Integrating train scheduling and delay management in real-time railway traffic control

• Published in: Transportation research. Part E, Logistics and transportation review Logistics and transportation review, 2017-09, Vol.105, p.213-239
• Main Authors: Corman, Francesco, D’Ariano, Andrea, Marra, Alessio D., Pacciarelli, Dario, Samà, Marcella
• Format: Article
• Language: English
• Subjects: Delay management; MILP; Min-cost flow; Passenger routing; Train scheduling
• ISSN: 1366-5545; 1878-5794
• DOI: 10.1016/j.tre.2016.04.007

•The paper integrates passenger-centric and operations-centric models for train traffic control.•A MILP model is developed addressing the integrated problem.•Several heuristics and lower bounds are developed.•Algorithms are evaluated through several realistic test cases.•Good quality solutions are found for large networks with heavy traffic. Optimization models for railway traffic rescheduling tackle the problem of determining, in real-time, control actions to reducing the effect of disturbances in railway systems. In this field, mainly two research streams can be identified. On the one hand, train scheduling models are designed to include all conditions relevant to feasible and efficient operation of rail services, from the viewpoint of operations managers. On the other hand, delay management models focus on the impact of rescheduling decisions on the quality of service perceived by the passengers. Models in the first stream are mainly microscopic, while models in the second stream are mainly macroscopic. This paper aims at merging these two streams of research by developing microscopic passenger-centric models, solution algorithms and lower bounds. Several fast heuristic methods are proposed, based on alternative decompositions of the model. A lower bound is proposed, consisting of the resolution of a set of min-cost flow problems with activation constraints. Computational experiments, based on multiple test cases of the real-world Dutch railway network, show that good quality solutions and lower bounds can be found within a limited computation time.