Adaptive Set Membership Photovoltaic Energy Generation Forecasting

Integrating Distributed Energy Resources, such as solar photovoltaic systems, is increasingly critical for modern power grids. However, their inherent variability presents significant challenges for accurate prediction and uncertainty handling, making real-time forecasting essential for efficient energy management. This paper presents the development of an adaptive forecasting model for solar energy generation using a Set Membership approach, focusing on generating multiple forecast scenarios to address uncertainty in stochastic Model Predictive Control frameworks. The method leverages a two-year dataset of photovoltaic generation in southeastern Finland sampled at a 15-minute rate. The proposed solution dynamically updates the set of feasible model parameters to capture the inherent variability of solar generation. Results demonstrate the model's ability to adapt to dynamic changes. The Adaptive Set Membership approach outperforms classic auto-regressive models, improving forecasting accuracy up to 4.3% across various prediction horizons. Furthermore, based on the uncertainty bounds on model parameters, the methodology provides effective tools to generate scenarios that properly represent the variability of future generation profiles.