Re-enacting rare multi-modal real-world grid events to generate ML training data sets

Today's energy grids are facing huge challenges caused by the growing diversity of energy consumers and producers as well as an ongoing increase of renewable energy sources and e-mobility. Hence, it is essential that the grids continuously evolve by introducing new monitoring, protection and optimization concepts including machine learning (ML) approaches. To overcome the lack of existing monitoring data for rare real-world grid events, this paper presents a concept for generating training data sets for ML approaches based on a multi-modal grid simulation tool. The simulation tool as well as the proposed semi-automated data generation approach are introduced and the concept is verified based on a real-world battery storage maintenance event.