Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/362362432

Optimal design of the modelling parameters of photovoltaic modules and


array through Metaheuristic with Secant Method

Article  in  Energy Conversion and Management X · August 2022


DOI: 10.1016/j.ecmx.2022.100273

CITATION READS

1 131

5 authors, including:

Patrick JUVET Gnetchejo Salomé Ndjakomo Essiane


University of Douala University of Douala/University of Yaounde I
16 PUBLICATIONS   148 CITATIONS    65 PUBLICATIONS   280 CITATIONS   

SEE PROFILE SEE PROFILE

Abdouramani Dadje Daniel Eutyche Mbadjoun Wapet


University of Ngaoundere University of Yaounde I
21 PUBLICATIONS   213 CITATIONS    18 PUBLICATIONS   117 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

This project aims to show the importance of injecting decentralized energy into the grid and to try to solve the problems caused by injection. View project

Optimization of thermal plant by reducing consumpion of fuel and gas emission View project

All content following this page was uploaded by Patrick JUVET Gnetchejo on 02 August 2022.

The user has requested enhancement of the downloaded file.


Energy Conversion and Management: X 15 (2022) 100273

Contents lists available at ScienceDirect

Energy Conversion and Management: X


journal homepage: www.sciencedirect.com/journal/energy-conversion-and-management-x

Optimal design of the modelling parameters of photovoltaic modules and


array through metaheuristic with Secant method
Patrick Juvet Gnetchejo a, *, Salomé Ndjakomo Essiane a, b, Abdouramani Dadjé c, *,
Daniel Mbadjoun Wapet d, Pierre Ele d
a
Laboratory of Technologies and Applied Sciences, University of Douala, Cameroon
b
Signal Image and Systems Laboratory, Higher Technical Teacher Training College of Ebolowa, University of Ebolowa, Cameroon
c
School of Geology and Mining Engineering, University of Ngaoundéré, Cameroon
d
Laboratory of Electrical Engineering, Mechatronic and Signal Treatment, National Advanced School of Engineering, University of Yaounde 1, Cameroon

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

Keywords: Nowadays, the performance of photovoltaic (PV) panels is a priority research topic. Obtaining the best perfor­
Photovoltaic parameters mance of these panels requires an adequate and accurate model. This paper employs a new approach based on a
PV array modified social network search algorithm combined with the Secant method (MSNS-SEC) to produce the best
Estimated currents
parameters of a photovoltaic cell, module, and array. To improve the performance of the parameters to be
Optimization
Solar cell
estimated, a control parameter via a Gaussian and Cauchy distribution is randomly added to the search space to
allow the agents to converge to the optimal solution. Then the Secant method is inserted into the objective
function to calculate the best-estimated currents. The application of the proposed model on three different
systems, namely a PV array, cell, and module, and the subsequent comparison with existing methods exhibit the
high accuracy of the proposed method, with the best root mean square error of 6.6851 × 10− 4 for the RTC cell,
1.5411 × 10− 3 for the Photowatt PWP module and 0.0134 for the experimental 18 PV array.

methods have been developed in the literature to estimate the best PV


parameters. These methods can be: numerical, analytical, evolutionary,
Introduction
and hybrid. Among the existing methods to obtain the parameters of a
PV, the current–voltage characteristic equation is the most suitable, as it
The global energy deficit has led to the development of new energy
contains all the properties and parameters of the PV. However, the
sources, including solar energy, which is a clean, available and free
transcendence of this equation makes it difficult to solve, as it leads to an
source of energy [1]. The key element in solar energy production is the
optimisation problem. To overcome this problem, metaheuristics are
photovoltaic (PV) cell [2]. The cells are assembled in series and/or
probably the most effective methods, as they have been proved to solve
parallel to form the photovoltaic module [3]. To produce the appro­
several problems in various fields [5–7].
priate electrical power, the panels are connected in series and/or par­
Lambert’s W function was introduced by [8] for the extraction of the
allel to form the photovoltaic array. The design, prediction, sizing,
parameters a photovoltaic model. The model is based on two principles:
diagnosis and maintenance of photovoltaic system installations require a
Thevenin’s theorem the first one, used to identify the linear part of the
reliable and accurate model of the cell, the module and the PV array [4].
PV cell model. A piecewise linear function, for the second part of the
Three different models, namely with one, two, and three diodes, have
nonlinear function. In the works done by [9], authors used a new
been established in the literature. These models are characterised by
empirical relationship to extract parameters of one diode model. The
parameters that have to be extracted accurately. Obtaining these
empirical method calculates the initial value of the series resistance from
different parameters remains a crucial and difficult problem. Several

Abbreviations: PV, Photovoltaic; MSNS-SEC, Modified Social Network Search Algorithm combined with the Secant method; BES, Bald Eagle Search; SNS, Social
Network Search; RMSE, Root Mean Square Error; DEDCF, Differential evolution with dynamic control factors; DE, Differential Evolution; GAMS, General Algebraic
Modeling System; OLGBO, Orthogonally-Adapted Gradient-Based; ABC-Ls, Artificial Bee Colony Algorithm based; ODGB, Opposition Decided Gradient-Based;
ADHHO, Adaptive Harris Hawks Optimization with persistent trigonometric differences; HBO, Heap-Based Optimizer; NEPO, New Emperor Penguin Optimisation-
based.
* Corresponding authors.
E-mail addresses: patrijuvet@yahoo.fr (P.J. Gnetchejo), abdouramanid@yahoo.fr (A. Dadjé).

https://doi.org/10.1016/j.ecmx.2022.100273
Received 18 April 2022; Received in revised form 25 July 2022; Accepted 26 July 2022
Available online 29 July 2022
2590-1745/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

Nomenclature Yi viewpoint or position


OF objective function
Symbols P firmware perturbation
θ vector of parameters q electron charge [C]
Iph photoelectric current [A] RMSE root mean square error
I0 , I01 , I02 Diode reverse saturation currents [μA] STC Standard testing condition (1000 W/m2, 25 ◦ C)
n, n1 , n2 Diode ideality factors T Temperature [K]
RS series resistance [Ω] TC trial coordinates
RP parallel resistance [Ω] Ub Upper bound
Ip current through parallel resistor [A] V cell output voltage [V]
Imes measured current [A] Vd Voltage Applied across the diode [V]
I cell output current [A] Vmes measured voltage [V]
Iiext (θ) is the estimated current Pi disturbance
Id , Id1 , Id2 diode currents [A] Voc open-circuit voltage [V]
Iiext (θ) estimated current [A] Gi A random function generated from the Cauchy distribution
K Boltzman constant [J/K] F(θ) objective function to minimize
N number of the experimental I-V data pairs Nr number of users in the group
NS number of cells connected in series Lb Lower bound
β: stopping criteria

the slope of the current–voltage characteristic provided by the manu­ differential algorithm with a dynamic control parameter. The control
facturer. In general, the numerical data of the current–voltage charac­ factor includes crossover and mutation, allowing the dynamic fit to
teristic is never provided at the time of purchase of a PV, so for this converge to the best solution. Although metaheuristic algorithms are
method, it will be necessary to use image processing to obtain this data more suitable for solving an optimization problem, many of them have
from the technical documentation: this data will not always be accurate. drawbacks such as: premature convergence to a local minimum, high
Analytical methods work well under normal weather conditions, but the execution time when the search space increases, instability of the solu­
changing of atmospheric condition make the analytic method useless tion for several trials. In order to overcome the drawbacks mentioned
[4]. In addition, approximating the equations considerably degrades the above, in this paper, a new approach based on the modified Social
accuracy of the results. Furthermore, [10] proposed a new approach Network Search (SNS) algorithm combined with the Secant method
with good accuracy for the single diode model of a PV module. The (MSNS- SEC) is used to extract the best parameters of a photovoltaic cell,
approach is implemented in Matlab/Simulink environment based on the module and array. The Social Network Search (SNS) algorithm simulates
Levenberg Marquardt algorithm. A method based on the examination of the behaviour of individuals in social networks in search of popularity.
the current–voltage characteristic was used by [11]. A system of five Like most metaheuristics, the SNS algorithm sometimes gets stuck in a
equations is solved using the Newton-Raphson algorithm. However, local minimum. To avoid this premature convergence, a perturbation
with the Newton-Raphson algorithm, a system of five equations is equation is randomly added in the search space to allow the agents to
required to determine the five parameters of the one-diode model; in converge to the optimal solution. Then, the Secant method is inserted in
addition, the solution of the Jacobian matrix is required in the Newton- the objective function for the computation of the best estimated current.
Raphson algorithm. This Jacobian matrix contains twenty-five first and The originality and the main contribution of this works are follows:
second derivative terms that make the method more complex. The main
disadvantage of numerical techniques such as Newton Raphson is the - We have Combined for the first time the social network search al­
need to perform extensive calculations for convergence; they do not give gorithm with the secant method to extract the best parameters of a
accurate results when the number of parameters to be estimated in­ PV cell, module and array.
creases. Despite the effeciency of numerical methods, their slow - To improve the research space of social network search algorithm, a
convergence does not always guarantee the best result, as they may control parameter via a Gaussian and Cauchy distribution is
converge to a local minimum and the choice of the initial condition is randomly added to the search space to allow the agents to converge
often not easy [4]. to the optimal solution.
[12] Used the penguin emperor optimisation method to estimate the - The Secant method is inserted into the objective function to calculate
five parameters of a single diode model. In order to reduce the execution the best-estimated currents.
time of the algorithm, two parameters (Iph and Is) were obtained - The proposed method is implemented on a real PV array of 18 PV
analytically. In [13], the differential evolution method is combined with modules.
an analytical method to estimate the best parameters of a PV when the - The obtains results, compared with previous works show the supe­
atmospheric conditions change. The authors in [14] adapted the Harris riority of the methods.
Hawk algorithm with trigonometric persistence to reduce the search
time for the global optimum; the algorithm successfully extracted the The rest of this paper is organised as follows. In the second part, the
best parameters for one, two, and three diode models. A modification of different methods used will be presented; the third section will present
the three-diode model was introduced by [15], where a new objective the different results obtained.
function is used to estimate the different parameters using the Heap-
based algorithm. In [16], the bee colony algorithm is used with a local Materials and methods
search procedure to improve the exploration capability of the basic al­
gorithm. A hybrid method was used by [17] to determine the parameters Problem formulation
of a two-diode model; to; to improve the convergence of the wind al­
gorithm, another algorithm, the Fruit Fly algorithm, was introduced to The main objective is to derive the mathematical model of a PV cell
improve the exploration capabilities of the latter. [18] proposed a or module by accurately estimating the various unknown parameters

2
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

Fig. 1. One diode PV cell.

Fig. 2. Double diode PV cell.

[
that characterize this model. These different parameters are:θ = IPh , I0 , The SNS algorithm
] [
n, Rs , Rp for the single diode model (Fig. 1), and θ = IPh , I01 , I02 , n1 , n2 ,
]
Rs , Rp for the two-diode model (Fig. 2). Inspiration
The Social Network Search (SNS) algorithm simulates the behaviour
The objective function defined as the difference between the
of individuals in social networks in search of popularity [20]. Like all
measured and estimated currents is usually quantified by the mean
metaheuristic algorithms, the SNS algorithm consists of a population,
square error [19,20].
which represents the views of each user in social networks. By sharing
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

√1 ∑ N their views, each user improves their position in the network.
RMSE = √ (Imes − Iext )2 (1)
N i=1
Principle and model
One user’s viewpoint can be affected by another user’s viewpoint in
N : the number of measurement points of the current–voltage
four ways [21]:
characteristic;
Imes : the set of points of the experimentally measured current;
Iext : the estimated currents. Imitation. This is the fact that as one user’s idea is better, others will seek
In order to have an estimated current–voltage characteristic (Vi , Ii ) to imitate it during their expression. The mathematical model for this is
very close to the measured (Vmes , Imes ), the expression of the estimated defined by:
current for the one-diode model should be defined as follows: Yi,New = Yj + rand(− 1, 1).rand(0, 1).(Yj − Yi ) (5)
[ ( ) ]
Ii,ext = Iph − I0 exp
q(Vi + Ii,ext .Rs )
− 1 −
Vi + Ii,ext .Rs
(2) Yi : is the vector of the viewpoint or position of ith the user.
n.k.T Rp Yj : is the vector of the viewpoint or position of jth the user.
Since the non-linearity of Eq. (2) does not allow for an explicit so­ Yi,New :is the new position of user ith in the search space.
lution, Eq. (2) is written as Eq. (3) below:
[ ( ) ] Conversation. The conversation is the phase of exchange with other
f (Ii,ext ) = Iph − I0 exp
q(Vi + Ii,ext Rs )
− 1 −
Vi + Ii,ext .Rs
− Ii,ext (3) users to choose the best idea. The mathematical model of this viewpoint
n.k.T Rp is defined by:
Where all estimated currents (Ii,ext ) is found by solving the equation. Yi,New = Yl + rand(0, 1).(Yj − Yi ).sign(fi − fj ) (6)
f (Ii,ext ) = 0 (4) Yl : represents the vector of the problem, that is randomly chosen to
This paper uses the Secant method to solve Eq. (4). be discussed.
sign(fi − fj ): represents the difference in opinion between users.

Dispute. During the dispute phase, users can discuss their opinions with

3
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

Fig. 3. Illustration of the Secant method.

a group of users. The mathematical model of this viewpoint is defined problems, three solutions have been proposed in this paper: (i) a control
by: parameter via the Gaussian and Cauchy distribution is randomly added
(∑Nr ) to the search space; (ii) a historical memory function is added; (iii)
t Yt Secant has been used to minimize the error between measured and
Yi,New = Yi + rand(0, 1). − (1 + round(rand).Yi ) (7)
Nr estimated currents.
∑Nr
Yt
M = Nt r : is the average of comments made by other users in the Modification
group AF:1 + round(rand) : is the emphasis a user places on their Randomly select one of the four equations (imitation, conversation,
opinion. disputation, innovation) disperses the search space. In this paper, from
Nr : is the number of users in the group. the original SNS algorithm model, we used only the conservation
equation (Eq. (6)). By focusing only on the conversation, and choosing
Innovation. This is the fact that a user can share an idea from their new the user with the best point of view, the user’s view at the end of these
experience. The mathematical model for this is defined by: exchanges must be better.
Eq. (6) of the conversation is therefore modified as shown below:
d
Yi,New = rand2 × Yid + (1 − t).(lbd + rand1 × (ubd − ldd )) (8)
Yi = Yl + rand.(Ybest − Yi ).sign(fi − fj ) − Pi (12)
d: is the 10th randomly chosen variable in the interval of decision
variables. Ybest is the vector of the user with the best view after evaluation of the
lb: is the lower limit of the variable d. objective function.
ub: is the upper limit of the variable d. Pi is the disturbance equation for achieving a better balance between
In social networks, users have to respect predefined rules. These rules exploration and exploitation. This disturbance equation is defined by:
consist in staying within defined limits. Thus, the boundary of each Pi = Gi *(YM1 − YM2 ) (13)
viewpoint is defined by:
Gi is a random function generated from the Cauchy distribution.
yi = min(yi , ubi )
(9)
yi = max(yi , lbi ) Gi = randchi (MC,ri , 0.1) (14)

During the exchange mechanism, a user’s opinion may change. Then with randch the Cauchy distribution, ri a random integer between [1-
the user may adopt a new idea. The objective function will evaluate the 0], YM1 and YM2 are selected candidates in the search space.
value of the new position, and the following equation will make a
choice: The Secant method
{ The objective function of Eq. (1) is a transcendental equation.
Yi , f (Yi ) < f (Yi,new )
Yi = (10) Several authors considered the estimated current equal to the measured
Yi,new , f (Yi,new )⩾f (Yi )
current to avoid this transcendence [22]. In order to have more accurate
Before running the algorithm, the initial positions must be generated results, we have solved this transcendental problem by inserting the
by the following equation: secant method when calling the objective function.

Yi = lb + rand(0, 1).(ub − lb) (11)


The method. In numerical analysis, the secant method is an algorithm for
At each iteration of the objective function, one of the four moods finding a zero of a functionf .
(imitation, argument, conversation, and innovation) is randomly The secant method is an efficient alternative to Newton’s method
selected. when the function f is not derivable or when no explicit expression of
Complete details of the SNS algorithm are available in [21]. f (x) is known. We therefore replace the tangent of Newton’s method

with the Secant to the curve.

Proposed method: The MSNS- SEC algorithm Principle. Let f have a zero in the interval[x− 1 , x0 ].To obtain a first
approximation of this zero, the idea is to replace f by its linear inter­
Like many metaheuristics, the SNS algorithm suffers from the prob­ polate on[x− 1 , x0 ], i.e. by:
lem of instability, premature convergence, and speed. To solve these

4
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

f (x0 ) − f (x− 1 )
Y(x) = f (x0 ) + (x − x0 ) (15)
x0 − x− 1
The unique linear function whose values coincide with those of f at
x− 1 andx0 . The x1 approximation is then obtained by solving:Y(x1) = 0.
Then:
x0 − x− 1
x1 = x0 − f (x0 ) (16)
f (x0 ) − f (x− 1 )
Geometrically as shown in Fig. 3, this is equivalent to replacing the
curve of equation y = f(x) by the Secant AB and xn+1 is the intersection
of AB with the line(Ox).
As shown in Fig. 3, xn+1 appears to be closer to the zero sought than
xn− 1 orxn . We repeat the process using the points (xn , xn+1 ) to find a
better approximation. Continuing, we obtain the sequence of Eq. (17):
xn − xn− 1
xn+1 = xn − f (xn ) n = 0, 1, 2.....∞ (17)
f (xn ) − f (xn− 1 )
xn converges to the solutionx∞ , and a stopping criterion with preci­
sion β is therefore defined by:
|xn+1 − xn |⩽β (18)

Application of Secant method for the calculation of the best estimated


current.
Considering Eq. (4), the solution of this equation is the estimated
current Iext When calling the objective function (Eq. (1)), the images of
the two points between which the best current is located are calculated
by the following Eqs. (19) and (20):
[ ( ) ]
q(Vi + a.Rs ) Vi + a.Rs
fi (a) = Iph − I0 exp − 1 − − a (19)
n.k.T Rp
[ ( ) ]
q(Vi + b.Rs ) Vi + b.Rs
fi (b) = Iph − I0 exp − 1 − − b (20)
n.k.T Rp

Then the estimated current is iterated by a from the Eq. (21):


b− a
a = a− fi (a) (21)
fi (b) − fi (a)
The best estimate current is determined by iterating Eqs. (19,20,21)
several times with precisionβ.
The algorithm for calculating the best estimated current based on
Secant is given in Fig. 4. Algorithm 1 shows the pseudocode of the
proposed MSNS- SEC method.
Fig. 4. Estimated current based on the Secant method.
Algorithm 1: Pseudocode of MSNS-SEC algorithm

1: Define MSNS- SEC initial parameters (MaxIter,Nuser, UB,LB)


2: for i=1 to Nuser
3: Generate initial population Eq. (11)
4: Evaluate objective function Eq. Eqn 1
(continued on next page)

Table 1
Comparison of RTC with literature.
Methods Parameters
4
Iph(A) Io1(µA) Io2(µA) n1 n2 Rs(Ω) Rp(Ω) Best RMSE × 10−

Proposed 0.7608 0.1501 5.0738 1.4153 2.4640 0.0397 57.1679 6.8514


DSO [22] 0.7608 0.0869 2.1772 1.3712 1.999 0.0380 58.3713 7.3255
DEDCF [18] 0.7608 0.06428 0.9999 1.3577 1.7869 0.0378 56.3793 7.4196
DE [27] 0.7605 0.4232 0.1872 1.8757 1.4360 0.02061 51.9345 7.6300
MPA [23] 0.7608 0.2704 0.2676 1.9488 1.4648 0.03667 53.5615 7.6499
GAMS [24] 0.7607 0.2259 0.7494 1.4510 2.0000 0.03674 55.4854 9.8248
OLGBO [25] 7.6078 0.74391 0.2265 2.0000 1.4512 1.4512 55.3151 9.8248
ABC-Ls [16] 0.7608 0.2279 0.7273 1.4518 1.9952 0.0367 53.5381 9.8257
ODGB [26] 0.7608 02,202 0.8020 1.4489 2.0000 0.0368 55.8326 9.8258
ADHHO [14] 0.76078 2.4672 3.6648 1.4584 2.0000 0.0366 55.09 9.8398
HBO [15] 0.7606 0.6700 0.1970 1.9087 1.4407 0.0368 51.6761 10.4972

5
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

Fig. 5. Current–voltage characteristic.

Algorithm 1: Pseudocode of MSNS-SEC algorithm (continued ) an experimental array consisting of 18 PV modules. For a fair evaluation,
5: End for the parameters were identical for each case study. The maximum
6: for i=1 to MaxIter number of iterations was set to 1000; the number of users Nu = 50, and
7: initialise memory parameters 10 runs were performed for each case.
8: for i=1 to Nuser
9: generate the permutation function Eq. (13) , Eq. (14)
10: Update the new position Eq. (12) Case study 1: RTC France PV cell
11: Clamp the new solution Eq. (9)
12: Evaluate objective function. Eq. Eqn 1
The RTC France cell is a test cell widely used in the literature. The
13: Calculate the best estimated currents Eq(4), Eq. (19) to Eq. (21)
14: Update the best fitness data of the current–voltage characteristic is the one submitted to an
15: Update the memory parameter with Cauchy and Gaussian distribution irradiance G = 1000 W/m2 and a temperatureT = 33◦ C. The manufac­
16: Select the best position and the best finest turer’s and characteristic data are available in [13].
17: if NfBest < f(i)
Table 1 shows the best results obtained by the proposed method. A
18: Update the new position Eq. (10)
19: end if comparative study is also made with [14–16,18,23–27] on the basis of
16: Select the best position and the best finest RMSE.
17: if fBest < precision In this Table 1, we see that the proposed method presents the best
18: end if result with an RMSE = 6.6851 × 10− 4, followed by 7.3255 × 10− 4 for
19: end for
[22] and against 9.8248 × 10− 4 for [24,25] and finally 10.4972 × 10− 4
20: end for
[15] for the worst. We recall that these different comparison results are
the best currently available in the literature.
The best parameters thus obtained were used to plot the cur­
rent–voltage characteristic in Fig. 5. This curve represent the experi­
Results and comments mental measurement of the I-V characteristic (pink) and the estimated I-
V characteristic (green) obtained from the algorithm, which is ploted
To evaluate the proposed method, 3 cases study have been imple­ from Eq. (2). These two characteristics demonstrate the fact that the
mented. The first case is the RTC France cell with a two-diode model, the MSNS-LAG algorithm has succeeded in obtaining better parameters as
second case is the Photowatt module with one diode, and the last case is close as possible to the experimental measurement. We can also observe

Table 2
Comparison of PWP with literature.
Methods Parameters

Iph(A) Io(µA) n Rs(Ω) Rp(Ω) Best


3
RMSE × 10−

Proposed 1.0331 1.1863 45.3536 1.5002 681.9921 1.5411


DSO [22] 1.0323 2.4965 1.3148 1.2405 748.3230 2.0399
MPA [23] 1.0323 2.5127 1.3689 1.2392 744.7016 2.0467
DEDCF [18] 1.0314 2.6380 47.5980 1.2356 821.6413 2.0529
DE [27] 1.0314 2.6380 1.3139 0.0343 22.8238 2.0529
NEPO [12] NA NA 48.4720 1.1720 982.4500 2.2000
FB-LLSEM [28] 1.0315 3.1436 1.3411 1.2181 858.4100 2.1321
EPO [12] 1.031 0.2090 48.4720 1.1720 982.450 2.2000
ODGB [26] 1.0305 3.4769 48.6369 1.2014 980.5942 2.4115
ABC-Ls [16] 1.0305 3.4742 48.6338 1.2016 984.1798 2.4250
OLGBO [25] 1.0305 3.48226 4.8642 1.20127 981.9830 2.4250
ADHHO [14] 1.0304 3.5062 48.6690 1.2007 999.4300 2.4252
GAMS [24] 1.0320 3.2681 1.3445 1.2062 828.2928 2.4426

6
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

that the algorithm predicts with certainty the experimental character­


istic: this is justified with the root means squared error presented in the
Table 1.
The corroboration between the measurement (pink) and the estimate
(green) demonstrates the algorithm’s accuracy.

Case study 2: PhotowattPWP PV module

This model is an 11.5 W panel, for which the current–voltage char­


acteristic data was taken during irradiance G = 1000 W/m2 and tem­
peratureT = 45◦ C. The manufacturer’s and characteristic data are
available in [14].
Table 2 shows the results of the best parameters obtained by the
MSNS- SEC algorithm. The RMSE obtained is 1.5411 × 10− 3. The gap
can be observed between the original SNS and the proposed MSNS- LAG.
It is obvious from these results that adding the Gauchy function in the
search space and the LaGrange algorithm in the objective function
provides better accuracy.
Fig. 6. Convergence curves.
In Table 2, a comparative study with other recently published
methods such as: [14,16,22,18,23–27] has been performed. On the base
of the RMSE, it is clear that the MSNS- SEC algorithm has the best op­
timum: with a value of 1.5411 × 10− 3, compared to 2.0399 and 2.4250

Fig. 7. Experimental platform.

7
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

Table 3
Results of the experimental platform of 18 PV array.
Operating condition Algorithms Best parameters
2
Irradiance (w/m ) Temperature ( C)

Iph(A) Io(µA) n Rs(Ω) Rp(Ω) Best RMSE

553 41.4 MSNS - SEC 10.0076 0.005976 5.99754 2.6917 360.6267 0.0254
ABC-TRR [29] 10.00 0.0057 215.87 2.699 368.2 0.0580
511 52.5 MSNS - SEC 9.2413 0.0098 6.1851 2.5401 412.1607 0.02213
ABC-TRR [29] 8.00124 0.01276 6.46768 2.54568 426.17692 0.0580
442 36.7 MSNS - SEC 8.0012 1.2604 6.3684 2.5506 426.4955 0.0166
ABC-TRR [29] 8.00 0.0099 230.29 2.568 419.8 0.0313
390 35.9 MSNS - SEC 7.0612 0.0036 6.0254 2.6665 487.4974 0.0128
ABC-TRR [29] 7.06 0.0057 225.77 2.630 515.5 0.0291
333 32.4 MSNS - SEC 6.0219 0.0024 6.1708 2.6608 561.7872 0.0097
ABC-TRR [29] 6.02 0.0034 225.75 2.634 580.2 0.0182
281 30.3 MSNS - SEC 5.0859 0.0022 6.2488 2.6531 603.3562 0.0069
ABC-TRR [29] 5.08 0.0032 228.83 2.620 621.8 0.0134

Fig. 8. I-V characteristic of PV array (GL100) at different temperature.

for the best recently proposed. It can also be noticed that the initial SNS accuracy of the MSNS- SEC algorithm in estimating the parameters when
algorithm presents a result of 2.4242 × 10− 3, which is much lower than the PVs are subjected to different temperature and irradiance
that obtained by the proposed method. conditions.
In Fig. 6, we have the convergence curves of the initial SNS algorithm The curves of the experimental measurements and the estimated
and the MSNS- SEC algorithm. curves of the PV array’s current–voltage characteristic are presented in
It can be observed in this figure that the convergence curve of the Fig. 8.
MSNS- SEC algorithm is the first to reach its minimum before the six We can also observe on the curves of Fig. 8 the corroboration be­
hundredth iteration: this demonstrates the speed of convergence of the tween the experimental and the estimated curves: this shows us the
algorithm. Obviously, the initial SNS algorithm crashes in a local min­ accuracy of the results obtained.
imum of 2.4242 × 10− 3 while MSNS- SEC reach its optimum of 1.5411
× 10− 3. Conclusion

This paper used a new approach based on the modified Social


Case study 3: 18 PV experimental field
Network Search (SNS) algorithm combined with the Secant method
(MSNS- SEC) to find the best internal parameters of a photovoltaic cell,
To confirm the validity of the MSNS- SEC algorithm, we applied it to
module, and array. The Social Network Search (SNS) algorithm simu­
extract the five model parameters to a diode of the PV field of the
lates the behavior of individuals in social networks in search of popu­
experimental platform of Fig. 7 at different temperatures and irradiance.
larity. For the modification of the SNS algorithm, three approaches were
Details of the parameters of this experimental field are available in
taken. Firstly, a control parameter via Gaussian and Cauchy distribution
[29]. The best extracted parameters and atmospheric conditions of the
was randomly added in the search space; secondly, a historical memory
PV field are given in Table 3.
function was added. When calling the objective function, the Secant
In this table, we can observe that the order of magnitude of the
method was used to increase the accuracy of the best current estimates.
RMSEs is lower than in the reference paper [29]; this demonstrates the

8
P.J. Gnetchejo et al. Energy Conversion and Management: X 15 (2022) 100273

The method was implemented on three different systems: a PV cell, a PV [10] Tossa AK, Soro YM, Azoumah Y, Yamegueu D. A new approach to estimate the
performance and energy productivity of photovoltaic modules in real operating
panel, and an 18 PV array. The different results obtained on the one
conditions. Sol Energy 2014;110:543–60. https://doi.org/10.1016/j.
hand, and the comparisons with other methods in the literature on the solener.2014.09.043.
other hand, show that the proposed method is far superior to all other [11] Ghani F, Fernandez EF, Almonacid F, O’Donovan TS. The numerical computation
methods in the literature with an RMSE of 6.8514 × 10− 4 for the PV cell, of lumped parameter values using the multi-dimensional Newton-Raphson method
for the characterisation of a multi-junction CPV module using the five-parameter
and 1.5411 × 10− 3 for the PV module. Furthermore, the convergence approach. Sol Energy 2017;149:302–13. https://doi.org/10.1016/j.
curves reaching the optimum before the five hundredth iteration also solener.2017.04.024.
demonstrate the algorithm’s speed; the results of the experimental PV [12] Chauhan A, Prakash S. A new emperor penguin optimisation-based approach for
solar photovoltaic parameter estimation. Int Trans Electr Energ Syst 2021;31(7).
array consisting of 18 panels also demonstrate the algorithm’s robust­ https://doi.org/10.1002/2050-7038.12917.
ness when panels are subjected to different environmental conditions. [13] Mengue FD, Kammogne AST, Siewe MS, Yamapi R, Fotsin HB. A new hybrid
The proposed method used in this paper, would help to optimal design a method based on differential evolution to determine the temperature-dependent
parameters of single-diode photovoltaic cells. J Comput Electron 2021;20(6):
PV cell, module and system, which is essential for the efficiency of the 2511–21. https://doi.org/10.1007/s10825-021-01785-6.
PV system. These parameter can also be used to detect or predict fault in [14] Song S, Wang P, Heidari AA, Zhao X, Chen H. Adaptive Harris hawks optimization
a PV system. with persistent trigonometric differences for photovoltaic model parameter
extraction. Eng Appl Artif Intell 2022;109. https://doi.org/10.1016/j.
engappai.2021.104608.
CRediT authorship contribution statement [15] AbdElminaam DS, Houssein EH, Said M, Oliva D, Nabil A. An Efficient Heap-Based
Optimizer for Parameters Identification of Modified Photovoltaic Models. Ain
Shams Eng J 2022;13(5). https://doi.org/10.1016/j.asej.2022.101728.
Patrick Juvet Gnetchejo: Conceptualization, Methodology, Soft­
[16] Tefek MF. Artificial bee colony algorithm based on a new local search approach for
ware, Writing – review & editing, Writing – original draft. Salomé parameter estimation of photovoltaic systems. J Comput Electron 2021;20(6):
Ndjakomo Essiane: Project administration, Supervision, Validation. 2530–62. https://doi.org/10.1007/s10825-021-01796-3.
Abdouramani Dadjé: Validation, Visualization, Investigation, Writing [17] Ibrahim IA, Hossain MJ, Duck BC. A hybrid wind driven-based fruit fly
optimization algorithm for identifying the parameters of a double-diode
– review & editing. Daniel Mbadjoun Wapet: Data curation, Writing – photovoltaic cell model considering degradation effects. Sustain Energy Technol
original draft, Formal analysis. Pierre Ele: Supervision, Validation. Assess 2022;50. https://doi.org/10.1016/j.seta.2021.101685.
[18] Parida SM, Rout PK. Differential evolution with dynamic control factors for
parameter estimation of photovoltaic models. J Comput Electron 2021;20(1):
Declaration of Competing Interest 330–43. https://doi.org/10.1007/s10825-020-01617-z.
[19] Gnetchejo PJ, Ndjakomo Essiane S, Ele P, Wamkeue R, Mbadjoun Wapet D, Perabi
The authors declare that they have no known competing financial Ngoffe S. Reply to comment on “Important notes on parameter estimation of solar
photovoltaic cell. Energy Convers Manage 2019;201. https://doi.org/10.1016/j.
interests or personal relationships that could have appeared to influence enconman.2019.112132.
the work reported in this paper. [20] Gnetchejo PJ, Ndjakomo Essiane S, Dadjé A, Ele P. A combination of Newton-
Raphson method and heuristics algorithms for parameter estimation in
photovoltaic modules. Heliyon 2021;7(4). https://doi.org/10.1016/j.
References heliyon.2021.e06673.
[21] Talatahari S, Bayzidi H, Saraee M. Social Network Search for Global Optimization.
[1] Gnetchejo PJ, Ndjakomo Essiane S, Ele P, Abdouramani D, Chen Z. Faults diagnosis IEEE Access 2021;9:92815–63. https://doi.org/10.1109/ACCESS.2021.3091495.
in a photovoltaic system based on multivariate statistical analysis. Energy Sources [22] Gnetchejo PJ, et al. A Self-adaptive Algorithm with Newton Raphson Method for
Part A 2021:1–22. https://doi.org/10.1080/15567036.2021.1919792. Parameters Identification of Photovoltaic Modules and Array. Trans Electr Electron
[2] Gnetchejo PJ, Ndjakomo Essiane S, Ele P. Diagnostic des défauts d’un champ Mater 2021;22(6):869–88. https://doi.org/10.1007/s42341-021-00312-5.
photovoltaïque par analyse statistique multivariée Fault detection in PV array [23] Sattar MAE, Al Sumaiti A, Ali H, Diab AAZ. Marine predators algorithm for
based on statistical analysis. The 1st International Conference on Local Resource parameters estimation of photovoltaic modules considering various weather
Exploitation. 2021. conditions. Neural Comput Appl 2021;33(18):11799–819. https://doi.org/
[3] Ndjakomo Essiane S, Gnetchejo PJ, Ele P, Chen Z. Faults detection and 10.1007/s00521-021-05822-0.
identification in PV array using kernel principal components analysis. Int J Energy [24] Gnetchejo PJ, Ndjakomo Essiane S, Ele P, Wamkeue R, Mbadjoun Wapet D, Perabi
Environ Eng 2021. https://doi.org/10.1007/s40095-021-00416-x. Ngoffe S. Important notes on parameter estimation of solar photovoltaic cell.
[4] Gnetchejo PJ, Essiane SN, Ele P, Wamkeue R, Mbadjoun Wapet D, Perabi Ngoffe S. Energy Convers Manage 2019;197. https://doi.org/10.1016/j.
Enhanced Vibrating Particles System Algorithm for Parameters Estimation of enconman.2019.111870.
Photovoltaic System. JPEE 2019;07(08):1–26. https://doi.org/10.4236/ [25] Yu S, Heidari AA, Liang G, Chen C, Chen H, Shao Q. Solar photovoltaic model
jpee.2019.78001. parameter estimation based on orthogonally-adapted gradient-based optimization.
[5] Wapet DEM, Essiane SN, Wamkeue R, Gnetchejo PJ. Hydropower Production Optik 2022;252. https://doi.org/10.1016/j.ijleo.2021.168513.
Optimization from Inflow: Case Study of Songloulou Hydroplant. JPEE 2020;08 [26] Premkumar M, Jangir P, Elavarasan RM, Sowmya R. Opposition decided gradient-
(08):37–52. https://doi.org/10.4236/jpee.2020.88003. based optimizer with balance analysis and diversity maintenance for parameter
[6] Mbadjoun Wapet DE, Ndjakomo Essiane S, Wamkeue R, Bisso D, Gnetchejo PJ. identification of solar photovoltaic models. J Ambient Intell Human Comput 2021.
Optimal management of hydropower production: Case of Memve’ele hydropower https://doi.org/10.1007/s12652-021-03564-4.
reservoir policy. Energy Rep 2022;8:1425–56. https://doi.org/10.1016/j. [27] Kharchouf Y, Herbazi R, Chahboun A. Parameter’s extraction of solar photovoltaic
egyr.2021.12.047. models using an improved differential evolution algorithm. Energy Convers
[7] Mbadjoun Wapet DE, Ndjakomo Essiane S, Wamkeue R, Bisso D, Gnetchejo PJ, Manage 2022;251. https://doi.org/10.1016/j.enconman.2021.114972.
Bajaj M. Optimal policy of hydroelectric reservoir integrated spill flow. J Appl [28] Sreekantha Reddy S, Yammani C. Parameter extraction of single-diode
Water Eng Res 2022:1–52. https://doi.org/10.1080/23249676.2022.2093794. photovoltaic module using experimental current–voltage data. Circuit Theor Apps
[8] Dehghanzadeh A, Farahani G, Maboodi M. A novel approximate explicit double- 2022;50(2):753–71. https://doi.org/10.1002/cta.3133.
diode model of solar cells for use in simulation studies. Renew Energy 2017;103: [29] Wu L, et al. Parameter extraction of photovoltaic models from measured I-V
468–77. https://doi.org/10.1016/j.renene.2016.11.051. characteristics curves using a hybrid trust-region reflective algorithm. Appl Energy
[9] Senturk A, Eke R. A new method to simulate photovoltaic performance of 2018;232:36–53. https://doi.org/10.1016/j.apenergy.2018.09.161.
crystalline silicon photovoltaic modules based on datasheet values. Renewable
Energy 2017;103:58–69. https://doi.org/10.1016/j.renene.2016.11.025.

View publication stats

You might also like