Machine Learning-Based Multi-Objective Optimization for Maximum Power Point Tracking and Fault Diagnosis in Intelligent Solar Photovoltaic Systems

Abstract

Photovoltaic (PV) systems require effective maximum power point tracking (MPPT) under rapidly changing environmental conditions while simultaneously maintaining reliable fault detection. Existing approaches commonly address MPPT optimization and fault diagnosis independently, limiting their ability to coordinate energy extraction and system health. This study proposes an integrated machine learning–multi-objective particle swarm optimization (ML-MOPSO) framework that combines environmental prediction, adaptive MPPT, converter optimization, and fault diagnosis within a unified closed-loop architecture. MOPSO simultaneously optimizes power extraction, convergence time, steady-state oscillation, tracking efficiency, fault-detection rate, and false-alarm rate, while machine learning predicts irradiance and temperature variations and identifies abnormal operating conditions. Simulation results demonstrate an average power extraction efficiency of 98.7%, with MPPT convergence time and steady-state oscillations reduced by approximately 34% and 41%, respectively. Under partial shading, the proposed framework achieved a 99.1% global maximum power point detection rate. The fault-diagnosis subsystem achieved 97.9% overall classification accuracy, with detection times below approximately 0.18 s. Robust performance was maintained under rapid irradiance changes, partial shading, measurement noise, parameter uncertainty, and multiple PV fault conditions. These findings demonstrate that coordinated predictive MPPT and fault-aware optimization can improve PV energy extraction, dynamic response, and operational reliability within a unified control framework.