Enhancing Intra-Hour Solar Irradiance Forecasting for Solar Applications: A Blended Model of Satellite, Sky Imager, and Persistence

Abstract

The increasing integration of solar power requires highly accurate intra-hour solar irradiance forecasts. This study aims to significantly improve intra-hour solar irradiance forecasts by developing and evaluating a blending approach that integrates distinct forecast sources. Our methodology involves extending the horizon of an All-sky imager (ASI) data-driven transformer-based model up to 1 h ahead. The outputs of this ASI model are blended with a Heliosat-3-based satellite forecast and a persistence forecast via linear regression as well as with distinct advanced machine learning algorithms. We assess the hybrid system’s performance across varying sky conditions and analyze the impact of temporal aggregation schemes and the effective spatial coverage of a single ASI installation. Results demonstrate that this integrated multisource hybrid approach provides substantial benefits by reducing the overall root mean squared error and mean absolute error over the standalone satellite forecast by 13.6% and 17.0%, respectively. This is attributed to the complementary strengths of the individual models: ASI excels under dynamic conditions, satellite offers broader spatial coverage, and persistence provides a robust baseline for the immediate future. Furthermore, the strong generalization capability of the ASI model is shown through its effective performance across climatically distinct sites (training in southern Spain and validation in northern Germany).