A Regime-Aware Hybrid for Photovoltaic Power Forecasting Using Physical Modeling and Deep Learning

Photovoltaic power production is inherently unstable, as it depends on rapidly changing meteorological conditions. Accurately predicting these variations enables energy providers to reduce reliance on costly backup power. This presentation addresses this issue through a regime-aware hybrid model that combines the strengths of a Temporal Convolutional Network and a physics-based model. The physics-based model is adjusted for solar output prediction under stable weather conditions, through estimating the Sun’s relative motion and the optical thickness of the cloud layer. Integrating sky classification, the hybrid highlights the potential of merging physics and computer science to predict power output 10 minutes to 2 hours in advance.