A Global Method for Crew Planning in Railway Applications

Abstract

Crew planning is a typical problem arising in the management of large transit systems (such as railway and airline companies). Given a set of train services to be performed every day, the problem calls for a set of crew rosters covering the train services at minimum cost. Although the cost may depend on several factors, the main objective is to minimize the number of crews needed to perform the rosters. The process of constructing the rosters from the train services has been historically subdivided into three independent phases, called pairing generation, pairing optimization, and rostering optimization, as is the case for the approach presented in Caprara et al. (1999a) for the Italian railways. In that paper, it is suggested that a feedback between the last two phases may significantly improve the quality of the final solution. In this paper, we illustrate the implementation of a new crew planning system within the EU Project TRIO. In particular, we describe the design of a new module for pairing generation, as well as an effective technique for integrating the pairing and rostering optimization phases into a unique one. The improvements over the previous approach are shown through computational results on real-world instances.