Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 12, No. 1, February 2022, pp. 82 91 ISSN: 2088-8708, DOI: 10.11591/ijece.v12i1.pp82-91 r 82 P arametric estimation in photo v oltaic modules using the cr o w sear ch algorithm Oscar Danilo Montoya 1,2 , Carlos Alberto Ram ´ ır ez-V anegas 3 , Luis F er nando Grisales-Nor e ˜ na 4 1 F acultad de Ingenier ´ ıa, Uni v ersidad Distrital Francisco Jos ´ e de Caldas, Bogot ´ a D.C., Colombia 2 Laboratorio Inteligente de Ener g ´ ıa, Uni v ersidad T ecnol ´ ogica de Bol ´ ıv ar , Cartagena, Colombia 3 F acultad de Ciencias B ´ asicas, Uni v ersidad T ecnol ´ ogica de Pereira, Pereira, Colombia 4 Departamento de Electromec ´ anica y Mecratr ´ onica, Instituto T ecnol ´ ogico Metropolitano, Medell ´ ın, Colombia Article Inf o Article history: Recei v ed Feb 1, 2021 Re vised May 20, 2021 Accepted Jun 16, 2021 K eyw ords: Cro w search algorithm Manuf acturer information Metaheuristic optimization P arametric estimation Photo v oltaic modules Single-diode model ABSTRA CT The problem of parametric estimation in photo v oltaic (PV) modules considering man- uf acturer information is addressed in this research from the perspecti v e of combinato- rial optimization. W ith the data sheet pro vided by the PV manuf acture r , a non-linear non-con v e x optimization problem is formulated tha t contains information re g arding maximum po wer , open-ci rcuit, and short-circuit points. T o estimate the three param- eters of the PV model (i.e., the ideality diode f actor ( a ) and the parallel and series resistances ( R p and R s )), the cro w search algorithm (CSA ) is emplo yed, which is a metaheuristic optimi zation technique inspired by the beha vior of the cro ws searching food deposits. The CSA allo ws the e xploration and e xploitation of the solution space through a simple e v olution rule deri v ed from the classical PSO method. Numerical simulations re v eal the ef fecti v eness and rob ustness of the CSA to est imate these pa- rameters with objecti v e function v alues lo wer t han 1 10 28 and processing times less than 2 s. All the numerical simulations were de v eloped in MA TLAB 2020 a and compared with the sine-cosine and v orte x search algorithms recently reported in the literature. This is an open access article under the CC BY -SA license . Corresponding A uthor: Oscar Danilo Monto ya F acultad de Ingenier ´ ıa, Pro yecto Curricular de Ingenier ´ ıa El ´ ectrica, Uni v ersidad Distrital Francisco Jos ´ e de Caldas Cra 7 # 40B-53, Bogot ´ a D.C., Colombia Email: odmonto yag@udistrital.edu.co 1. INTR ODUCTION The presence of photo v oltaic (PV) sources has increased rapidly in the past tw o decades in lo w , medium and high-v oltage le v els, and their accelerated de v elopment has decreased their production, mainte- nance, and operati v e costs [1]-[3]. Moreo v er , these rene w able ener gy resources h a v e reduced the ener gy pur - chase costs in urban areas and greenhouse g as emiss ions in rural netw orks po wered by diesel generators [4]. The inte gration of these PV sources into electrical grids generally requires the po wer electronic con v erters to manage their ener gy production to maximize the producer’ s profit [5]. This ener gy management is achie v ed through linear and non-linear control strate gies applied to find and maintain the operation of the PV module in the maximum po wer point (MPP) [6]-[8]. In the literature, dif ferent models are utilized to represent the PV modules, which are composed of one, tw o, or three diodes. Each one of them has multiple parameters that ha v e to be found prior to determining the panel beha vior re g arding current, v oltage, and po wer outputs [2], [9]. The most accepted model to represent the PV module is the single-diode representation. In this model, we need J ournal homepage: http://ijece .iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 83 to find three parameters of the PV module that are associated with the ideality diode f actor and the series and parallel resistances [10]. T o find these parameters, in the current literat ure, se v eral optimization techniques are a v ailable that use the data sheet information pro vided by the panel manuf acturer , where three main operati v e points can be highlighted: i) open-circuit point, ii) short-circuit point, and iii) MPP . These points formulate a non-linear non-con v e x optimization problem to determine the best combination of the model parameters to represent the complete electrical beha vior of the PV module [11], [12]. In the literature, methods such as the adapti v e dif ferential e v olution algorithm [5], the sine-cosi ne algorithm [10], the analytical method [13], the genetic algorithm [14], the cuck oo search algorithm [15], the least square method [16], the electrom agnetic-lik e algori thm [17], and the grasshopper optimization algorithm [18], [19], among others, can be found. It should be noted that this re vision demonstrates that the non-linear , non-con v e x nature of the parametric estimation problem mak es the appli cation of po werful optimization tech- niques necessary to find the optimal parameters that ensure the correct operation of the PV modules’ equi v alent electric circuit. Based on this state-of-the-art re vision, we propose a ne w optimization algorithm to estimate the elec- trical parameters of the PV module using the single-diode model representation. The proposed algorithm corresponds to the cro w search algorithm (CSA), which has not been pre viously applied to this problem using the PV data sheet with the main adv antage that only four parameters can be tuned. Moreo v er , numerical results demonstrate the v alues of objecti v e functions that are lo wer than 1 10 28 , which are clearly better than the results reported in [5] and [10], where the v alues of objecti v e functions were 1 10 12 and 1 10 15 , respec- ti v ely . An additional adv antage of the proposed approach is that it reaches the optimal solution in less than 2 s by ensuring optimal funding through the non-parametric W ilcoxon test. It is important to mention that the CSA has pre viously been reported in [2] to determine the parameters of the PV modules. Ho we v er , the authors of [2] focus their study on an optimization model that tak es into account only the po wer tracking error , considering v ariations in the temperature and irradiance inputs. It dif fers from our proposal since we are w orking directly with the manuf acturer nameplate in which three operati v e points are considered to determine the general model of the PV system, which correspond to short-circuit point, open-circuit point, and MPP . The remainder of this paper is or g anized as follo ws. Section 2 presents the formulation of the para- metric estimation problem in PV modules considering the data sheet information pro vided by the manuf acturer . Section 3 presents the general description of the proposed CSA. Section 4 presents the primary characteristics of the test system and the computational v alidation features. Section 5 sho ws the numerical v alidation of the proposed methodology and its analysis and discussion. Finally , section 6 presents the main concluding remarks deri v ed from this research as well as some possible future w orks. 2. OPTIMIZA TION MODEL The nomenclature of the mathematical optimization model presented for the parametric estimation in PV modules using a single-diode model has been listed belo w: V ariables and parameters I pv : Define the photoelectric current a : Ideality f actor of the diode I 0 : In v erse saturation current in the PV module N c : Number of PV cells connected in series I mpp : Maximum po wer point current a min : Lo wer bound of the ideality diode f actor I sc : Short-circuit current a max : Upper bound of the ideality diode f actor V mpp : Maximum po wer point v oltage R min p : Lo wer bound of the parallel resistance V oc : Open-circuit v oltage R max p : Upper bound of the parallel resistance R s : Series equi v alent resistance R min s : Lo wer bound of the series resistance R p : P arallel equi v alent resistance R max s : Upper bound of the series resistance q : Electron char ge (e.g., 1 : 60217646 10 19 Coulomb) T : Absolute temperature in the diode union ( 273 : 15 + 25 K elvin) k : Boltzmann constant ( 1 : 38064852 10 23 Joules/K elvin) The parametric estimation problem in PV modules considering N c cells connected in series is de- v eloped based on their ideal model using a single-diode representation [10]. The schematic modeling of the P ar ametric estimation in photo voltaic modules using the cr ow sear c h algorithm (Oscar Danilo Montoya) Evaluation Warning : The document was created with Spire.PDF for Python.
84 r ISSN: 2088-8708 single-diode model, that is, the electrical circuit equi v alent, has been presented in Figure 1. Figure 1. Equi v alent circuit of a photo v oltaic module T o analyze PV modules with the electrical circuit equi v alent depicted in Figure 1, the e xponential relation is used to define the output v oltage and current as in (1) [20], [21]. I = I pv I 0 exp q V + R s I ak N c T 1 V + R s I R p : (1) W ith the objecti v e to determine all the parameters of the PV module, that is, the series and parallel resistances and the diode ideality f actor , the three main operati v e points pro vided by the module manuf acturer ha v e been considered. These operati v e points include: i) open-circuit point, ii ) short-circuit point, and iii ) MPP . The analysis of each one of these points has been presented belo w . 2.1. Open-cir cuit operati v e point The open-circuit operati v e point of the PV module presented in Figure 1 implies that the v oltage in its terminals is V = V oc with a null current flo w through them, that is, I = 0 . W ith this operati v e condition, it is possible to obtain an e xpression for I pv from (1) as (2): I pv = I 0 exp q V oc ak N c T 1 V oc R p : (2) 2.2. Short-cir cuit operati v e point The second operati v e point pro vided by the PV m odu l e manuf acturer corresponds to the s hort-circuit scenario at the terminals of the module, which implies that I = I sc and V = 0 . W ith these operati v e conditions, in (1) assumes in (2): I sc = I pv I 0 exp q R s I sc ak N c T 1 R s I sc R p : (3) No w , if we combine (2) and (3), and some algebraic manipulations are made, the follo wing equation for the in v erse saturation current is deri v ed: I 0 = I sc + R s I sc R p + V oc R p exp q V oc ak N c T exp q R s I sc ak N c T : (4) It should be noted that if we substitute (4) in (2), we can obtain a general representation of the PV current as (5): I pv = I sc + R s I sc R p + V oc R p h exp q V oc ak N c T 1 i exp q V oc ak N c T exp q R s I sc ak N c T V oc R p : (5) 2.3. Maximum po wer point In the information pro vided by the PV module manuf acturer , an operational point named MPP is a v ailable. This point presents the information re g arding the maximum possible po wer transferred from the panel to the system with which this is interconnected, that is, ( I mpp ; V mpp ) . If this point is substituted in (1), a general equation for I mpp is found as (6): I mpp =   I pv I 0 h exp q V mpp + R s I ak N c T 1 i V mpp + R s I mpp R p : ! : (6) Int J Elec & Comp Eng, V ol. 12, No. 1, February 2022: 82–91 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 85 2.4. Optimization model Based on the information pro vided by the PV module manuf acturer , that is, the three aforement ioned operati v e points, it is possible to formulate an optimization model that allo ws the estimation of the parameters of the electrical equi v alent circuit, that is, series and parallel resistances and t he ideality diode f actor using a single-objecti v e formulation that minimizes the mean square error between the manuf acturer data and the calculated v alues. The general structure of the objecti v e function assumes the follo wing form: min z = E 2 oc + E 2 sc + E 2 mpp ; (7) where E oc = I 0 exp q V oc ak N c T 1 V R p I pv ; (8) E sc = I pv I 0 exp q R s I sc ak N c T 1 R s I sc R p I sc ; (9) E mpp =   I pv I 0 h exp q V mpp + R s I ak N c T 1 i V mpp + R s I mpp R p ! I mpp : (10) It should be noted that to find the v alue of the objecti v e function defi n e d in (7), it is necessary to kno w the v alues of the parameters a , R s , and R p (decision v ariables) in conjunction with the simultaneous solution of the (4) and (5) for the in v erse saturation and the PV current. T o complete the optimization model for parametric estimation in PV modules considering manuf acturer data, we assign the lo wer and upper bounds for the decision v ariables as in (11) [10]: a min a a max ; R min p R p R max p ; R min s R s R max s : (11) It should be observ ed that the propos ed optimization model defined from (7) to (11) added with the equality constraints (4) and (5) corresponds to a continuous non-linear non-con v e x op t imization problem. It implies that multiple solutions for the parameters a , R s , and R p can e xist with the same numerical perfor - mance, that is, multimodal optimization beha vior [22]. Due to this reason, to reach an adequate solution with minimal computational ef fort, t h i s paper proposes the application of the methaeuristic optimization technique kno wn as CSA [23], [24], which has not been pre viously reported in the literature to address the parametric es- timation probl em in P V modules. This is the main contrib ution of this research based on its e xcellent numerical performance. In the ne xt section, we will present the CSA and its application of the studied problem. 3. CR O W SEARCH ALGORITHM The CSA is a recently de v eloped combinatorial optimization technique to solv e continuous non-linear comple x and non-con v e x optimizati on problems with multiple constraints [25], [26]. This technique belongs to the f amily of the bio-inspired optimization deri v ed from the con v entional particle sw arm optimization (PSO) methodology . It is commonly kno wn that cro ws observ e other birds to kno w where the y hide their food in order to steal it once the o wner lea v es. In the case a cro w commits thie v ery , this cro w will tak e additional precautions such as mo ving hiding places to reduce the possibility of being a future victim [27]. In f act, the y use their o wn e xperience of ha ving been a thief to forecast the beha vior of possible thie v es and can determine the safest course to protect their caches from being pilfered. The main characteristics of the cro ws are: i) the y li v e in flocks; ii) the y can memorize the position of their hiding places; iii) the y can follo w each other to commit thie v ery; and i v) the y can protect their caches from being pilfered by a probability . In this paper , we present the mathematical adaptation of this beha vior to solv e comple x optimization problems as originally proposed in [23]. The primary steps in the implementation of the CSA ha v e been discussed belo w . 3.1. Initialization of the pr oblem and selection of the adjustable parameters The optimization problem is defined, that is, the optimization model (7) to (11) is added with the equality constraints (4) and (5). Then, the adjustable parameters of the CSA are s elected , that is, the flock size n , the maximum number of iterations t max , the flight length f l , and the a w areness probability A p . It should be P ar ametric estimation in photo voltaic modules using the cr ow sear c h algorithm (Oscar Danilo Montoya) Evaluation Warning : The document was created with Spire.PDF for Python.
86 r ISSN: 2088-8708 noted that the selection of these parameters is heuristic and depends on the kno wledge that the programmer has about the optimization problem under study . These are typically selected using multiple simulations to identify the best trade-of f between response quality and processing times. 3.2. Initial position and memory of the cr o ws In a d -dimensional s pace, all the n cro ws are initially positioned as the members of the flock. It is w orth mentioning that each cro w represents a feasible solution of the optimization problem that is composed by d decision v ariables. Cro ws = 2 6 6 6 4 x 11 x 12 x 1 d x 21 x 22 x 2 d . . . . . . . . . . . . x n 1 x n 2 x nd 3 7 7 7 5 (12) It should be noted that the generation of each component of the cro w i associated with the v ariable j tak es the follo wing form: x ij = x min j + r x max j x min j ; 8 i = 1 ; 2 ; :::; n j = 1 ; 2 ; :::; d (13) where, r is a random number between 0 and 1 generated with a normal distrib ution. In the be ginning of the search process, t he memory of each cro w is initialized. Ho we v er , in the i nitial iteration, the cro ws ha v e no search e xperience. Then, it is assumed that the y ha v e hidden their foods at their initial positions. Memory = 2 6 6 6 4 m 11 m 12 m 1 d m 21 m 22 m 2 d . . . . . . . . . . . . m n 1 m n 2 m nd 3 7 7 7 5 (14) 3.3. Fitness function e v aluation As a con v entional metaheuristic optimization algorithm, the CSA w orks with a fitness function instead of the original objecti v e function of the problem to deal with infeasibilities in the solution space [28]. Ho we v er , in the case of the parametric estimation of PV modules, due to the structure of the optimization model , it is possible to directly e v aluate the objecti v e function for each cro w . 3.4. Generation of the new position f or each cr o w The generation of ne w positions for the cro ws in the search space proceeds as follo ws: consider that the cro w i w ants to pass to a ne w position. F or this goal, this cro w randomly selects one of the flock cro ws (e.g., cro w k ) to follo w in order to disco v er the position where the y ha v e hidden their food (i.e., m k ). F ollo wing this procedure, the ne w position of the cro w i is obtained as (16): x t +1 i = x t i + r i f i;t l ( m t k x t i ) ; r k A k ;t p a random position (see (13)) ; otherwise (15) where r j is a random number with uniform distrib ution between 0 and 1, and A k ;t p denotes the a w areness probability of cro w k at iteration t . It should be observ ed that this procedure is repeated for all the cro ws. It is w orth mentioning that each cro w generated with (15) is re vised if the lo wer and upper bounds of the decision v ariables are fulfilled. In the case that the cro w i violates these bounds, in (13) is used to correct it. 3.5. Ev aluation of the objecti v e function and updating of the memories F or each one of the positions of the cro ws, the objecti v e function is e v aluated, which is used to update their memories as (16): m t +1 i = x t +1 i ; z x t +1 i is better than z ( x t i ) m t i ; otherwise (16) It is observ ed that if the fitness function v alue of the ne w position of the cro w i is better than the fitness function v alue of the memorized position, the cro w updates its memory by the ne w position. ”Better” in the case of parametric estimation in PV modules implies ”lo wer”. Int J Elec & Comp Eng, V ol. 12, No. 1, February 2022: 82–91 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 87 3.6. Check termination criterion The steps defined in subsections 3.4. and 3.5. are repeated until the maximum number of iter ations is reached (i.e., t max ). When the termination criterion is met, the best position of the memory in terms of the objecti v e function v alue (i.e., the minimum v alue) is reported as the solution of the optimization problem. 3.7. Implementation of the CSA Algorithm 1 presents the general pseudo-code that defines the steps necessary to implement the CSA to solv e the problem of parametric estimation in PV modules [23]. Algorithm 1: Application of the cro w search algorithm to the problem of the parametric estimation in PV modules Data: Read the data of the PV module pro vided by the manuf acturer; Define the algorithm parameters; Randomly initialize the position of a flock of n cro ws in the search space; Ev aluate the fitness function for all the cro ws; Initialize the memory of each cro w; f or t = 1 : t max do f or i = 1 : n do Randomly select one of the cro ws to follo w (e.g., cro w k ); Define an a w areness probability; if r k A k ;t p then x t +1 i = x t i + r i f i;t l ( m t k x t i ) ; else x t +1 i = a random position (apply Eq. (13)); end Check the feasibility of ne w positions; Ev aluate the ne w position of the cro ws; Update the memory of cro ws; end end Result: Return the best solution stored in the memory of the cro ws 4. TEST SYSTEM AND COMPUT A TION AL V ALID A TION The implementation of the proposed CSA to the problem of the parametric estimation in PV m o dul es represented with its single-diode model has considered the information pro vided by the manuf acturer of the K y- ocera KC200GT [10] as sho wn in T able 1, in the softw are MA TLAB 2020 a using a desk computer INTEL(R) Core(TM) i 5 3550 3 : 5 -GHz, 8 GB of RAM v ersion 64-bit with Microsoft W indo ws 7 Professional. T o v alidate the ef fecti v eness of the CSA to find a high-quali ty solution to the problem of the para- metric estimation in PV modules, a population of 20 indi viduals w as considered, that is, n = 20 , with 100000 consecuti v e iterations; an a w arenes s probability fix ed as 0 : 75 , and a v ariable fly length defined by the follo w- ing rule f l = 2 r and (1 t=t max ) . Moreo v er , the lo wer and upper bounds for the decision v ariables are 0 : 5 a 2 , 0 : 001 R p 1 , and 50 R s 200 , which ha v e been tak en from [10]. T able 1. K yocera KC200GT manuf acturer information (T ak en from [5]) P arameter Symbol V alue Open-circuit v oltage V oc 32.900 V T emperature coef ficient for V oc K V oc -0.123 V/ o C Short-circuit current I sc 8.210 A T emperature coef ficient for I sc K I sc 3.180 10 3 A/ o C V oltage on the MPP V mpp 26.300 V Number of cell in series N c 54 Current on the MPP I mpp 7.610 A P ar ametric estimation in photo voltaic modules using the cr ow sear c h algorithm (Oscar Danilo Montoya) Evaluation Warning : The document was created with Spire.PDF for Python.
88 r ISSN: 2088-8708 5. NUMERICAL RESUL TS The application of the CSA to the problem of parametric estimation in PV modules produced the results reported in T able 2, where the best 10 solutions are presented after 100 consecuti v e e v aluations. From results in T able 2, we can observ e that: i) the v alue of the objecti v e function related with the mean square error (see (7)) that e v aluates the error re g arding the open-circuit point, short-circuit point, and MPP pro vided by the manuf acturer of the PV module and the c alculated v alues using the single-diode model are lo wer (i.e., better) than 1 10 29 , which can be considered null for an y practical implementation. In this conte xt, as mentioned in [5], all the parameters represent optimal solutions, moreo v er , these impro v e the concl u s ion reported in [10] wherein v alues lo wer than 15 were considered optimal; ii) the solutions in the range from 3 to 8 present the same objecti v e function v alue, that is, 7 : 8886 10 31 , which confirm the multimodal nature of the problem of the parametric estimation in PV modules since there are dif ferent combinations of the decision v ariables that ha v e the same numerical performance; iii) the electrical parameter that presents more v ariations along the optimal solutions is the parallel resistance since the minimum v alue reached for this parameters is found in the solution 10 with a v alue of 55 : 0001 and the maximum v alue is found in the solution 5 with a v alue of 188 : 3342 , that is, a dif ference superi o r than 120 between both solutions; and i v) the a v erage processing times reported by the CSA to find the numerical results reported in T able 2 w as about 1 : 80 s with a standard de viation of 0 : 20 s, which demonstrates the ef ficienc y of the CSA to find the global optimal solution. In Figure 2 is presented the V I curv e of the PV module for each one of the ten solutions reached by the CSA and presented in T able 2. These curv es were obtained making a sweep in the v oltage v ariable from 0 to V oc in steps of 0 : 10 V by solving the (1) for all the combinations o f a , R s and R p parameters of the single-diode model of the PV module presented in Figure 1. From the numerical results presented in Figure 1, it can be noted that the points P 1 , P 2 , and P 3 correspond to the open-circuit point, MPP , and short-circuit operati v e points, which confirms that the information pro vided by the PV module manuf acturer is suf ficient to estimate with minimum errors the complete beha vior of the panel in all its operati v e range; moreo v er , when the W ilcoxon test w as applied for ten independent samples, each one of them with 10 optimal solutions, a mean v alue for p of about p of 0 : 5486 with a v alue of h = 0 w as obtained; this implies that the null h ypothesis of the W ilcoxon test is confirmed, and therefore, the analyzed samples present the same median wit h a significance le v el of 100 %, which demonstrates that the CSA has the ability to find the global optimal solution at each e v aluations with 100 consecuti v e search through the solution space. T able 2. T en best results reached by the CSA N o a R s ( ) R p ( ) f f 1 0.65018 0.3938 56.8600 0 2 0.57158 0.5199 81.3492 0 3 0.59211 0.4996 76.4901 7.8886 10 31 4 0.64397 0.5071 113.2315 7.8886 10 31 5 0.50549 0.6038 188.3341 7.8886 10 31 6 0.51566 0.5436 74.2629 7.8886 10 31 7 0.58256 0.5137 81.6568 7.8886 10 31 8 0.60683 0.5451 159.1443 7.8886 10 31 9 0.67940 0.3646 55.8052 2.2877 10 29 10 0.67623 0.3605 55.0001 2.5243 10 29 5.1. Comparison with combinatorial methods In this section, we present the comparati v e results among the proposed CSA and dif ferent combina- torial methods that solv e optimization problems in the continuous domain. These methods are: v orte x search algorithm [12], sine-cosine algorithm [10], PSO [29], and genetic algorithm [30]. F or each one of these com- parati v e methods, 100 consecuti v e e v aluations were made to obtain the best 10 results reported in T able 3. From results listed in T able 3 it is possible to observ e that: The CSA and the VSA optimization approaches present the best numerical performance with respect to the objecti v e function v alues lo wer than 10 28 and 10 25 , respecti v ely . The solutions found by the SCA and the CGA methods can also be considered optimal, since the objecti v e function is in practical terms null. Ho we v er , we can mention that the SCA presents a better numerical performance when compared with the CGA as w as demonstrated in [10], since the best objecti v e function is 9 : 7192 10 28 for the SCA and 3 : 2207 10 12 for the CGA, which implies a dif ference higher than 1 10 5 between them. Int J Elec & Comp Eng, V ol. 12, No. 1, February 2022: 82–91 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 89 The list of dif ferent solutions reported in 3 confirms the non-linear non-con v e x nature of the problem of the parametric estimation in PV systems, since the CGA and the SCA are stuck in local optimums, while the PSO, the VSA, and the proposed CSA present a better numerical performance reaching high-quality objecti v e function v alues. T o complement the analysis among the CSA and the comparati v e metaheuristic m ethods, the parameters for the optimal solutions are listed in T able 4 with label number 1 presented in T able 3. Figure 2. V I curv e obtained by solving (1) for the parameters presented in T able 2 T able 3. T en best solutions reported by each comparati v e method N o CSA VSA SCA PSO CGA 1 0 0 9.7192 10 18 2.5243 10 29 3.2207 10 12 2 0 0 1.0823 10 17 1.0097 10 28 5.1049 10 11 3 7.8886 10 31 0 1.3799 10 16 1.6155 10 27 2.3014 10 10 4 7.8886 10 31 0 1.9364 10 16 2.5243 10 27 2.7546 10 10 5 7.8886 10 31 7.8886 10 29 3.5309 10 16 6.0647 10 27 3.9197 10 10 6 7.8886 10 31 1.9248 10 28 5.2122 10 16 2.9181 10 26 5.1134 10 10 7 7.8886 10 31 1.3067 10 26 5.4738 10 16 3.5498 10 26 5.2351 10 10 8 7.8886 10 31 1.3354 10 26 8.4393 10 16 5.0553 10 26 6.0393 10 10 9 2.2877 10 29 2.9614 10 26 9.3922 10 16 7.6361 10 26 6.2646 10 10 10 2.5243 10 29 4.8872 10 26 1.5123 10 16 1.4485 10 25 6.8811 10 10 T able 4. Optimal solutions reported by the proposed and comparati v e methods Method a R s ( ) R p ( ) CSA 0.650181877806710 0.393806884684579 56.8600309871423 VSA 0.502572297672421 0.505917172241395 57.6920080257133 PSO 0.681740933460645 0.508343460036893 170.884395321487 SCA 0.917140758347724 0.146864384999908 52.6718647012995 CGA 0.985461664181760 0.234102605899478 70.5159926098178 6. CONCLUSION In this research, the CSA w as implemented to find the optimal parameter combination to represent PV modules with its single-diode model. Numerical results demonstrated that this algorithm finds solutions with v alues lo wer than 1 10 28 re g arding the objecti v e function v alue after 100 consecuti v e e v aluati ons, which were better in comparison with the classical metaheuristic methods used to solv e this problem; these methods were the VSA, SCA, PSO, and CGA respecti v ely . The first 10 solutions reached by the CSA confirm that the problem of the parametric estimation in PV modules is a multimodal non-linear optimization problem with dif ferent combinations of the decision v ariables that present the same numerical performance. Re g arding the processing times, the proposed CSA tak es about 1 : 80 s to find the optimal solution of the studied problem with the main adv antage being that based on t he W ilcoxon test, after 100 consecuti v e e v aluations, the possibility P ar ametric estimation in photo voltaic modules using the cr ow sear c h algorithm (Oscar Danilo Montoya) Evaluation Warning : The document was created with Spire.PDF for Python.
90 r ISSN: 2088-8708 of finding the global optimum is ensured. Moreo v er , the CSA is easily implementable in an y programming language with only 4 parameters to be tuned. In the future, it will be possible to de v elop the follo wing re- search w orks: i) to e xte n d the proposed CSA to the parametric estimation in induction motors and distrib ution transformers that are modeled with non-linear non-con v e x optimization models and ii) to apply the proposed optimization model to the estimation of parameters in PV modules considering real measures of v oltages and currents including v ariable weather conditions. A CKNO WLEDGMENTS This w ork w as supported in part by the Centro de In v estig aci ´ on and Desarrollo Cient ´ ıfico de la Uni- v ersidad Distrital Francisco Jos ´ e de Caldas under grant 1643-12-2020 associated with the project: “Desarrollo de una metodolog ´ ıa de optimizaci ´ on para la gesti ´ on ´ optima de recursos ener g ´ eticos distrib uidos en redes de dis- trib uci ´ on de ener g ´ ıa el ´ ectrica. and in part by the Direcci ´ on de In v estig aciones de la Uni v ersidad T ecnol ´ ogica de Bol ´ ıv ar under grant PS2020002 associated with the project: “Ubicaci ´ on ´ optima de bancos de capacitores de paso fijo en redes el ´ ectricas de distrib uci ´ on para reducci ´ on de costos and p ´ erdidas de ener g ´ ıa: Aplicaci ´ on de m ´ etodos e xactos and metaheur ´ ısticos. REFERENCES [1] M . S. Cengiz and M. S. Mamis ¸ , “Price-ef ficienc y relationship for photo v oltaic systems on a global basis, International Journal of Photoener gy , v ol. 2015, pp. 1–12, 2015, doi: 10.1155/2015/256101. [2] A. Omar , H. M. Hasanien, M. A. Elgendy , and M. A. L. Badr , “Identification of the photo v oltaic model parameters us ing the cro w search algorithm, The Journal of Engineering, v ol. 2017, no. 13, pp. 1570–1575, Jan. 2017. [3] A. P az-Rodrıguez, J. F . Castro-Ordo˜nez, O. D. Monto ya, and D. A. Giral-Ramırez, “Optimal inte gration of photo v oltaic sources in distrib ution netw orks for daily ener gy losses minimization using the v orte x search algorithm, Applied Sciences, v ol. 11, no. 10, May . 2021, Art. No. 4418, doi: 10.3390/app11104418. [4] O. D. Monto ya, L. F . Grisales-Nore˜na, W . Gil-Gonzalez, G. Alcala, and Q. Hernandez-Escobedo, “Optimal location and sizing of PV sources in DC netw or ks for minimizing greenhouse emissions in diesel generators, Symmetry , v ol. 12, no. 2, Feb . 2020, Art. No. 322, doi: 10.3390/sym12020322. [5] P . P . Bisw as, P . Sug anthan, G. W u, and G. A. Amaratung a, “P arameter estimation of solar cells using datasheet information with the application of an adapti v e dif ferential e v olution algorithm, Rene w able Ener gy , v ol. 132, pp. 425-438, 2019, doi: 10.1016/j.renene.2018.07.152. [6] D. G. Monto ya, C. A. R. P aja, and R. Giral, “Maximum po wer point tracking of photo v oltaic systems based on the sliding mode control of the module admittance, Electr . Po wer Syst. Res., v ol. 136, pp. 125–134, Jul. 2016, doi: 10.1016/j.epsr .2016.02.001. [7] M. Al-Soeidat, D. D.-C. Lu, and J. Zhu, An analog BJT -tuned maximum po wer point tracking technique for PV systems, IEEE T ransactions on Circuits and Systems II: Express Briefs, v ol. 66, no. 4, pp. 637–641, Apr . 2019, doi: 10.1109/TCSII.2018.2865804. [8] M. Arjun, V . V . Ramana, R. V isw ade v , and B. V enkatesaperumal, An iterati v e analytical solution for calculating maximum po wer point in photo v oltaic systems under partial shading conditions, IEEE T ransactions on Circuits and Systems II: Express B riefs, v ol. 66, no. 6, pp. 973–977, Jun. 2019, doi: 10.1109/TCSII.2018.2867088. [9] A. R. Jordehi, “P arameter estimation of solar photo v oltaic (PV) cells: A re vie w , Rene w able Sustainable Ener gy Re v ., v ol. 61, pp. 354–371, Aug. 2016, doi: 10.1016/j.rser .2016.03.049. [10] O. D. Monto ya, W . Gil-Gonzalez, and L. F . Grisales-Nore˜na, “Sine-cosine algorithm for parameters’ estimation in solar cells using datasheet information, Journal of Ph ysics: Conference Series, v ol. 1671, Oct. 2020, Art. No. 012008. [11] U. Jadli, P . Thakur , and R. D. Shukla, A ne w parameter estimation method of solar photo v oltaic, IEEE Journal of Photo v oltaics, v ol. 8, no. 1, pp. 239–247, 2018, doi: 10.1109/JPHO T O V .2017.2767602. [12] W . J. Gil-Gonzalez, O. D. Monto ya, and J. M. Lopez-Lezama, “V orte x sear ch algorithm applied to the parametric estimation in pv cells considering manuf acturer datasheet information, IEEE Latin America T ransactions, v ol. 19, no. 9, pp. 1581–1589, Sep. 2021, doi: 10.1109/TLA.2021.9468612. [13] A. Elkholy and A. A. El-Ela, “Optimal parameters estimation and modelling of photo v oltaic modules using analytical method, Heliyon, v ol. 5, no. 7, p. e02137, Jul. 2019, doi: 10.1016/j.heliyon.2019.e02137. [14] M. A. A w adallah and B. V enkatesh, “Estimation of PV module parameters from datasheet information using optimization techniques, in 2015 IEEE International C onference on Industrial T echnology (ICIT), Mar . 2015, doi: 10.1109/ICIT .2015.7125507. [15] J. Ma, T . O. T ing, K. L. Man, N. Zhang, S.-U. Guan, and P . W . H. W ong, “P arameter estimation of photo v oltaic models via cuck oo search, J. Appl. Math., v ol. 2013, pp. 1–8, 2013, doi: 10.1155/2013/362619. [16] N. Maouhoub, “Photo v oltaic module parameter estimation using an analytical approach and least squares method, J. Comput. Electron., v ol. 17, no. 2, pp. 784–790, Jan. 2018, doi: 10.1007/s10825-017-1121-5. [17] H. M. Ridha, C. Gomes, and H. Hizam, “Estimation of photo v oltaic module model’ s parameters using an impro v ed electromagnetic- lik e algorithm, Neural Computing and Applications, v ol. 32, no. 16, pp. 12627–12642, Jan. 2020, doi: 10.1007/s00521-020-04714- z. [18] O. S. Elazab, H. M. Hasanien, I. Alsaidan, A. Y . Abdelaziz, and S. M. Muyeen, “P arameter estimation of three diode photo v oltaic model using grasshopper optimization algorithm , Ener gies, v ol. 13, no. 2, Jan. 2020, Art. No. 497, doi: 10.3390/en13020497. Int J Elec & Comp Eng, V ol. 12, No. 1, February 2022: 82–91 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 91 [19] J. Montano, A. F . T obon, J. P . V ille g as, and M. Durango, “Grasshopper optimizat ion algorithm for parameter estimation of photo- v oltaic modules based on the single diode model, International Journal of Ener gy and En vironmental Engineering, v ol. 11, no. 3, pp. 367–375, Feb . 2020, doi: 10.1007/s40095-020-00342-4. [20] A. K. Abdulrazzaq, G. Bognar , and B. Plesz, Accurate method for PV solar cells and modules pa- rameters e xtraction using i–v curv es, Journal of King Saud Uni v ersity-Engineering Sciences, Jul. 2020, doi: 10.1016/j.jksues.2020.07.008. [21] R. Nde gw a, J. Simiyu, E. A yieta, and N. Odero, A f ast and accurate analytical method for parameter determination of a photo v oltaic system based on manuf acturer’ s data, Journal of Rene w able Ener gy , v ol. 2020, pp. 1–18, May . 2020, doi: 10.1155/2020/7580279. [22] R. Chelouah and P . Siarry , “Continuous genetic algorithm designed for the global optimization of multimodal functions, Journal of Heuristics, v ol. 6, no. 2, pp. 191–213, 2000, doi: 10.1023/A:1009626110229. [23] A. Askarzadeh, A no v el metaheuristic method for solving constrained engineering optimization prob- lems: Cro w search algo- rithm, Computers and Structures, v ol. 169, pp. 1–12, Jun. 2016. [24] Y . Meraihi, A. B. Gabis, A. Ramdane-Cherif, and D. Acheli, A comprehensi v e surv e y of Cro w Search Algorithm and its applica- tions, Artificial Intelligence Re vie w , Sep. 2020. [25] B. Zolghadr -Asli, O. Bozor g-Haddad, and X. Chu, “Cro w search algorithm (CSA), in Adv anced Optimization by Nature-Inspired Algorithms, Springer Sing apore, Jul. 2017, pp. 143–149, doi: 10.1007/978-981-10-5221-7-14. [26] A. A. E. Ela, R. A. El-Sehiemy , A. Shaheen, and A. Shalaby , Application of the cro w search algorithm for economic en vironmental dispatch, in 2017 Nineteenth International Middle East Po wer Systems Conference (MEPCON), Dec. 2017, doi: 10.1109/MEP- CON.2017.8301166. [27] S. Shekha w at and A. Sax ena, “De v elopment and applications of an intelligent cro w search algorithm based on opposition based learning, ISA T ransactions, v ol. 99, pp. 210–230, Apr . 2020, doi: 10.1016/j.isatra.2019.09.004. [28] O. D. Monto ya, W . Gil-Gonzalez, and L. F . Grisales-Norena, “V orte x search algorithm for optimal po wer flo w analysis in DC resisti v e netw orks with CPLs, IEEE T ransactions on Circuits and Systems II: Express Briefs, v ol. 67, no. 8, pp. 1439–1443, Aug. 2020, doi: 10.1109/TCSII.2019.2938530. [29] V . Khanna, B. K. Das, D. Bisht, V andana, and P . K. Singh, “Estimation of photo v oltaic cells model parame- ters using particle sw arm optimization, in Ph ysics of Semiconductor De vices, Springer International Publishing, 2014, pp. 391–394, doi: 10.1007/978-3-319-03002-9-98. [30] Y . Zhang, S. L yden, B. L. de la Barra, and M. Haque, A genetic algorithm approach to parameter es- timation for PV modules, in 2016 IEEE Po wer and Ener gy Society General Meeting (PESGM), Jul. 2016, doi: 10.1109/PESGM.2016.7741781. P ar ametric estimation in photo voltaic modules using the cr ow sear c h algorithm (Oscar Danilo Montoya) Evaluation Warning : The document was created with Spire.PDF for Python.