Optimizing Voltage and Frequency Stability in Renewable Microgrids Using Reinforcement Learning

Microgrids come with high penetration of renewable energy like solar become obvious to meet sustainable development goals. However, renewable energy has no rotational inertia like the conventional power grid, which shows resilience against voltage and frequency change to maintain grid stability.  To establish a microgrid as an independent and reliable power system a virtual inertia must be mounted. In this paper, a reinforcement learning-based droop-virtual inertia is incorporated into the microgrid system based on a vehicle-to-grid (V2G) control approach. The voltage magnitude and frequency at the point of common coupling constitute the microgrid states. The Q-learning algorithm is employed to enhance activities according to the status of the microgrid system.  To validate the algorithm in one case at fixed 1 kW/m² solar irradiation, a load of 180 kW and 40 kVAR was added for 1.5 s. In another, solar irradiation is varied for 2-6 seconds at the power generation end. In both cases, upon activation of the devised control mechanism, the output voltage and current are effectively regulated around their reference values. The suggested microgrid architecture and its associated control mechanisms have been rigorously tested in OPAL-RT to confirm their efficacy, and the real-time simulation outcomes suggest several possible applications in the advancement of the smart grid.