Abstract: Large-scale electric vehicles (EVs) scheduling models are complex and require high calculation capacity. To solve these problems, machine learning methods have attracted more and more attention in electric vehicle charging and navigation scheduling. For the photovoltaic-storage-charging integrated energy station, a scheduling strategy of the energy stations based on deep reinforcement learning (DRL) is proposed in this paper. Firstly, the operation strategy of energy station and the basic theory of deep reinforcement learning are analyzed. Secondly, the users psychological state of time and cost for different charging schemes are described based on regret theory, and the agent perception model of user-EV-station state environment is established. To improve the convergence speed of the algorithm, time varying ε-greedy strategy is introduced as action selection method of agent. Finally, multi-scenario simulations are designed based on the actual road network and energy stations in Nanjing. The results show that the proposed method effectively improves the photovoltaic consumption rate of the energy station under the condition of considering the psychological effect of various users. The proposed method provides a new idea for electric vehicle charging scheduling.