A policy safety-enhanced multi-agent reinforcement learning for cost-orient energy management of renewable energy powered multi electricity-hydrogen integrated charging stations

The coordinated management of multi-energy flows in integrated charging infrastructure is critical for enabling sustainable transportation networks with high renewable energy penetration. However, traditional methods struggle to simultaneously handle the dual intermittency of electricity and hydrogen demand, stochastic renewable generation, and stringent operational constraints, leading to sub-optimal cost and safety issues. To address it, this paper proposes a renewable energy-driven multi-electricity-hydrogen integrated charging station (multi-EHI-CS) system with a distributed energy resource framework. The energy management problem is formulated as a total cost minimization (covering operational, maintenance, and risk costs) with system constraints. To address the inherent uncertainties, it is formulated to a partially observable Markov game and solved via the multi-agent twin delayed deep deterministic policy gradient (MATD3) algorithm, which accounts for the stochastic nature of renewable generation and demand. Moreover, a policy safety enhancement mechanism is embedded into the multi-agent reinforcement learning framework to guide agents in learning safe and adaptive scheduling policies within feasible domains, ensuring operational safety during training and deployment. Simulation results demonstrate that the proposed approach reduces total costs by up to 12.85% compared to benchmark methods, effectively handling intermittency and uncertainties while ensuring safety.