Reliable intelligent control for electric drives

• Applicant: Professor Dr.-Ing. Armin Dietz
• Subject Area: Electrical Energy Systems, Power Management, Power Electronics, Electrical Machines and Drives ; Methods in Artificial Intelligence and Machine Learning
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 554851031

Energy-efficient electric drives are a key technology for achieving the European climate targets. Due to the high complexity of drive systems, operating strategies that do not achieve optimum energy efficiency are often used in practice. Methods from the field of artificial intelligence, in particular reinforcement learning (RL), enable automated learning of the control of complex electric drives and are therefore a promising approach to this problem. Own preliminary work and the work of other research groups show the efficiency of the methods on the one hand, but also the challenges on the other, especially for the desired training in the end application. Due to the methods and characteristic properties of the training, there is a high degree of variation in the control performance of the learned controllers. Furthermore, it has been shown that the training does not always reliably lead to a stable, usable operating strategy. For this reason, reliable and safe RL training is to be investigated in this project. In this context, the term “reliable” implies a reduction in the variance of the control quality of the learned controllers that are trained with the same setup, so that controllers with an optimal operating strategy are continuously trained. The term “safe” implies that training on the test bench is guaranteed within the physical system limits (e.g. temperature, current limit, etc.) and without damaging the system. To this end, possibilities for increasing reliability are initially researched in simulation. RL training can be influenced by a variety of parameters. In this project, research focuses on four points: the influences of the RL algorithm itself, exploration, buffer sampling and various environmental parameters, such as the reward function. Furthermore, strategies are being researched that comply with relevant limits in different operating ranges and dynamically adapt the control strategy, taking into account other operating point-dependent specifications. The knowledge gained in simulation is validated in test bench measurements and expanded with empirical findings. To this end, hardware acceleration is being researched to enable real-time control using neural networks and learning on the test bench, as well as the advantageous implementation of the training process on the control platform. The project could make RL-based control methods for electric drives suitable for practical use and thus make a significant contribution to increasing efficiency and saving energy in industry. The planned open-source publication of the software will enable the research results to be widely used and further developed.

• DFG Programme: Research Grants (Transfer Project)
• Application Partner: Zohm Control GmbH