Accurate short-term forecasts of wind and solar power are essential for an optimal integration of these energy sources in power systems and electricity markets. Commonly, these forecasts are based on current output at the forecast location and/or output from physical models. With the substantial deployment of generation capacities, high amounts of data have become available at numerous generation sites and at a high temporal resolution.
This project aims in improving power generation forecasts by employing the space-time dependencies in these data. Therefore high-dimensional forecasting models are developed that filter out the most relevant information in the data, adapt to changes in the underlying processes and are still computationally feasible in an operational setting.