Abstract
Photovoltaic (PV) systems are increasingly being deployed as a clean and sustainable energy source, but their efficiency is significantly impacted by environmental factors, particularly under partial shading conditions (PSC). These conditions introduce multiple local maxima in the power-voltage ($\mathrm{P}-\mathrm{V}$) curve, making it challenging for conventional maximum power point tracking (MPPT) techniques such as Perturb and Observe (P&O) to reliably track the global maximum power point (GMPP). This paper proposes a supervised machine learning (ML)-based MPPT strategy using a Multi-Layer Perceptron (MLP) regressor to directly predict the optimal duty cycle of a boost converter, leveraging inputs such as irradiance, temperature, and shading variation. A dynamic simulation environment is developed to evaluate and compare the proposed ML-based controller against the conventional P&O method. The model demonstrates improved tracking performance, especially under dynamic and partially shaded conditions. Quantitative performance metrics, including predicted maximum power point (MPP) accuracy, duty cycle optimization, and overall tracking efficiency, confirm the superiority of the ML-based approach for real-time MPPT applications.