An Economic Optimization Scheduling Method for Photovoltaic Storage Charging Stations Based on Improved Proximal Policy Optimization Algorithm

To address the optimization scheduling problem of photovoltaic storage charging stations, the traditional Proximal Policy Optimization (PPO) algorithm, which uses a fixed penalty coefficient constraint violation function, limits the agent's ability to distinguish between similar strategies during training, making it difficult to obtain a globally optimal strategy. Therefore, this paper proposes an improved PPO algorithm considering a stepwise penalty coefficient. The method first constructs an economic optimization scheduling model for photovoltaic-storage charging stations aimed at maximizing operational revenue and creates a reinforcement learning interaction environment, where model parameters define the agent's state and action spaces. The constraint violation amount is then calculated based on the agent's actions and the current environmental state, determining the stepwise penalty coefficient used to construct the constraint penalty function. Finally, through extensive offline training, the agent generates the optimal strategy. Case study results indicate that compared to the traditional PPO algorithm, the scheduling scheme obtained using the improved PPO algorithm further enhances the photovoltaic consumption rate and economic benefits of the photovoltaic storage charging station.