Improved MPPT for PV Systems Combining Thevenin Equivalents with Simple Neural Networks

khessam medjdoub; Marwa Bouazza; Mohamed Habbab; Abdeldjebar Hazzab; Anis Njeh; Hafedh Trabelsi

doi:10.5935/jetia.v12i59.3587

Authors

khessam medjdoub University of Naama Salhi Ahmed, Naama
Marwa Bouazza Department of Electrical Engineering, University of Naama Salhi Ahmed, Naama , Algeria
Mohamed Habbab Laboratory of CAOSEE, Tahri Mohamed University,Bechar, Algeria
Abdeldjebar Hazzab Electrical and Computer Engineering, Department Royal Military College of Canada
Anis Njeh CES, University of Sfax, Sfax Engineering School,Sfax, Tunisia
Hafedh Trabelsi CES, University of Sfax, Sfax Engineering School,Sfax, Tunisia

DOI:

https://doi.org/10.5935/jetia.v12i59.3587

Abstract

Efficient MPPT (Maximum Power Point Tracking) algorithms are required for PV (Photovoltaic) systems in order to obtain maximum power under temperature and sun irradiation variations. Single-diode models are complex and computationally intensive, while traditional methods, such as the perturbation and observation (P&O) method with oscillatory behavior and slow convergence. This study offers a simplified hybrid system that combines an integrated artificial neural network (ANN) controller with a Thevenin equivalent circuit description of the PV modules. While the NN forecasts the ideal operating cycle of the step-up transformer using only G and T inputs, the Thevenin technique reduces nonlinear current-voltage curves to and parameters at the maximum power point. Comprehensive validation using MATLAB/Simulink demonstrates a steady-state efficiency of 98.2%, 30% faster tracking than the P&O method, and robust partial shading recovery using only 800 training samples for application-ready performance.