Energy Management for isolated renewable-powered microgrids using reinforcement learning and game theory

This paper presents a decentralized energy management system (EMS) solving for optimal load-response strategy applying reinforcement learning (RL) and game theory for islanded renewable-powered microgrids. The EMS enables the consumers in a microgrid to independently evaluate the tradeoff between satisfying load demand and maintaining sufficient stored energy to make load-response decisions correspondingly. The evaluation and decision-making process consists of two parts: an instant virtual two-player load-response game and a long-term linear-reward inaction (LR-I) learning process adjusting consumer power/load models. The virtual two-game solving process is an instantaneous decision-making system so that the consumers could make real-time decisions, while the LR-I process gradually improves the consumer payoff based on the system feedback during the operation. Simulation of a microgrid powered by PV cells and battery banks is conducted to evaluate the EMS performance. It is shown that the game-learning EMS has a better performance compared to both the direct virtual two-player game and the naïve LR-I approach. Additionally, compared to the naïve LR-I approach, the proposed game-learning algorithm has a faster converging-speed.