The digitization of integrated, regional energy systems leads to the emergence of so-called cyber-physical energy systems, which are based on the integration of software components and physical processes. In the analysis and optimization of these systems, modeling and simulation methods and tools are of central importance. New requirements for modeling and simulation are due to a steadily growing amount of operating data from Internet-of-Things (IoT) sensors and the blurring of the sector boundaries between electricity, gas, heating, cooling and mobility. Data-based machine learning (ML) methods are successfully used for the simulation and modeling of physical systems. However, traditional methods of machine learning reach their limits, especially with dynamic physical processes 5 7. Especially in situations in which data acquisition is complicated or expensive, it is often difficult to learn complex relationships from sparse or incomplete data. Furthermore, the physical plausibility of the results is a central framework condition for many ML applications. Techniques and methods that can meaningfully combine established physical modeling as well as existing specialist and domain knowledge with machine learning (e.g. in the architecture of deep learning networks that are influenced by the equations of the underlying physics) are promising candidates for this problem to solve. The development of domain-informed, interpretable and robust ML methods and algorithms were defined as a central research requirement in the field of artificial intelligence 6. Machine learning and physical modeling models are very often in direct competition. While it is unlikely that purely physical models will be completely replaced by data-driven models, the combination of the two modeling paradigms opens up new possibilities.
|Effective start/end date
|1/02/22 → 31/01/23
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.