PV cells design. Reports indicates that heuristic algorithms are the most effective methods for accurately determinining the values of parameters. However, local concentration is against recent heuristic methods, and they are inhibited producing optimal results. This paper seeks to show that combining the heuristics algorithms with the Newton Raphson method can considerably increased the accuracy of results. An inspired artifact technique from the drone squadron simulation from control center is proposed for the extraction of the best constitutive parameters. This study equally provides clarifications on the approaches recently reported and proposed to build objective function. Furthermore, comparative evaluation of the current ten best heuristics algorithms that are published in the PV estimation domain is also undertaken. Moreover, this study investigates the convergence of algorithms when points of the number of current-voltage characteristics are varied. The results from this study highlight the differences between the two formulation, and it shows the best formulation accuracy. The results obtained from seven study cases that are considered in this present study, with the combined Newton Raphson performance method and Drone Squadron optimisation, were employed to extract precise PV module parameters.The study of the numbers of points reveals that the algorithm converges and is more precise when the numbers of points of the I-V characteristic are reduced. However, if these points are minimal, the algorithm will be hindered from returning optimal results.

Introduction

Electricity production from solar energy has undergone continnous changed in the 21st century [1]. The increase in solar energy production is attributed to advanced field research and to the free availability of solar energy in the day time in many countries in Africa and elsewhere [2]. Therefore, field development is important for an accurate design of cells and PV modules that can predict the power delivery from the PV system [3, 4].

PV modeling takes into account, well-known parameters such as the ideality factor of diode, the current saturation, photocurrent generation, series resistance and shunt resistance [5]. The drawn PV cell equivalent circuit and mathematic model enable the understanding of information provided from a basic technical data sheet, to estimate unknown parameters and generate maximum power. These data sheets present the open circuit voltage (Voc), the short-circuit current (Isc), voltages at points of maximum power (Vmp), the current at the point of maximum power (Imp) and the power at the point of maximum power (Pmp). Different techniques (analytic, numeric, metaheuristic and hybrids) have been used in the literatures for different parameters extractions. Metaheuristic algorithms have been considered as the best method [4]. The modelling of PV uses sets of nonlinear equations to find the unknown values of PV cell. A detailed literature review on extraction techniques is summarized by [7] which shows that PV equations can be modelled using three different approaches, namely : analytic, numeric and progressive computational techniques [7].

Precise results are gotten from the prime method with little computation time. Analytic method are very simple and more often, just an iteration is necessary to obtain the desired result. A develop method of approximation, which serves as explicit model capitalizing for VI characteristics and estimation of single PV model of diode parameters has been proposed by [8]. Others methods like Lambert's W-function [9], the expansion of Taylor series [10], set of approximate explicit equations [11] and curve-fitting approaches [12] have been considered in the litterature. Some methods enable 5 parameters estimation and others extract just series resistance and shunt resistance [11]. Major weakness of this analytical method is that they suits only the condition of standard test and collapse with conditions variations. Numeric methods with techniques of curve fitting are best than analytic and various considered algorithms of these methods enable acquisition of results accuracy by the assessment of all PV-IV curves points using the algorithm of Levenberg to get solar cell parameters [13]. Method of Newton Raphson implies large computing times and memories; it fails for inaccurate initial guess convergence even though the method to delay the convergence is efficient [14]. PV parameters estimation include constraints optimization was carried out using metaheuristic. These algorithms are mainly nature inspired, following natural environmental processes. These algorithms are considered the best, because they give optimal results whatever the conditions. Numerous available algorithms in literatures portray different modelling techniques such as conventional GA and basic progressive computing techniques. Double PV parameters estimation for diode model was done using: improved the Optimization of Harris hawks (HHO) [15], improved adapted differential and evolution with repairing crossover rate (RcrIJADE) [16], genetic algorithm (GA) [17], algorithm of the enhanced vibration of particles system (EVPS) [6], artificial algorithm of bee swarm optimizations (ABSO) [18] and algorithm optimization of the improved JAYA (IJAYA) [19]. In addition we have the use of self-adaptive teaching-learning-based optimization (SATLBO) [20], improved optimization of whale algorithm (IWAO) [21] algorithm of the improved shuffled and complex evolutions (ISCE) [22], algorithm of hybrid pollination of flower (GOFPANM) [23] and grasshopper optimization algorithm (GOA) [24], the improved teaching and learning based optimization (ITLBO) [25], the algorithm of hybrid cuckoo search (HBCS) [26]. There is also improved algorithm of opposition-based cosine (ISCA) [27], the algorithm of hybrid reflective of trust-region (ABC-TRR) [28], teaching-learning artificial based bee colony photovoltaic estimation of parameters (TLABC) [29] and the algorithm of hybrid differentia-evolution with the optimization of whale (DE/WOA) [30]. Other algorithm used are algorithm of enhanced shuffled and complex evolution by opposition-based learning improvement (ESCE OBL) [31] and the advanced onlooker-ranking based and improved adaptive and differential evolution (ORcr-IJADE) [32], the optimization of cat swarm algorithm (CSO) [33], the algorithms of firefly hybrid search pattern (HFAPS) [34] and the improved mothflame (IMFO) [35]. The mimic adaptive and differential evolution (MADE) [36], algorithm of Coyote optimization (CAO) [37], algorithm of multiple learning and back-tracking search (MLBSA) [38], symbiotic-organisms-search (SOS) [39], successive algorithm discretization (SDA) [40], algorithm of hybrid adaptive and nelder-mead simplex (EHA-NMS) [41, 42], nonhomogeneous algorithm cuckoo search (NoCuSa) [43] and analytical-sunflower algorithm optimization (SFO) [44] were used. Guided JAYA performance (PGJAYA) [45] and improved adaptive and differential evolution algorithm (IADE) [46]. PSO was proven limited in areas of selecting initial parameters as well as high numbers of iteration. The modified PSO proposed to subdue these limitations: Heterogeneous-Chaotic, comprehensive and learning particle optimizer of swarm (C-HCLPSO) [47], the swarm intelligence collaborative (WDOWOAPSO), the enhanced optimisation leader particle of swarm (ELPSO) [48], mutation particle with swarm adaptive optimisation (MPSO) [49] and time variation acceleration coefficients of swarm particle optimization (TVACPSO) [50]. Recent studies investigated other aspects of PV parameters [51], authors investigated on the identification of PV parameters in operation conditions instead of standard. The determined shunt resistance and series resistance in working condition with a few point of IV characteristics around power maximum point. Other authors [52] investigate on influence of the degradation on PV parameters where they used an explicit method to identify single model of diode parameters [53] using current and voltage, temperature coefficient with three specific points include open circuit point, power maximum point, and short circuit point, to extract a solar cell's best parameter. The method enable PV performance under standard reporting conditions prediction. In [54], a new multispiral leader's particles swarm was used to boost PSO convergence while estimating PV parameters. In [55], Boosted Harris Howk's optimization was proposed to determine stable and accurate results [56]. Investigate the different methods used for the estimate of PV parameters, and then they proposed an exact based method on Lambert W-function.

Even with the improvement of these algorithms to date, there is no algorithm for answering all the problems of optimization; moreover, as expressed earlier in the theorem of "no-free-lunch", solving optimization problems with new algorithms are always welcome. That is why we proposed a new method, which is DSO in combination with the method of Newton Raphson, for PV parameters extraction. This algorithm is different from metaheuristics, which are nature inspired; the DSO algorithm is of human and scientific development.

The objective function used by many authors is reducing errors between current measurements and estimation. Studies carried out by researchers in 2019 have proposed method of checking best accuracy results from PV parameters extraction. The obtained results pointed out that some proposed methods found in literatures were not accurate [3]. Nevertheless, the given answer of these authors reveal some inaccuracies in two distinct formulations that had always been presented in literature as similar [57]. This current paper aims firstly to provide insights on the two reported current estimated (accuracy and approximation formulation), and then hybrid methods related to Newton Raphson and self-adapting algorithm is proposed. The DSO is a simulation technique of drone squadron from control station [58], that is used here for the extraction of the best PV modules and cells parameters. At the end, the study also investigates on the convergence of the algorithm when the number of points of the current voltage characteristic is varied.

This paper contributes to:

the clarification of current estimation of two formulations.

the two analogue formulations that are often used in literatures were introduced and compare with recently published ten algorithms.

the application of self-adapting algorithm in combination with the method of Newton-Raphson considered as first time applied method for best PV cells and PV modules parameters extraction.

the investigation on the convergence of the algorithms when the number of points of the current voltage characteristics are varied.

the tested algorithm of DSO hyper-heuristic on known sets of PV data (RTC France solar cell and PWP201) and results determined compared with literatures exhibiting the algorithm efficiency.

DSO application on the experimental module of Sharp ND-R250A5 PV data [59].

Other aspects of the research present at section 2 a solar energy cell general model, various formulations at section 3, the DSO models at section 4, different study cases at section five and conclusion in section 6.

General model

Many models are used for the representing of the electrical PV cell characteristics. Two models such as single and dual diode model were mainly used and presented in Figure 1 and Figure 2 respectively [15, 60].

Figure 1

Single diode electrical circuit from PV cell.

Figure 2

PV cell, dual diode electrical circuit.

Eq. (1) present a single diode current output model [61]:Where: is the number of cells connected in series; k is the constant of Boltzman [J/K], T the temperature [K] and q the charge of electron [C].

is an output current of cell [A], an output voltage of cell [V], a photoelectric current [A], a reverse diode saturation currents [μA], the ideality factor of diode, R the resistance series [Ω], the resistance shunt [Ω] and thermal voltage [V] given by the expression of Eq. (2).

For dual diode model, output current is expressed by.

Objective function

In other to implement heuristic algorithm to obtain a photovoltaic cell's intrinsic parameters, we should expressed the problem as optimisation problem. Some authors used root means square errors (RMSE) defined by [18, 34]:

is the objective function to minimize; the set of experimental number of I–V pairs; data set of current estimated; represents the vector of parameters to be estimated; represents single diode vector of parameters estimated for dual diode model.

Results from literatures and their implementation in this research reveals two formulations for current estimation: formulation of approximation and formulation of accuracy.

Approximation formulation

For output current complexity at Eq. (1), many of the authors [25, 33] have chosen to use the approximation formulation for current estimation described by Eq. (5):

For single model of diode,

The expression of the objective function estimating five parameters from single model of diode is given by:

For a dual model of diode,

For the double diode, the objective function is given by the expression:With and representing measured voltage and current respectively;

The major advantage of the approximation formulation is that it focuses directly on obtaining the least square error and parameters from the listed optimization algorithms. However, the results determined are not reliable; they are presented at section 5.

The best formulation

The estimated characteristic of I–V is very close to the experimental measurements, for the experimental use set (,) using , the estimated current by the formulation of a single and dual diode respectively and described by:

Non-linear Eqs. (9) and (10) can not be solve to determine a precise solution: which is a limit solving this equation. Lambert function (W) and digital methods are applied to overcome this limitation [59].

Eq. (9) is then rewritten as Eq. (11)

We determine by solving with Newton Raphson's method described in Figure 3.

Figure 3

Structural plan of the current calculated on the basis of the Newton-Raphson method.

And:

derivative from Eq. (11) which is expressed by:

is a total estimation of current output; the stops criterion in obtaining better precision value to be chosen .

Then comes set of which corresponds to measures of ( ,), a final objective formulation as function of single and dual diode models expressed by Eqs. (13) and (14) respectively

Optimization of Drone Squadron

Inspiration

Algorithms of Drones Squadron are influenced by drones navigation at controlled space or command center. It was originally developed by Melo and Banzhaf in 2018 [58]. Like many other flying devices, the DSO is made up of a squadron of drones from different teams and control center. The control center do maintain partial surveillances and information collected to develop new firmwares (Figure 4).

Figure 4

DSO cyclic process.

The drones must be sectioned in teams and size for a good update of drone firmware [62]. Drone's mission consist of searching to locate landscaped targets.

Command and control center

Considering parts of intelligent DSO that orders tasks to be performed through commands of drones from the control center is controlled.

Firmwares

The firmware plays the role, that of generating only by disturbance, new test coordinates (TC) [58]; The following relationships are used to randomly generate the firmware disturbance pattern:Where is the complete perturbation formula which returns trial coordinates, the solution point in the search space (the coordinate) and a function that returns the actual perturbation movement.

Where P is total perturbation related to test coordinates returns, Departure solution point at search space (coordinate) and Offset a returns function of the actual perturbation motion.

Drones movements

In calculating targets positions, drones used autonomous system to redirect information to the control center. Scalable techniques of optimization are applied to calculate targets positions.

The gaol is generating better position for each drone/team (). With the travelling of drones within spaces boundaries, Eq. (17) is applied as correction for each exterior reference point.Where:

N: drones number/team,

D: dimensions

UB: upper bounds

LB: lower bounds.

Firmware upgrade and update

Algorithms could possibly end with this step. For the assessment of the quality of parameters/team, two elements are evaluated from the control center: target classified algorithm value and level of violations generated algorithm [58]. Eq. (18) determines each team quality.

The drones can during landscape exploration, generate two coordinates or identical information. For this purpose, a stagnation in relation to Eq. (19) is established [62].Where:

: objective team's function

: best current coordinates

: objective function of the best current coordinates

: details of teams coordinates

: uniform random distribution within the interval of 1 and 0

: accepting solution probability

Melo in 2017 described the complete DSO algorithm [62] and the detailed description of all DSO steps determined by the relationship described by Melo and Banzhaf in 2018 [58].

Experiments carried out and results

This section presents various experiments results including the approximation formulation, the accurate formulation results that are presented in section 5.1 and Section 5.2 respectively, and the 10 recent algorithms published: EVPS [6], TVACPSO [50], IMFO [35], IJAYA [19], ITLBO [25], CAO [37], SOS [39], MLBSA [38], MADE [36], GAMS [3]. The experiment of these algorithms was made based on the well known set of RTC France solar cell, with various parameters presented at Table 1. Irradiation and temperature are 1000W/m2 and 33 °C respectively. The starting parameters of these algorithms were mentioned in each paper, which are set as follows:

Table 1

Parameters of commercialized silicon PV cells RTC France.

Parameters	Isc(A)	Voc(V)	Imp(A)	Vmp(V)	Ns	Ki for Isc (%/C)	T(°C)	G (W/m2)
Values	0.760	0.573	0.691	0.450	1	0.0350	33	1000

EVPS: number of agents or population (NP) = 50; maximum iterations (MaxIt) = 1000;

SOS: Np = 50; the maximum number of function evaluations (Max_FEs) = 50,000

COA: number of groups (NP) = 10; number of coyotes (Nc) = 10; MaxIt = 1000.

TVAPSO: Np = 1000; MaxIt = 100

IJAYA: NP = 20; Max_FES = 50,000;

MLBSA: NP = 50; Max_FEs = 50,000;

IMFO: Np = 100; Max_FEs = 50,000;

DONE: NP = 50; Max_FEs = 50,000;

ITLBO: Np = 50; Max_FEs = 50,000;

GAMS: MaxIt = 100; Max_FEs = 188.

The DSO is valid for six different cases, taking into account single and double diode model. Selected algorithms are executed on software Matlab 2017b, installed on a computer with Intel (R) Core (TM) i7-2670QM CPU @ 2.20HHz, RAM 8.0GB, in an operating system Windows10.

Results combination of 10 algorithms for approximation formulation

Table 2 shows best results from twenty tests from different independent algorithms. We observe the nearly similar results from 10 algorithms with best RSME lying between 9.860 ∗ 10-4. Results of determined parameters are more or less similar to the ones obtained by the different authors in their publication with the exception of TVACPSO [50], which applied best formulations.

Table 2

Best results of approximation formulation from ten algorithms.

Algorithms	Best parameters
	Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)	Best RMSE × 10−4
GAMS[3]	0.760775	0.323020	1.481183	0.036377	53.718524	9.860218
MADE [36]	0.760775	0.323020	1.481183	0.036377	53.718538	9.860218
ITLBO[25]	0.760775	0.323020	1.481183	0.036377	53.718523	9.860218
TVACPSO[50]	0.760775	0.323020	1.481183	0.036377	53.718524	9.860218
IJAYA [19]	0.760875	0.322847	1.481168	0.036281	52.218878	9.860751
EVPS [6]	0.760772	0.325040	1.481811	0.036352	53.896023	9.860934
IMFO [35]	0.760774	0.323468	1.481322	0.036371	53.760868	9.860255
CAO[37]	0.760775	0.323018	1.479167	0.036377	53.718289	9.860218
SOS [39]	0.760778	0.322467	1.481013	0.036379	53.571905	9.860547
MLBSA[38]	0.760775	0.323020	1.479168	0.036377	53.718523	9.860218

Results combination of 10 algorithms for precision formulation

Table 3 shows the best results of the precision formulation of 20 independent tests from the different algorithms. Table 3 shows best results of precision formulation of twenty independent tests of different algorithms. From there, we observe that the RMSE values of 9.86 ∗ 10-4 from the approximate formulation is now 7.73 ∗ 10-4. we observ a significant error reduction, that also provides a good precision in various parameters extracted. Table 4 gives summary statistics and performances of two formulations based on algorithms considered.

Table 3

Best results from 10 algorithms of accuracy formulation.

Algorithms	Best parameters
	Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)	Best RMSE x 10−4
GAMS[3]	0.760788	0.310684	1.477268	0.036547	52.889789	7.730062
MADE[36]	0.760787	0.310684	1.475258	0.036546	52.889734	7.730062
ITLBO[25]	0.760787	0.310684	1.475258	0.036546	52.889790	7.730062
IMFO[35]	0.760787	0.310830	1.475305	0.036544	52.904381	7.730066
MLBSA[38]	0.760787	0.310684	1.475258	0.036546	52.889790	7.730062
TVACPSO[50]	0.760788	0.310684	1.475258	0.036546	52.890001	7.730062
IJAYA [19]	0.760822	0.305965	1.473717	0.036634	52.920663	7.731828
CAO [37]	0.760787	0.310684	1.475258	0.036546	52.889778	7.730062
SOS[39]	0.760786	0.310641	1.475244	0.036548	52.905131	7.730076
EVPS[6]	0.760780	0.317061	1.477295	0.036458	53.337698	7.764575

Table 4

Statistics of ten algorithms.

Methods	Algorithms	BestRMSE × 10−4	MeanRMSE × 10−4	WorstRMSE x10−4	std	BestCpu time (S)
Approximate formulation	GAMS [3]	9.860218	9.860218	9.860218	0	0.016
	MADE [36]	9.860218	9.860218	9.860218	3.207e-15	0.17
	ITLBO [25]	9.860218	9860218	9860218	1.49e-17	0.36
	TVACPSO [50]	9.860218	9.860793	9.870897	2.381e-7	2.61
	IJAYA [19]	9.860751	9.933388	10.072718	6.22e-6	1.83
	EVPS [6]	9.860934	9.861713	9.86242194	4.416e-8	270
	IMFO [35]	9.860255	9.920388	10.47119	1.316e-05	0.71
	CAO [37]	9.860218	10.324154	12.173784	6.929e-5	2.44
	SOS [39]	9.860547	9.871444	9.920406	1.89e-6	3.71
	MLBSA [38]	9.860218	9.860218	9.860218	2.387e-12	1.07
Accurate formulation	GAMS [3]	7.730062	7.730062	7.730062	0	0.032
	MADE [36]	7.730062	7.730062	7.730062	4.187e-15	0.22
	ITLBO [25]	7.730062	7.730062	7.730062	1.431e-17	0.98
	IMFO [35]	7.730066	7.7352115	7.772781	9.890e-7	1.32
	MLBSA [38]	7.730062	7.730062	7.730062	1.086e-16	1.53
	TVACPSO [50]	7.730062	7730063	77300762	3.01e-10	5.12
	IJAYA [19]	7.731828	7.756194	7.854054	2.864e-6	1.14
	CAO [37]	7.730062	7.755962	8.195525	1.036e-5	4.05
	SOS [39]	7.730076	7.730448	7.731448	4.36e-8	7.67
	EVPS [6]	7.764575	7.762812	7.785691	1.52e-6	1024

Table 4 shows the statistics of results from twenty independent tests obtained from 10 algorithms of two formulations with:

Best: representing minimum values of RMSE from twenty tests;

Average: RMSE average of 20 tests;

Worst: optimal RMSE value from twenty tests,

std: standard of deviation.

Cpu: the optimum calculation time of 20 tests.

From Table 4, we can concluded that the computation time of the approximation formulation is reduced compared to that of the accurate formulation.

Case study 1

The essential step of this case is the consideration DSO for single diode model of RTC France. Table 5 shows 5 parameters with best values. Figures 5 and 6 present IV and PV curves measurements and estimations respectively.

Table 5

Best results from RTC France SD.

Parameters	Best solutions
Ir(A)	0.760787
Io(μA)	0.310684
n	1.475258
Rs(Ω)	0.036546
Rp(Ω)	52.889786
OF(RMSE × 10−4)	7.730062

Figure 5

I–V characteristic of I–V Curves measured and estimated.

Figure 6

Characteristics of P–V curves measured and estimated.

Table 5 shows a better RMSE achieved by DSO. The comparison of values with the ones determined at Table 2 and Table 3 clearly shows the better results of DSO compared to other proposed methods. Characteristics of IV and PV curves in Figure 5 and Figure 6 are plotted from the determined DSO parameters, representing data measurements and estimations. We observe on each curve, measured data, which correspond to estimated data: this clearly points to the precision parameters of DSO algorithm.

Case study 2

The case study concerns dual diode model from RTC France combined with DSO. Best results from 7 estimated RTC France parameters were compared to recent literatures presented at Table 6. Best values determined are observed from Table 6. DSO provides best values from 17 recently known algorithms. Statistic at Table 7 shows for each independent tests, the algorithm accuracy.

Table 6

Best results from RTC-France DD in relation to recent literature.

Algorithms	Iph(A)	Io1(μA)	Best parameters
			Io2(μA)	n1	n2	Rs(Ω)	Rp(Ω)	BestRMSE × 10−4
DSO	0.760813	0.086953	2.177280	1.371206	1.9999	0.038033	58.371326	7.325513
MPSO [49]	0.760812	0.008971	2.136189	1.373644	2	0.037994	58.24134	7.3257
C-HCLPSO-III8 [47]	0.76081	0.008797	0.97376	1.3795	1.817	0.037640	55.796	7.4259
TVACPSO [50]	0.760809	0.004046	0.092746	1.327160	1.735315	0.037973	56.54905	7.4365
ELPSO [48]	0.76080	1e-6	0.099168	1.386091	1.835767	0.037551	55.920471	7.4240
EHA-NMS [41]	0.760781	0.225974	0.749346	1.451017	2.000000	0.036740	55.485441	9.824848
DE-WAO [30]	0.760781	0.225974	0.749346	1.451017	2.00000	0.036740	55.485437	9.82484
ABC-TRR [28]	0.760781	0.225974	0.749349	1.451017	2.000000	0.036740	55.485438	9.824849
ISCE [22]	0.760781	0.225974	0.749348	1.451016	2.0000	0.036740	55.485444	9.824849
BHCS [26]	0.76078	0.74935	0.22597	1.45102	2.0000	0.03674	55.48544	9.82485
PGJAYA [45]	0.7608	0.21031	0.88534	1.4450	2.0000	0.0368	55.8135	9.8263
HFAPS [34]	0.760781	0.225974	0.749358	1.45101	2.0000	0.036740	55.4855	9.8248
TLABC [29]	0.76081	0.42394	0.24011	1.45671	1.9075	0.03667	54.66797	9.84145
GOFPANM [23]	0.760781	0.749347	0.225974	1.451016	2.0000	0.036740	55.485448	9.82484
ORcr-IJADE [32]	0.760781	0.225974	0.749348	1.451017	2.00000	0.036740	55.485438	9.824858
IWAO [21]	0.7608	0.6771	0.2355	1.4545	2.0000	0.0367	55.4082	9.8255
CSO [33]	0.76078	0.22732	0.72785	1.45151	1.99769	0.036737	55.3813	9.8252
ABSO [18]	0.76078	0.26713	0.38191	1.46512	1.98152	0.03657	54.6219	9.8344

Table 7

Best Photo Watt-PWP201 SD module results compared to actual methods.

	Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)	BestRMSE × 10−4
DSO	1.032357	2.496596	1.314835	1.240547	748.32309	2.039992
MPSO [49]	1.032230	2.552134	1.31884	1.238450	762.9058	2.041
WDOWOAPSO [34]	1.032382	2.512911	1.317304	1.239288	744.71435	2.046535
GCPSO [59]	1.032382	2.512922	1.317305	1.239288	744.71663	2.046535
TVACPSO [50]	1.031435	2.6386	1.321018	1.235611	821.59514	2.0530
SDA [40]	1.030517	3.481614	1.349970	1.201288	981.59961	2.425074
EHA-NMS [41]	1.030514	3.482263	1.35119	1.201271	981.98225	2.425075
DE-WAO [30]	1.030514	3.482263	1.35119	1.201271	981.98214	2.425075
ABC-TRR [28]	1.030514	3.482263	1.35119	1.201271	981.98223	2.425075
ISCE [22]	1.030514	3.482263	1.35119	1.201271	981.98228	2.425075
PGJAYA [45]	1.0305	3.4818	1.351177	1.2013	981.8545	2.425075
HFAPS [34]	1.0305	3.4842	1.351247	1.2013	984.2813	2.4251
TLABC [29]	1.03056	3.4715	1.35087	1.20165	972.93567	2.42507
GOFPANM [23]	1.030514	3.482263	1.35119	1.201271	981.98232	2.425075
ORcr-IJADE [32]	1.030514	3.482263	1.35119	1.201271	981.98224	2.425074
(IWAO) [21]	1.0305	3.4717	1.350869	1.2016	978.6771	2.4251

This second phase of case study concerns the dual diode model for DSO and RTC France. Table 6 shows best estimation results for the seven parameters from RTC France in comparison to recent literature. Observation from Table 6 shows that out of 17 recently proposed algorithms, best values were determined by DSO algorithm. The statistics at Table 7 equally show for each individual test, precision of the algorithm.

Case study 3

The present case study has for reference model of PV module Photo Watt-PWP 201. In series having 36 cells with environmental conditions such as G = 1000W/m2 for the irradiation and the temperature of T = 45 °C. the study PV is also known as PV test model. Measurements and electrical characteristics are from [38]. Table 7 shows best results from 5 parameters determined from this model considering recent publications. The RMSE as well as the determined parameters support the hypothesis that DSO algorithm gives best PV module parameters. Tables 7 and 9 show clearly that DSO exhibits good precision and stability compared to other more recent algorithms proposed. Figure 7 shows the best DSO convergence curve from twenty independent tests. Objective function of each experiment was several times used to determine best solutions. Five (5) plotted parameters (overall current, current mean, current median…) present DSO statistic for each function evaluated. Figure 7 explicitly shows objective function convergence after 400 of iterations of each evaluated function: showing the precision and DSO algorithm stability described by standard of deviation at Table 9.

Table 9

Statistics of RTC-France SD and DD, Photo Watt-PWP 201, STM6- 40/36 SD in recent literature.

	Algorithms	Best RMSE × 10−4	Mean RMSE × 10−4	Worst RMSE x10−4	std
RTC FRANCE SD	DSO	7.730062	7.730062	7.730062	1.272E-16
	TVACPSO [50]	7.7301	7.7301	7.7301	5.580E-10
	C-HCLPSO-III8 [47]	7.73006	NA	NA	8.235E-10
	EHA-NMS [41]	9.824848	9.826829	9.860219	5.190E-07
	DE-WAO [30]	9.860219	9.860219	9.860219	3.545E-10
	ABC-TRR [28]	9.86021	9.86021	9.86021	6.150E-17
	ISCE [22]	9.860219	9.860219	9.860219	3.98E-17
	BHCS [26]	9.86022	9.86022	9.86022	2.612E-17
	TLABC [29]	9.86022	9.98523	10.3970	1.860E-05
	GOFPANM [23]	9.86021	9.860219	9.860219	5.590E-15
	CSO [33]	9.8602	9.8602	NA	5.494E-9
RTC FRANCE DD	DSO	7.325513	7.488326	7.683173	1.385E-05
	MPSO [49]	7.3257	NA	NA	NA
	C-HCLPSO-III8 [47]	7.4259	NA	NA	5.549E-11
	TVACPSO [50]	7.4365	7.5883	7.8476	1.104E-05
	ELPSO [48]	7.4240	7.5904	7.9208	9.749E-06
	EHA-NMS [41]	9.824848	9.826829	9.860219	5.190E-07
	DE-WAO [30]	9.82484	NA	NA	NA
	ABC-TRR [28]	9.824849	9.825556	9.860219	4.950E−07
	ISCE [22]	9.824849	9.827740	9.861092	4.610E-07
	HBCS [26]	9.82485	9.83800	9.86865	1.538E-06
	PGJAYA [45]	9.8263	9.85821	9.949852	2.537E-06
	HFAPS [34]	9.8248	NA	NA	NA
	TLABC [29]	9.84145	10.5553	15.0482	1.550E-04
	GOFPANM [23]	9.82484	9.9548	1.34053	6.518E-05
	ORcr-IJADE [32]	9.824858	NA	NA	NA
	IWAO [8, 21]	9.8255	9.9693	10.889	1.9297E-05
	CSO [33]	9.8252	9.9619		3.467E-05
	ABSO [18]	9.8344	NA	NA	NA
	Algorithms	Best RMSE × 10−3	Mean RMSE × 10−3	Worst RMSE x10−3	std
Photo Watt-PWP 201 SD	DSO	2.039992	2.039992	2.039992	2.704E-12
	MPSO [49]	2.041	NA	NA	NA
	WDOWOAPSO [34, 42]	2.046535	2.046535	2.046535	1.267E-12
	GCPSO [59]	2.046535	2.046535	2.046536	1.105E-10
	TVACPSO [50]	2.0530	2.0530	2.0537	1.340E-7
	SDA [40]	2.425074	NA	NA	NA
	EHA-NMS [41]	2.425075	2.425075	2.425075	5.530E-17
	DE-WAO [30]	2.425075	2.425092	2.425442	6.270E-08
	ABC-TRR [28]	2.425075	2.425075	2.425075	9.680E-17
	ISCE [22]	2.425075	2.425075	2.425075	2.470E-17
	PGJAYA [45]	2.425075	2.425144	2.426764	3.071E-07
	HFAPS [34]	2.4251	NA	NA	NA
	TLABC [29]	2.42507	2.42647	2.44584	3.998E-06
	GOFPANM [23]	2.425075	2.425075	2.425075	2.918E-16
	ORcr-IJADE [32]	2.425074	NA	NA	NA
	IWAO [21]	2.4251	2.4269	2.4335	2.236E-06
	ABSO [18]	9.8344	NA	NA	NA
STM6- 40/36 SD	DSO	1.721921	1.721921	1.72192151	5.983E-16
	SDA [40]	1.729645
	RcrIJADE [16]	1.729813	1.729813	1.729813	1.180E-14
	BHCS [26]	1.72981	1.83648	3.32985	4.059E-04
	NoCuSa [43]	1.72981	4.12985	3.10757	9.222E-02
	ELPSO [48]	2.1803	2.2503	3.7160	2.9211E-4
	TVACPSO [50]	2.1803	2.1803	2.1803	6.3747E-9
	HFAPS [34]	1.97
	EHA-NMS [41]	1.729813	1.729814	1.729814	2.300E-17
	MADE [36]	1.7298	1.7298	1.7298	8.490E-14
	ISCE [22]	1.729813	1.729814	1.729814	2.430E-17
	TLABC [29]	1.80612	1.97321	2.21357	8.767E-05

Figure 7

Photowatt-PWP201 module with DSO convergence curves.

Case study 4

This case refers to monocrystalline PV module STM6-40/36 which consists of 36 cells in series. The environmental measurement conditions are G = 1000W/m2, and T = 51 °C. The electrical characteristic and the set of measures have been taken from [25]. Table 8 shows the best results obtained for five parameters evaluated from STM6- 40/36 PV module of recent literature. These parameters determined by DSO compared to 11 propositions of recent algorithms is observed from Table 8. These parameters are very close to those determined [16] and [40].RMSE = 1.721921 × 10−3 is better compared to the one of 11 propositions of algorithms.

Table 8

Compared recent methods of literatures.

Algorithms	Best parameters
	Iph(A)	Io(μA)	n	Rs (mΩ)	Rp(Ω)	Best RMSE × 10−3
DSO	1.663903	1.741245	1.51839	153.640	573.53392	1.721921
SDA [40]	1.663949	1.702581	1.51783	156.421	570.22744	1.729645
RcrIJADE) [16]	1.663904	1.738656	1.52030	153.8557	573.41858	1.729813
HBCS [26]	1.66390	1.7386	1.5203	153.72	573.418	1.72981
NoCuSa [43]	1.66390	1.73866	1.5203	153.72	573.418	1.72981
ELPSO [48]	1.666268	45.9614	1.40162	0.5	497.74731	2.1803
TVACPSO [50]	1.666268	45.96154	1.40162	0.5	497.74866	2.1803
HFAPS [34]	1.6663	1.0703	1.47266	0.24849	490.03	1.97
EHA-NMS [41]	1.663904	1.738656	1.520302	4.273771	573.418	1.729813
MADE [36]	1.6639	1.7387	1.5203	154.8	573,4188	1.7298
ISCE [22]	1.663904	1.738656	1.520302	4.273771	573.4185	1.729813
TLABC [29]	1.66317	2.14043	1.54354	130.68	621.34272	1.80612

Table 9 presents the statistics of the recent literatures of RTC-France SD and DD, Photo Watt-PWP201 and STM6- 40/36 SD specifying DSO stability for PV parameters estimation. Results determined from 20 independent tests are: the minimum reached RSME best values; the highest value is the worst RMSE value; the average RMSE shows average value after 20 tests with standard of deviation (std). The smallest standard deviation values for each case expresses DSO algorithm reliability.

Case study 5

The fifth case study therefore concerns experimental PV module data of Sharp ND-R250A5 [59]. Two models composed of single and dual diode used to express DSO performance. Table 10 presents electrical characteristics for PV module Sharp ND-R250A5 at STC. The measured data are available in [42], [59]. The limit of lower and upper bound of Iph, Io, n, Rs, Rp is available in [59] Tables 11 and 12 show the best single and dual diode parameters of Sharp ND-R250A5 PV-module from different irradiations. Figures 8 and 9 present characteristics of IV and PV curves at various operating conditions. Observation from these figures shows that measured data merge with estimated data under various irradiation and temperature. RMSE value observed from each condition from Tables 11 and 12, show estimated low and fit curve which present DSO to achieve best PV module parameters under various environmental conditions.

Table 10

Module parameter of Sharp ND-R250A5.

Parameters	Values
Isc(A)	8.68
Voc(V)	37.6
Imp(A)	8.10
Vmp(V)	30.9
Ns	60

Table 11

Results of SD model of ND-R250A5.

Irradiation/temperature	Best parameters
	Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)	Best RMSE × 10−3
G = 1040w/m2T = 59 °C	9.144865	0.995851	1.204937	0.591870	5000	7.697717
G = 924w/m2T = 68 °C	8.151384	1.46680	1.199462	0.590546	5000	7.72753142
G = 743w/m2T = 67 °C	6.560009	0.180341	1.078311	0.633776	5000	8.93948385
G = 544w/m2T = 53 °C	4.771023	6.089760e-2	1.105081	0.615500	5000	4.828055534

Table 12

Result of different SD models of ND-R250A5.

Irradiation/temperature	Iph(A)	Io1(μA)	Best parameters
			Io1(μA)	n1	n2	Rs(Ω)	Rp(Ω)	Best RMSE × 10−3
G = 1040w/m2T = 59 °C	9.144865	1.91425E-6	9.9584E-1	1.204905	1.204937	0.591870	4999.9999	7.697717191
G = 924w/m2T = 68 °C	8.151384	2.849850	1.466799	1.199462	1.199462	0.590546	5000	7.727531
G = 743w/m2T = 67 °C	6.560009	1.5959E-1	2.07428E-2	1.078311	1.078311	0.633776	5000	8.939483
G = 544w/m2T = 53 °C	4.771023	4.679367E-2	1.4103E-2	1.105081	1.105081	0.615500	4999.9999	4.828055

Figure 8

I–V curves measurement and estimation of the characteristics.

Figure 9

P–V curves measurement and estimation of the characteristics of SD model of ND-R250A5.

Case study 6

This case study concerns STP6-120/36 module, PV module (STE 4/100) and the Leybold solar module (LEYBOLD 664 431). PV module STP6-120/36 consists of 36 monocrystalline cells connected in series has environmental measurement conditions G = 1000W/m2, T = 55 °C, and electrical measurements and characteristic can be taken [25].

The polycrystalline PV module (STE 4/100) consists of 4 cells in series. The environmental measurement conditions is G = 900W/m2 and T = 22 °C. The polycrystalline PV module (LEYBOLD 664 431) consists of 20 cells in series; the environmental measure conditions are G=360W/m2, T=24°C. The electrical characteristic and the set of measures of (STE 4/100) and Leybold Solar Module (LEYBOLD 664 431) have been taken from [63]. Table 12, Table 13 shows the best-estimated results of 5 parameters for each PV module from the current literature. The RMSE as well as the determined parameters confirm DSO algorithm reliability hypothesis, which gives best precision for PV modules parameters.

Table 13

Best STP6-120/36, PV (STE 4/100) and Leybold solar (LEYBOLD 664431) modules results.

Photovoltaics module	Algorithms	Best parameters					Best RMSE
		Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)
STPE6-120/36	DSO	7,4752	1,9308	1,2818	0,004691	15.83880	1.42510 × 10−2
	MADE [36]	7,4725	2.335	1.2601	0.0046	22.2199	1.6601 × 10−2
	ITBLO	7.4725	2.3350	1.2601	0.0046	22.2199	1.6601 × 10−2
	IJAYA	7.4672	2.2536	1.2571	0.0046	27.5925	1.6731 × 10−2
	STBLO	1.1679	7.4814	1.2048	0.0055	9.8	1.6211 × 10−2
LEYBOLD 664 431	DSO	0.1540	1.228x10−2	1.3865	5.1552	2202.828	3349651 × 10−4
	Brent [63]	NA	NA	NA	NA	NA	8.3838 × 10−4
STE 4/100	DSO	0.0264	2.011x10−3	1.2015	1.4787	2128.8	2.9852 × 10−4
	Brent [63]	NA	NA	NA	NA	NA	3.339 × 10−4

Case study 7

This case refers to model GL-100 PV module containing 36 cells. The study was done on a PV module and a string containing 6 PV modules. The study aims at showing convergence of the method when numbers of points N of the characteristic is varied. The I–V characteristic's initial number of points is N = 113 points and was obtained in [28]. Points were reduced to obtain different following values: N = 72; N = 36; N = 29; N = 18; N = 9. Table 14 shows that the algorithm converges and is more precise when numbers of points of the I–V characteristic were reduced. However, if these points are minimal, the algorithm will have difficulty reaching its optimal result, as seen in cases where NP = 9 (T = 33.7).

Table 14

Different result of SD model of GL-100 at different N.

Irradiation/Temperature	N	Best parameters						Time(s)
		Iph(A)	Io(μA)	n	Rs(Ω)	Rp(Ω)	Best RMSE × 10−9
PV moduleG = 328 w/m2T = 26 °C	143	1.97971	0.001291	1.10769	0.58512	359.50787	6.46551	42.67
	72	1.97971	0.001291	1.10769	0.58512	359.50787	4.26242	31.80
	36	1.97971	0.001291	1.10769	0.58512	359.50788	3.14067	25.15
	29	1.97971	0.001291	1.10769	0.58512	359.50786	2.84234	19.87
	18	1.97971	0.001291	1.10769	0.58512	359.50788	2.12389	17.84
	9	1.97971	0.001291	1.10769	0.58512	359.50787	1.16520	14.63
PV moduleG = 499w/m2T = 33.7 °C	143	3.00787	0.002054	1.08407	0.62145	253.16931	6.54356	56.07
	72	3.00787	0.002054	1.08407	0.62145	253.16931	4.68112	27.86
	36	3.00787	0.002054	1.08407	0.62145	253.16931	2.71226	23.20
	29	3.00787	0.002054	1.08407	0.62145	253.16931	1.48724	22.7
	18	3.00787	0.002054	1.08407	0.62145	253.16935	0.47719	21.44
	9	3.00211	0.046068	1.22799	0.14022	254.88638	13355.83	13.55
PV moduleG = 700w/m2T = 40.1 °C	143	4.21951	0. 29322	1.05044	0.65662	161.83471	6.35424	66.54
	72	4.21951	0. 29322	1.05044	0.65662	161.83471	4.20372	32.02
	36	4.21951	0. 29322	1.05044	0.65662	161.83471	2.94262	31.02
	29	4.21951	0. 29322	1.05044	0.65662	161.83471	2.91417	30.62
	18	4.21951	0. 29322	1.05044	0.65662	161.83471	2.27727	15.50
	9	4.21951	0. 29322	1.05044	0.65662	161.83471	1.46543	13.89
PV stringG = 754 w/m2T = 47.7 °C	143	4.54615	0.00357	4.54615	3.97906	889.53643	6.89827	69.39
	72	4.54615	0.00357	4.54615	3.97906	889.53643	4.97891	34.36
	36	4.54615	0.00357	4.54615	3.97906	889.53643	3.12043	25.08
	29	4.54615	0.00357	4.54615	3.97906	889.53643	3.09454	24.28
	18	4.54615	0.00357	4.54615	3.97906	889.53643	1.93749	18.00
	9	4.54615	0.00357	4.54615	3.97906	889.53643	1.27052	15.31
PV stringG = 658 w/m2T = 39.5 °C	143	3.96612	0.00171	6.19872	3.88369	936.64812	6.43359	61.81
	72	3.96612	0.00171	6.19872	3.88369	936.64812	4.68042	31.89
	36	3.96612	0.00171	6.19872	3.88369	936.64812	3.38296	30.45
	29	3.96612	0.00171	6.19872	3.88369	936.64812	2.76299	25.35
	18	3.96612	0.00171	6.19872	3.88369	936.64812	1.61304	17.36
	9	3.96612	0.00171	6.19872	3.88369	936.64812	1.06481	15.67

These results show how important is the choice of the I–V characteristic numbers of points N. From Table 14, by choosing a reduced number of points, one obtains a faster execution time and an optimal result. We also notice that number of points does not influence the pace and slopes of the characteristic.

Conclusion

This article describes assumptions made as well as the clarifications of two formulations for best parameters of photovoltaic estimation: formulation of approximation and formulation of precision. Ten current published algorithms have been implemented based on the two formulations. Algorithm convergence studies carried out considering the numbers of points of variation of voltage current characteristics. The determined results of these formulations present accuracy precision formulation. The new self-adaptation algorithm called DSO was proposed for simulation of squadron of drones of a control center to determine the best cell and photovoltaic module parameters. A comparative study of results determined from different studies cases based on the description of methods in literatures. This makes it possible to clarify DSO performance in extracting cell, module, and array PV parameters. From results of 10 algorithms and DSO, it was observed that precision formulation is the most recommended for the rest of the work. It should be noted that the proposed solutions in this article would allow scientific communities to choose a better formulation for PV parameters estimation. In addition, combining DSO and method of Newton Raphson is of great advantage in seeking optimal solution to the problems in power systems. The combine DSO and the method of Newton Raphson could be used for precise parameter extraction. Besides, accurated results can also be determined by combinating Newton Raphson's method with any heuristic algorithm.

Declarations

Author contribution statement

Patrick Juvet Gnetchejo: Conceived and designed the experiments; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Salomé Ndjakomo Essiane & Pierre Ele: Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data.

Abdouramani Dadjé: Performed the experiments; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Funding statement

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability statement

Data included in article/supp. material/referenced in article.

Declaration of interests statement

The authors declare no conflict of interest.

Additional information

No additional information is available for this paper.