Optimal Predictive Maintenance Scheduling for Photovoltaic Systems Using Weather Features and Hybrid Machine Learning: A Case Study of the 10 MW Tozeur PV Power Plant

Abstract The global transition to renewable energy necessitates efficient maintenance strategies for photovoltaic (PV) systems. This study optimizes maintenance scheduling for the 10 MW Tozeur PV power plant using a novel, integrated approach. Seasonal variations in PV system efficiency and power output are initially modeled using a Seasonal Autoregressive Integrated Moving Average (SARIMA) model. Subsequently, weather-induced anomalies impacting system performance are identified through autoencoder-based anomaly detection. A hybrid forecasting model, combining Autoregressive Integrated Moving Average (ARIMA) and Isolation Forest algorithms, predicts periods of abnormal system behavior. Finally, a Random Forest Classifier integrates weather data and predicted anomalies to create a predictive maintenance schedule. This integrated methodology enhances maintenance reliability and demonstrates the potential of predictive analytics for optimizing the operational efficiency and lifespan of large-scale PV systems.