A prioritized replay safe soft-actor critic (PR-sSAC) deep reinforcement learning for real-time energy dispatch of integrated energy systems

Integrated energy systems (IES) are complex systems that integrate multiple energy sources, converters, and loads to achieve optimal operation and carbon neutrality. However, conventional optimization methods for IES scheduling face challenges such as nonconvexity, uncertainty, and system dynamics. This paper proposes a novel deep reinforcement learning (DRL) approach for real-time energy dispatch of IES. The approach uses a prioritized replay safe soft-actor critic (PR-sSAC) algorithm that incorporates a safety network and a prioritized experience replay mechanism to improve the performance and reliability of the DRL agent. The safety network ensures that the agent’s actions do not violate the physical constraints of the IES components, while the prioritized experience replay mechanism enhances the sampling efficiency and convergence speed of the agent. Simulation experiments are conducted on a case study of an IES model with six components and the approach is compared with other DRL methods and a theoretical optimization model. The results show that the approach achieves near-optimal operation cost and tolerable constraint violation, demonstrating its effectiveness and applicability for IES scheduling.