A learning-based energy management strategy for hybrid energy storage systems with compressed air and solid oxide fuel cells

The intermittency and volatility of renewable energy have been major challenges in modern power systems. This paper proposes a self-adaptive energy management strategy based on deep reinforcement learning (DRL) to integrate renewable energy sources into a system comprising compressed air energy storage, battery energy storage systems, and solid oxide fuel cells. However, the basic deep deterministic policy gradient algorithm lacks sensitivity to environmental changes, particularly when there is a mismatch in module capacity within the system. This limitation may affect the proper selection of the charging and discharging actions for the hybrid energy storage system. Thus, some modifications are dedicated to the careful replay buffer design in the basic algorithm, improving the ability to identify subtle changes in the reward function. The proposed method is also compared with other DRL methods to validate the feasibility and effectiveness. The simulation results demonstrate the compatibility of the improved algorithm with the proposed energy management strategy and better performance in terms of economic benefits.