A bi-level optimization model for project scheduling and traffic flow routing in railway networks

Abstract

Long-term tactical infrastructure planning for a transportation network consists of deciding on renewals and major maintenance works. Such projects constitute large budget volumes and will impair the available traffic capacity during their execution, especially for railway systems. Quantitative methods that schedule and coordinate infrastructure projects together with traffic flow adaptations is however largely lacking today.

This paper addresses the joint planning of temporary capacity restrictions and traffic flow adaptions during track work closures, by proposing a bi-level optimization model which separates the problem into project scheduling (upper level) and traffic assignment (lower level). The latter model uses a novel traffic flow formulation for routing volumes of trains through the transportation network under the capacity restrictions given by the project scheduling. An aggregated network is used together with time discretized into uniform periods, which makes it possible to treat large national planning problems with a planning horizon of up to a year and a period length of a couple hours. The computational properties are evaluated, both for the individual models, and for their joint usage. Furthermore, results from applying the models on two case studies, concerning Northern and South-Western Sweden, are presented.

The main conclusion is that the model formulations are capable of solving realistic planning cases and to provide support for capacity planners at an infrastructure manager, even for a large national railway. The results show that a good overview over the collective traffic impact is obtained, but also that details of particular traffic relations or capacity usage over individual network links and their variation over time can be studied. One major deficiency has been identified in the flow-based traffic assignment model, which can lead to incoherent train flows over long traveling distances and many time periods.