Quantifying predictive uncertainty for energy Systems
JProf. Dr. Julian Quinting, Prof. Dr. Nikki Vercauteren, Prof. Dr. Jana Börner
Meteorology, Geophysics, Computer Science, Energy research
Motivation
As society pivots from fossil fuels toward a renewable-powered future, our energy systems hinge ever more critically on reliable forecasts for energy generation, storage, and consumption at various timescales. While sub-daily cloud cover and wind fluctuations require redispatch measures, multi-day Dunkelflauten need early planning of storage capacities, for example in the form of subsurface storage systems (e.g. geothermal), whose performance and uncertainties depend on reliable characterization and monitoring, to stabilize the energy grid. At the same time, breakthroughs in AI enable new possibilities that make such forecasts more computationally feasible. For example, deep learning algorithms now distill massive observational and reanalysis datasets into predictions with forecast skill exceeding that of numerical weather predictions models at considerably reduced computational costs. These fundamental changes—the rapid advancement of AI, the transition to clean energy and its associated search for suitable storage systems—create both an urgent need and a unique opportunity for interdisciplinary research connecting computer science, meteorology, geophysics and energy research. Only by developing holistic models that consider the complexities and uncertainties of the entire energy system will it be possible to achieve the transition to a reliable and sustainable energy supply.
To advance this research, we intend to establish a DFG research group that brings together existing expertise from three institutes at the University of Cologne (UoC) and selected partners in the surrounding region. First, the Institute of Geophysics and Meteorology (IGM) assembles expertise on atmospheric predictability across scales as well as geophysical methods to characterize and monitor uncertainty resulting from subsurface storage systems, providing complementary data streams for AI-based models. Second, the institute for Computer Sciences offers expertise on deep learning methods and on trustworthy AI models. Combining both areas of expertise offers unique opportunities to develop smart AI leading to uncertainty-aware predictive tools. Third, the Institute of Energy Economics has outstanding expertise in the field of energy‐system modeling. This expertise at UoC is complemented by the European Centre for Medium‐Range Weather Forecasts (ECMWF), a world leader in the development of AI‐based weather forecasting models that works closely with the IGM. Through our collaboration with the Institute of Technology, Resource and Energy Efficient Engineering at the Bonn‐Rhein‐Sieg University of Applied Sciences, we also secure a strong partner on the application side.