Project Summary

Accurate PV power forecasts are essential for the stable and efficient operation of the energy system. The PV-Fokusproject develops innovative machine learning methods to improve PV power forecasting and explores new meteorological input data. A particular focus is placed on high-resolution satellite-based information on solar irradiance and cloud cover, as well as newly developed predictors for the formation and dissipation of low-level clouds. A flexible ML architecture enables the combination of different data sources and models across forecasting horizons, from short-term intraday to day-ahead forecasts, and across different grid levels.

Our Contributions

KIT focuses in particular on low-level clouds, whose rapid and difficult-to-predict development can lead to significant PV forecasting errors. Using high-resolution satellite data, low-level stratiform and convective clouds are automatically identified and their development is analysed. Machine learning models and explainable AI methods are used to identify the key environmental drivers of cloud formation and dissipation.

The resulting insights are initially used for intraday forecasting and subsequently combined with numerical weather prediction models to enable day-ahead forecasts. In this way, PV-Fokus contributes to more accurate PV power forecasts and a more reliable integration of solar energy into the electricity grid.