Centralized Secondary Control Through Reinforcement Learning for Isolated Microgrids

Abstract

Microgrids (MGs) continue to be a subject of extensive research that can enhance their operation and controls. This study addresses frequency control in MGs operating in islanded mode, where distributed energy resource integration and system stability are crucial. Conventionally, secondary control in MGs has relied on proportional-integral controllers. This paper proposes a novel approach for centralized secondary control by employing a deep deterministic policy gradient agent based on reinforcement learning (RL). Through extensive simulations conducted in Simulink, the performance of the RL-based secondary control system was evaluated. The results obtained demonstrate the effectiveness of the RL-based approach in maintaining the frequency within acceptable limits, even in the presence of load variations. The proposed RL-based secondary control showcases the potential of advanced machine learning techniques to revolutionize frequency regulation in MGs. By enabling more precise and adaptive control strategies, this approach improves efficiency, stability, and overall performance of MGs.