Practical bus timetable optimization method based on deep reinforcement learning

Public transport travel can reduce air pollution, improve safety, and save many travel costs. However, as the passenger flow and traffic conditions change over time, it is difficult for bus schedules to respond in time. Therefore, it is meaningful to conduct real-time bus dispatching according to the current passenger flow and traffic conditions to reduce bus companies' costs and improve passengers' experience. Although deep reinforcement learning has made good progress in public transport control, most relevant research has made continuous decisions on whether to dispatch in a very short time interval. In this study, we consider bus timetable optimization as a reinforcement learning problem and thoroughly consider the practical application. The Proximal Policy Optimization (PPO) is used to determine the next departure interval. Therefore, the bus timetable is formulated dynamically and timely to cope with changes in traffic conditions and passenger demand. Specifically, we propose a novel deep reinforcement learning framework with a novel environment model, adding human experience to avoid learning from scratch and reconsider the step size, observation, and reward function to provide more effective information to the deep RL system. The experimental results show that our method is superior to the existing methods