Safe Reinforcement Learning-based Control for a Voltage Source Inverter Operating in an Unbalanced Grid

Unbalances in electric energy grids, e.g., caused by single-phase connected loads, can have a negative impact on the grid quality and can be challenging to handle for controllers. Data-driven control approaches such as reinforcement learning have recently gained significant momentum in inverter-based grid applications. However, unbalanced grid conditions have been not investigated yet. Therefore, in the considered scenario, a reinforcement learning controller is safely trained in a hardware experiment to run a three-level neutral-point-clamped voltage source inverter with an LC filter for supplying an unbalanced load. It is experimentally shown that the reinforcement learning controller can be applied to the unbalanced load application without structural changes while only the training scenarios need to be extended to unbalanced cases.