Chinese high-speed railway has developed rapidly in the more intelligent and automatic direction over the past few decades. In this paper, we consider the optimization problem of the train timetable for the high-speed railway to minimize the total train waiting time and total station occupied time. To deal with time-related constraints, we first establish the train operation environment based on Discrete Event Dynamic System (DEDS). Then, we reformulate the timetabling problem as a Markov Decision Process (MDP) problem and propose an improved Q-learning approach by redesigning Q-value function to solve the problem. Finally, we consider the Beijing-Shanghai high-speed railway as a numerical example, where the passenger flow and train running time are stochastic. We empirically show that our Q-learning method reduces over 30% total waiting time and 1.9% total occupied time compared with the well-known First-Come-First-Service (FCFS) scheduling strategy.
High-speed Train Timetabling Based on Reinforcement Learning
2022-12-04
875253 byte
Conference paper
Electronic Resource
English
Cyclic train timetabling model for high speed railway
IEEE | 2016
|Springer Verlag | 2018
|Energy-Efficient Train Timetabling
Springer Verlag | 2023
|Passenger centric train timetabling problem
Online Contents | 2016
|