High-speed Railway Timetable Rescheduling Under Random Interruptions Based on Reinforcement Learning

Research on high-speed train timetable rescheduling under interruption conditions is of significant importance for enhancing the real-time dispatching capabilities of railways and optimizing train operation efficiency. This study employs a data-driven optimization approach, specifically deep reinforcement learning, to explore methods for reconstructing train operation trajectories under interruptions. Using the Proximal Policy Optimization (PPO) model while considering train operation constraints, we propose a train rescheduling approach to minimize train delays. We establish a train operation simulation environment where the PPO intelligent agent continuously interacts with the environment, seeking the optimal strategy with minimal delay. To evaluate the PPO model's performance and efficiency, we conduct tests using scenarios involving random interruptions and actual data from the Wuhan-Guangzhou high-speed railway in China. The verification results demonstrate that the train rescheduling scheme derived from the PPO model outperforms those obtained from other common reinforcement learning models and even the decisions made by on-site dispatchers. It can reduce train delays by about 13%. PPO exhibits significantly faster convergence compared to other commonly used reinforcement learning models. Although the solution quality obtained by PPO is about 2% less than the optimal solution, the PPO model has a significant improvement in the computation speed of obtaining the near-optimal solution. This makes it a more practical choice for real-time decision-making. © 2023 Science Press. All rights reserved.