Railway Scheduling with Declarative Constraint Programming

Abstract

Simulation of train scheduling is a highly complex problem. Classical methods in this field are mainly designed for conflict resolution, which means that a solution or partial solution is generated and subsequently tested to determine whether the conditions are met (generate-and-test procedure). The main advantage of the proposed paradigm, Constraint Processing, is that its basic strategy is avoidance of conflicts. The use of the conflict-avoiding CP paradigm is advantageous, for example, in scheduling trains (track selection, global temporal situations, reservations), where strongly branched decision trees arise. Some examples are given illustrating the innovative aspects of the Constraint Processing paradigm. However, the size of real problems, in terms of track length, number and type of trains, different disposition rules, optimization or quality criteria, make it necessary to explore other methods to deal with the amount of data, to reduce the remaining search spaces, to ensure short response times and interactivity and to guarantee high-quality solutions.

We describe possible ways of coping with the above mentioned problems, especially to reducing the lateness of trains: automatic decomposition of large rail networks and distributed train scheduling, using a slice technique to improve the system’s backtracking behaviour with a view to finding faster, better solutions, and combining constraint processing and genetic algorithms to find alternative tracks in a station.