A Primal-Dual Deep Deterministic Policy Gradient-Based Dispatch Method for Electric-Hydrogen Coupling System Under Uncertainty

The electric-hydrogen coupling system (EHCS) has emerged as a fundamental infrastructure for modern clean energy, addressing the challenges posed by high renewable energy penetration in power systems. Based on deep reinforcement learning, this paper proposes a Primal-Dual Deep Deterministic Policy Gradient (PD-DDPG)-based dispatch method for EHCS. This method efficiently and reliably manages renewable energy uncertainties, ensuring the secure operation of the EHCS. Within the PD-DDPG, a constrained Markov Decision Process (CMDP) formulation for the EHCS is developed, which integrates an uncertainty-interactive environment and a constraint-learning training mechanism to enhance constraint satisfaction. The CMDP is parameterized using the DDPG baseline, and a primal-dual optimization approach is devised to facilitate adaptive safe exploration for decision policies. The effectiveness of the proposed dispatch method for the EHCS is validated using the modified IEEE-118 test system.