The boarding and alighting process of passengers in subway stations affects the distribution of passengers on the platform and the train dwell time. During peak hours, the high passenger density in the boarding and alighting areas of the platform often leads to conflicts and congestion. In this paper, on the basis of on-site field research, the Deep Q-Network algorithm in deep reinforcement learning is used to describe the passenger boarding and alighting process and analyze the impact of the proportion of boarding and alighting passengers on the boarding and alighting process, and the results show that the passenger walking paths are basically the same in different traffic conditions, and that the potential crowded areas and the location of dense crowds, which can be observed as a safety hazardous area within the scenario, can be used to improve service quality. The DQN model proves to be effective for modeling the boarding and alighting process, and future research may explore related topics in greater depth.