Solving the train dispatching problem via deep reinforcement learning

Every day, railways experience disturbances and disruptions, both on the network fleet side, that affect stability of rail traffic. Induced delays propagate through network, which leads to a mismatch in demand offer for goods passengers, and, turn, loss service quality. In these cases, it is duty human traffic controllers, so-called dispatchers, do their best minimize impact However, dispatchers inevitably have limited depth perception knock-on effect decisions, particularly how they areas are outside direct control. recent years, much work Decision Science has been devoted developing methods solve problem automatically support this challenging task. This paper investigates Machine Learning-based tackling problem, proposing two different Deep Q-Learning methods(Decentralized Centralized). Numerical results show superiority techniques respect classical linear based matrices. Moreover Centralized approach compared with MILP formulation showing interesting results. The experiments inspired data provided by U.S. class 1 railroad.