PhD defence: Towards Solar Nowcasting

Solar energy is becoming increasingly important in electricity grids and buildings, but its production can change rapidly when clouds pass over solar panels. These fluctuations make it more difficult to balance electricity supply and demand, manage congestion in the electricity grid, and use solar energy efficiently in buildings.

This PhD thesis investigates how short-term solar energy production can be predicted more accurately using sky imagers, that continuously observe the sky, known as all-sky imagers, together with artificial intelligence. The research develops several forecasting approaches that use sky images to track cloud movement and predict changes in solar irradiance and photovoltaic power from a few minutes up to one hour ahead.

The results show that combining sky images with machine learning and deep learning can significantly improve short-term solar forecasts compared with conventional forecasting methods. The developed approaches are particularly valuable during changing and difficult weather conditions, when solar production is harder to predict.

The thesis also examines how these improved forecasts can be used in practice. More accurate information about upcoming solar production can support congestion management, reduce power imbalances, and improve energy management in buildings through measures such as load shifting, battery storage optimization, and increased use of locally generated solar electricity.

Overall, the research explains how solar nowcasting can contribute to a more reliable, flexible, and efficient integration of solar energy into future energy systems.