Reinforcement Learning for Autonomous Control in Renewable Microgrids

The greater penetration of renewable energy sources into microgrids requires advanced control strategies that are able to effectively deal with the randomness and variability inherent in renewable power generation. This paper investigates the potential of reinforcement learning (RL) based approach to the autonomous control problem in renewable microgrids, and compares its performance to rule-based control approach. Through the extensive simulations, the RL based control method achieved a much higher RE utilization rate of 84.4 % in average, while the rule-based method has an average utilization rate of 70.8 %. Moreover, the RL methods decreased the overall operational costs by 37.9 %, and decrease in the generation cost, battery cycling cost, and grid importing cost can be observed. The ERI, which is one of the main metrics for system resiliency, was also increased by 11.8 %, demonstrating the effectiveness of the RLish method for preserving the energy availability and for handling the grid interactions. These results provide strong evidence for the efficacy of reinforcement learning in addressing optimal microgrid operation, which can lower costs and bolster system robustness, pointing to an auspicious direction for further sustainable energy management study and practice.