MCTS Based Approach for Solving Real-time Railway Rescheduling Problem

Abstract

Solving a real-time Railway Traffic Management Problem (rtRTMP) is a challenging task for human operators. To solve the traffic situation, many factors need to be considered. Traditionally, the most critical factor is the availability of the possible routes and the relative position of the vehicles to each other. Besides, additional constraints can be found, such as the velocity, the length, and railway company regulations. The human decision-making process is essential in case of any disturbance (deviation from the pre-planned timetable). The human operator may solve this situation, but generally, the solution is not optimal. In this paper, the authors present a new method, where they consider an MCTS based algorithm to solve the traffic situation in a fast way in a given station. The performance of the algorithm is examined in two abstraction levels. The main purpose is to execute an experimental study to examine the efficiency of the MCTS based algorithms to solve railway traffic situations.

Multi-Agent Reinforcement Learning for railway rescheduling

“…OpenTrack environment can provide a base for the implementation [9]. Another viable approach would be the Monte-Carlo tree search procedure, which can find a conflictfree route in a short time [10]. Using a graph representation and a Deep Q-Network (DQN) leads to a solution as well [11], [12].…”

Multi-Agent Deep Reinforcement Learning (MADRL) for Solving Real-Time Railway Rescheduling Problem

“…This technology may provide a fast solution for each train to find a conflict-free route through the network. MCTS based algorithms may find the best-first solution in a tree-based representation of the control problem [1]. Simple Q-learning algorithms also find the solution in a simplified infrastructure [2].…”