Energy optimization scheduling of grid-connected renewable energy hydrogen production system based on RGHS-PPO algorithm

To address the issue of "wind and solar curtailment" caused by the intermittency and volatility of wind and photovoltaic (PV) power in renewable-based hydrogen production, this study develops an integrated framework and equipment model for a power grid-coupled renewable hydrogen system (RGHS). The core contribution lies in proposing a novel economic dispatch approach based on deep reinforcement learning (DRL), which effectively overcomes the limitations of traditional CPLEX methods that rely heavily on accurate forecasting and underperform in dynamic state monitoring. Specifically, this study: (1) introduces, for the first time, an economic dispatch model framework for RGHS; (2) innovatively designs an RGHS-PPO approach based on Proximal Policy Optimization (PPO), leveraging its low-complexity advantage to efficiently optimize system performance; and (3) formulates a scheduling strategy aiming to maximize system profit and minimize environmental cost. The real-time dispatch problem is modeled as a Markov Decision Process (MDP), with a carefully designed state space, action space, and reward function, and the RGHS-PPO algorithm is employed for training. Simulation results demonstrate that the RGHS-PPO algorithm not only closely approximates the performance of traditional day-ahead optimization strategies using CPLEX but also significantly reduces total system operating costs compared to state-monitoring-based strategies. Moreover, it improves the utilization of wind and solar energy and exhibits strong robustness and adaptability under forecast uncertainty.