Model-Predictive Railway Traffic Management

Due to the increased traffic on the roads people are looking for attractive alternatives to reach their destination. Especially for longer trips railway transportation is an attractive alternative. Because of this more people are traveling by train, resulting in the railway networks being operated closer to their maximum capacity. This has negative effects on the robustness of the timetable: a few small delays in the network can cause numerous secondary delays and spread over large parts of the network. It is therefore important that these initial delays are detected and dealt with as efficiently as possible.

Current practice of many railway operators is to divide the network into several dispatching areas, each with their own dispatcher, who tries to deal with the problems of his dispatching area as efficiently as possible. The dispatcher can make changes in the order of the trains departing and arriving at stations, cancel trains and connections, or reroute trains. The dispatchers base their decisions on a given set of rules and their experience. These dispatching decisions might be optimal for the given dispatching area, but the dispatchers are often unable to get a clear picture of the situation in the entire network and as a result their actions might cause unnecessary delays in other parts of the network.

To address this problem we are developing a strategy to determine the dispatching actions that result in the largest reduction of the sum of delays in the entire network. We use a railway traffic model, described as a discrete event system, to predict the propagation of the delays and the effects of dispatching actions on this propagation. Based on the model we define a mixed integer linear programming problem. By solving the Mixed Integer Linear Programming (MILP) problem the optimal dispatching action are determined.

The main is solving this problem within a very short time period. The problem needs to be solved very quickly because the results need to be applied during the daily operation of the railway network and the operation cannot be delayed because the problem has not been solved yet. We aim to achieve a reduction in the computation time by reducing the size of the discrete event system and by doing so we also reduce the problem size of the MILP problem, resulting in shorter computation times. We are also performing research on distributed optimization, further reducing the computation time by distributing the problem over multiple computers, while still solving the original problem optimally.

Project members: B. Kersbergen (Bart), T.J.J. van den Boom (Ton), B. De Schutter (Bart)

Distributed and large-scale systems, Discrete-event systems, Model predictive control, Optimization-based control, Railway networks

Sponsored by: