Deep reinforcement learning-based control strategy for integration of a hybrid energy storage system in microgrids

Abstract

This study proposes a deep reinforcement learning-based control strategy for power management in hybrid energy storage-based microgrids. The proposed hybrid energy storage uses supercapacitors, batteries, and hydrogen storage to handle the power imbalance in microgrids. The major contribution of the present study is the implementation of deep reinforcement learning for optimal power-sharing among microgrid components considering the output response characteristics of the hybrid energy storage. The proposed control method is a two-layer deep reinforcement learning control strategy. The supervisory layer optimally distributes power among the hybrid energy storage components, while the local layer controls the switching control of the power electronics converters. The proposed control strategy is tested and validated with various operating scenarios. The experimental results demonstrate that the proposed local layers can significantly reduce overshoot and ripple in the DC bus voltage by up to 5% and 25%, respectively, compared to conventional method. In addition, the safe operation of expensive technologies, i.e., fuel cell and electrolyzer is ensured by controlling the rate of power change by the supervisory layer.