The expansion of renewable energy generation is crucial for achieving climate neutrality. In particular, wind and PV (photovoltaic) power-generating units are predicted to massively increase in power production. Due to the volatility of renewable electricity from wind and PV, it is necessary to balance peak loads in the power grid. This is essential to ensure supply security due to the lack of grid stability. Coordination of load balancing requires increased flexibility in future power systems. Machine Learning (ML), especially reinforcement learning (RL), can play a crucial role in this regard. With RL, optimized control strategies can be developed which allow adaptation to dynamic operating conditions. This can significantly enhance the performance and efficiency of energy systems, thereby contributing to grid stability.

Concept for the application of reinforcement learning on a digital twin platform.

Concept for the application of reinforcement learning on a digital twin platform.

The focus of the research project RELY lies on the flexibilization of energy systems through optimized control using ML technologies. A key objective is to combine RL algorithms with Digital Twins (DT). This combination is expected to improve the efficient, safe and stable control of critical processes in power systems and leverage the technology from theory to practice. To assess the potential and scalability of the developed methods, a laboratory-scale reversible pump turbine available in the IET laboratory serves as a use case. The model pump turbine reflects key characteristics of power generating units of real and future power systems, allowing the reliability of RL algorithm based process controls to be demonstrated in general.

With the methods developed in the RELY project, the efficient production of renewable energy from hydropower can be strengthened. This actively contributes to a climate-neutral future at the IET.

  1. C. Tubeuf, J. Aus der Schmitten, R. Hofmann , C. Heitzinger, & F. Birkelbach (2024) Improving Control of Energy Systems With Reinforcement Learning: Application to a Reversible Pump Turbine. In Proceedings of the ASME 2024