A chance-constrained and constraint-aware reinforcement learning methodology for hydrogen-integrated energy systems under uncertainty

Rapid growth of renewable resources and hydrogen infrastructure is reshaping multi-energy systems while amplifying operational uncertainty. This study introduces a decision-making structure for an electricity–heat–gas system with hydrogen production, storage, and refueling facilities, explicitly capturing variability in renewable output, electric-vehicle behavior, and hydrogen demand. A chance-constrained formulation governs interlinked energy flows and ensures probabilistic feasibility of network and equipment limits. To handle real-time operational adjustments, the model is paired with a constraint-aware reinforcement learning (RL) strategy that adapts to evolving conditions without violating safety or physical requirements. Hydrogen acts as a flexible carrier connecting power and gas networks via coordinated electrolysis and storage. Simulation results indicate that the proposed approach enhances system reliability, alleviates hydrogen demand fluctuations, and reduces operating costs compared with deterministic scheduling, demonstrating its value for resilient hydrogen-integrated energy systems.