Enhancing photovoltaic system modeling and control under partial and complex shading conditions using a robust hybrid DE-FFNN MPPT strategy

Abstract

Improving efficiency, reliability, and extending the life of Photovoltaic (PV) systems are some essential benefits of developing an effective Maximum Power Point Tracking (MPPT) control system. In this study, an intelligent MPPT approach based on the Differential Evolution (DE) algorithm, and an adaptive Feed-Forward Neural Network (FFNN) is developed to address various issues of the most commonly used MPPT algorithms that affect the performance of PV systems including, power generation efficiency, tracking time, and instabilities around the Global Optima (GO). Consequently, in order to determine the suitable hyper-parameters’ values for the optimization and training procedure, the suggested hybrid technique is evaluated in the first part of this study for a variety of the FFNN configuration’s possibilities. Afterwards, the best obtained neural model is implemented as an MPPT controller, and a performance appraisal has been made between the DE algorithm based Feed-Forward Neural Network (DE-FFNN) and the most efficient utilized MPPT techniques. The observed results in the optimization and training phases show that our proposed strategy with 11 neurons in the hidden layer presents the most optimal neural model with a Mean Square Error (MSE) of 1.4924 in 139.8613 s in the optimization phase, and 2.57e−14 in the training phase. Subsequently, the created DE-FFNN model based-MPPT with the proposed configuration illustrates excellent performance compared to the applied MPPT techniques with a tracking speed between 0.0237 and 0.0106 s, a remarkable reduction in instability, and a power efficiency that can go up to 99.9971% in four different irradiance scenarios.