An enhanced Gradient-based Optimizer for parameter estimation of various solar photovoltaic models

Abstract

The performance of a PhotoVoltaic (PV) system could be inferred from the features of its current–voltage relationships, but the PV model parameters are uncertain. Because of its multimodal, multivariable, and nonlinear properties, the PV model requires that its parameters be extracted with high accuracy and efficiency. Therefore, this paper proposes an enhanced version of the Gradient-Based Optimizer (GBO) to estimate the uncertain parameters of various PV models. The Criss-Cross (CC) algorithm and Nelder–Mead simplex (NMs) strategy are hybridized with the GBO to improve its performance. The CC algorithm maximizes the effectiveness of the population and avoids local optima trapping. The NMs strategy enhances the individual search capabilities during the local search and produces optimum convergence speed; therefore, the proposed algorithm is called a Criss-Cross-based Nelder–Mead simplex Gradient-Based Optimizer (CCNMGBO). The primary objective of this study is to propose a simple and reliable optimization algorithm called CCNMGBO for the parameter estimation of PV models with five, seven, and nine unknown parameters. Firstly, the performance of CCNMGBO is validated on 10 benchmark numerical optimization problems, and secondly, applied to the parameter estimation of various PV models. The performance of the CCNMGBO is compared to several other state-of-the-art optimization algorithms. The results proved that the proposed algorithm is superior in handling the numerical optimization problem and obtaining the uncertain parameters of various PV models and performs better during different operating conditions. The convergence speed of the proposed CCNMGBO is also better than selected optimization algorithms with highly reliable output solutions. The average objective function value for case 1 is 9.83E−04, case 2 is 2.43E−04, and the average integral absolute error and relative error values are 1.05E−02 and 3.51E−03, respectively, for all case studies. With Friedman’s rank test values of 2.21 for numerical optimization and 1.66 for parameter estimation optimization, the CCNMGBO stood first among all selected algorithms.