A Novel ML Approach for Real-Time Prediction of PV Production: Tracking and Self-Shading Effects

The present paper introduces a patented method for real-time prediction and performance monitoring of photovoltaic (PV) systems. The approach integrates module temperature, irradiance, and solar angles into two physically interpretable matrices. Together, these matrices capture shading losses and production variations. The model was trained on synthetic data from a 1 MW plant with single-axis North-South trackers and then validated at a second site with similar coordinates, achieving an$(\mathbf{R}^{\mathbf{2}})$value of$\mathbf{0.988}$. The design of the algorithm is both transparent and lightweight, thus facilitating the generation of precise power forecasts while ensuring seamless integration into O&M platforms. This enhances fault detection and operational optimization with minimal computational effort.