Robust railway crew schedules

On an intensively used network as the Dutch railway network, the problem of crew reschedul- ing is hard. Currently at NS, when disruptions occur there is no quantitative support for dispatchers to reschedule the duties. However, the manual approach of the dispatchers is not effective nor efficient. Therefore, operations researchers developed several methods that can support dispatchers in decision making. One of them, an approach using Column Generation, is very promising and is extensively tested in this thesis together with an heuristic approach that mimics the manual approach of the dispatchers and an agent-based approach. To test the three methods several disruption cases are constructed and combined with the schedules, either with or without reserve duties. After the comparison of the methods, the set of schedules is extended with some additional schedules. One of them is a schedule that is constructed as emergency schedule that can be used during extreme situations. Besides schedules, also different sets of stand-by duties are proposed. These new sets can be divided into two categories, where one is based on a location perspective and the other is based on a time perspective. Finally, we extended the scenarios such that a more realistic presentation of the practical situation, where multiple serious disruptions a day occur, is obtained. From the analysis it becomes clear the method based on Column Generation is in general the best approach. However, the agent-based approach performs particular well on smaller instances, sometimes even better than the Column Generation approach. The heuristic ap- proach does not perform well. Only rescheduling a schedule like the emergency schedule with the heuristic provides reasonable results. Apart from the emergency schedule there is very little difference in robustness between the schedules for all rescheduling methods. Robustness of a schedule is measured as the way schedules can be adapted after disruptions. Reserves make the process of rescheduling more easy, but the Column Generation approach performs also good without reserves. So the use of a good rescheduling method seems to contribute more to a robust railway crew schedule than changing the properties of a schedule.