Predictive Maintenance in Photovoltaic Systems Using Ensemble ML Empirical Analysis

This paper aims to enhance the effectiveness and sustainability of photovoltaic (PV) systems by employing ensemble machine learning empirical analysis (EMLEA) to predict regular maintenance schedules using a minimal set of features. The research utilizes a 99.9kW PV system dataset, with a diverse set of features such as DC voltage, DC current, instantaneous power generation, power factor, and frequency. Ensemble machine learning (ML) algorithms including RandomForest, XGBoost, CatBoost, and LightGBM were deployed to forecast the regular maintenance needs of the PV system. The proposed EMLEA system identified DC voltage and DC current as key determining factors for the system’s regular maintenance needs. Using these essential features, the maintenance schedule was estimated, and the performance metrics were evaluated by the EMLEA. Out of all models tested, XGBoost emerged as the superior model with an impressive accuracy of 98.62% and a precision of 94.37%.