A data-driven, variable-speed model for the train timetable rescheduling problem

Train timetable rescheduling - the practice of changing the routes and timings of trains in real-time to respond to delays - can help to reduce the impact of reactionary delay. There are a number of existing optimisation models that can be used to determine the best way to reschedule the timetable in any given traffic scenario. However, many of these models do not adequately account for the acceleration and deceleration required for trains to achieve the rescheduled timetable. The few models that do account for this are overly complex and cannot be solved to optimality in sufficiently short times. In this study, we propose a new model for train timetable rescheduling that uses statistical methods and historical data to parsimoniously take train speed into account. The model is tested using a new set of instances based on real data from Derby station in the UK. We show that the improved accuracy of the proposed model comes with little to no trade-off in terms of run time compared to fixed-speed timetable rescheduling models.