Comprehensive approach to photovoltaic power forecasting using numerical weather prediction data and physics-based models and data-driven techniques

Photovoltaic power forecasting is essential for maintaining electric grid stability and efficiently integrating solar energy power plants into the national power generation system. However, it remains challenging due to the complexity of accurately predicting solar radiation across varying weather conditions and diverse photovoltaic system configurations. This study addresses these challenges by developing a novel integrated forecasting algorithm that includes numerical weather prediction data, physics-based models, and artificial neural networks. The algorithm enhances direct normal irradiance forecasts, computes global tilted irradiance using an improved transposition model, and predicts photovoltaic output with a dynamic thermal-electric model. Losses and inverter efficiency are also incorporated. The algorithm provides 72-h power forecasts with customizable temporal resolution, without the need for on-site observations. Validation against 15-min data from a real photovoltaic plant demonstrated mean bias errors and root mean squared errors of 7.5 W/kWp and 123.7 W/kWp (DC), and 9.3 W/kWp and 121.0 W/kWp (AC), corresponding to relative errors of 1.8 %, 30.0 %, 2.3 %, and 29.9 %. The algorithm is scalable, adaptable to various system configurations, and effective for regions with limited data, thus supporting improved grid operations, enabling better management of photovoltaic generation variability and enhancing energy system efficiency.