The model-free and self-adaptive behavior of reinforcement learning (RL)-based control approaches offers the capability for highly dynamic and nonlinear control for complex systems and has recently shown convincing results for controlling standalone inverters at component level. Based on learning algorithms and general design rules, this setup enables fully automatic and model-free commissioning of an inverter-based microgrid without human intervention. As experimentally shown using 250 kVA industrial-grade inverters, the integration with other heterogeneous control concepts, such as model predictive control (MPC), is also possible in straightforward manner.
