Daniel Weber, a research assistant in the Department of Power Electronics and Electric Drive Systems (LEA), successfully defended his doctoral dissertation on 20 July 2026. In his thesis, he demonstrated that it is possible to use reinforcement learning based on artificial intelligence to train the voltage control of an inverter directly on the actual hardware without prior knowledge of the system parameters. To this end, a standard algorithm was extended to include a safety component that enables safe training during live operation, thereby allowing for a largely automated and model-free controller development process. The effectiveness of the approach was initially demonstrated on a three-phase 250 kVA inverter and validated in comparison with established controllers. The subsequent application to a 1.25 MVA system demonstrates the method’s potential for the automated commissioning of decentralised converter-based microgrids and, consequently, for future smart grids.
The Institute of Electrical Engineering and Information Technology extends its warmest congratulations to Dr Weber.
