Safe Bayesian Optimization for Data-Driven Power Electronics Control Design in Microgrids: From Simulations to Real-World Experiments

Micro- and smart grids (MSG) play an important role both for integrating renewable energy sources in electricity providing power supply remote areas. Modern MSGs are largely driven by electronic converters due to their high efficiency flexibility. Controlling is a challenging task requirements of availability, safety voltage quality within wide range different MSG topologies resulting demand comprehensive testing new control concepts during development phase. This applies, particular, data-driven approaches such as reinforcement learning, which the stability operating behavior can hardly be evaluated on analytical basis. Therefore, OpenModelica Microgrid Gym (OMG) package, open-source software toolbox simulation optimization MSGs, proposed. It capable modeling simulating arbitrary offers Python-based interface plug & controller testing. In standardized OpenAI allows easy optimization. The usage benefits OMG designing controllers demonstrated utilizing Bayesian Both current loops source inverter standalone, grid-forming mode automatically tuned given uncertain application environment. Finally, transfer real-world laboratory experiments successfully demonstrated.