MACHINE LEARNING APPROACHES FOR EFFICIENT MAXIMUM POWER POINT TRACKING IN SOLAR ARRAYS UNDER PARTIAL SHADING CONDITIONS

Abstract

Partial shading is still one of the most significant issues for photovoltaic (PV) systems, normally causing non-optimal energy generation because of the presence of more than one local maximum on the power-voltage (P–V) curve. Conventional Maximum Power Point Tracking (MPPT) methods, like Perturb and Observe (P&O), can’t find the global maximum power point (GMPP) efficiently under this condition. Here, we present the use of machine learning (ML) algorithms to enhance MPPT performance in partial shading conditions for solar panels. Synthetic data have been created using simulation under different irradiance, temperature, and shading conditions. Two ML algorithms, Gaussian Process Regression (GPR) and Multi Layer Perceptron Regressor (MLP), have been trained to predict, in real time, the optimal duty cycle for a boost converter. Comparative analysis with traditional P&O techniques showed that ML-based MPPT greatly enhanced tracking efficiency, especially in dynamic shading conditions, with an average efficiency enhancement of 5.3%. The results confirm the potential of ML algorithms to enable more robust, adaptive, and efficient solutions for solar energy harvesting maximization in complicated environmental conditions.