Hybrid Models for Photovoltaic Power Forecasting under Different Physical Modeling Approaches

Reliable solar Photovoltaic (PV) power forecasting is critical for efficient grid management. This study evaluates hybrid models integrating physical and Machine Learning (ML) techniques for day-ahead PV power forecasting. Two hybridization approaches are considered: physics-informed ML and physics-guided models. The study evaluates multiple ML techniques integrated with three distinct physical modeling approaches of PV output power: a combination of standard models, an optimized model choice, and a detailed energy balance-based model proposed in this work. Results show that the physics-guided approach leveraging the most detailed physical model achieves the highest accuracy. However, physics-informed ML models deliver comparable performance with lower complexity, making them suitable for day-ahead applications with limited domain knowledge of PV physical models. This work also provides insights into the formulation of appropriate hybrid techniques for broader forecasting scenarios, such as regional or long-term PV forecasts.