Low-Carbon Economic Dispatch of Hydrogen-Based Community Microgrid System Using Deep Reinforcement Learning

The hydrogen-based community microgrid system (HCMS) offers a new approach to transforming the energy system into a low-carbon one through complementary electricity-hydrogen energy conversion. However, the system faces operational challenges: external electricity and hydrogen supply affect system economics and produce additional carbon emissions during energy production and consumption, reducing hydrogen’s decarbonization potential as a clean energy source. To address these challenges, this paper develops a low-carbon economic dispatch framework that considers electricity-hydrogen price dynamics and real-time carbon emission intensity from the external grid. This framework enables carbon emission analysis across the entire chain, including electricity demand, hydrogen use in transportation, hydrogen production through electrolysis, and fuel cell power supply. To enhance feature extraction for key temporal data such as power load, photovoltaic generation, and carbon emission intensity, an improved deep reinforcement learning algorithm is proposed by integrating attention mechanisms and Long Short-Term Memory (LSTM) networks into the Proximal Policy Optimization (PPO) framework. Simulation results confirm the proposed algorithm’s faster convergence speed and better dispatch performance compared to traditional PPO methods.