Deep Reinforcement Learning for Integration of Train Trajectory Optimization and Timetable Rescheduling Under...

Area of Science:

• Railway Engineering
• Operations Research
• Artificial Intelligence

Background:

• High-speed trains face disruptions from weather and equipment failures, causing delays.
• Increasing traffic density exacerbates delays, leading to timetable conflicts.
• Current hierarchical control systems are inefficient for real-time rescheduling.

Purpose of the Study:

• To develop an integrated model for timetable rescheduling and train trajectory optimization.
• To address inefficiencies in existing hierarchical train control systems.
• To ensure the efficiency and punctuality of high-speed railways.

Main Methods:

• An integrated model combining timetable rescheduling and trajectory optimization was formulated.
• Minimum running time was introduced into rescheduling, serving as the objective for trajectory optimization.
• A deep reinforcement learning (DRL)-based approach was employed to solve the integrated problem.
• Constraints included train overtaking, platform availability, and train priority.

Main Results:

• The proposed method successfully generated integrated rescheduled timetables and optimized train trajectories simultaneously.
• Numerical experiments on the Beijing-Shanghai high-speed railway demonstrated the method's effectiveness.
• Computation time increased linearly with problem size, indicating scalability.

Conclusions:

• The integrated approach overcomes limitations of hierarchical systems for high-speed rail rescheduling.
• Simultaneous optimization of timetables and trajectories enhances railway operational efficiency.
• DRL provides an effective solution for complex real-time railway management problems.