Planning Additional Trains on Corridors

Abstract

Every train schedule entails a certain risk of delay. When adding a new train to an existing timetable, planners have to take the expected risk of delay of the trains into account. Typically, this can be a very laborious task involving detailed simulations. We propose to predict the risk of a planned train using linear regression models on the basis of extensive real world delay data of trains. We show how to integrate these models into a combinatorial shortest path model to compute a set of Pareto optimal train schedules with respect to risk and travel time. We discuss the consequences of different model choices and notions of risk with respect to the algorithmic complexity of the resulting combinatorial problems. Finally, we demonstrate the quality of our models on real world data of Swiss Federal Railways.