Algorithm Struggles to Optimize Rail Schedules at Scale, Study Finds

<p>A railway timetable is typically planned one year in advance, but may be revised several times prior to the time of operation in order to accommodate on-demand slot requests for inserting additional trains and network maintenance. Revising timetables is a computationally demanding task, given the many dependencies and details to consider. In this paper, we focus on the potential of using optimization-based scheduling approach for revising train timetables during short term planning, from one week to few hours before the actual operation. The approach relies on a MILP (Mixed Integer Linear Program) model which is solved by using the commercial solver Gurobi. In a previous experimental study, the MILP approach was used to revise a significant part of the annual timetable for a sub-network in Southern Sweden to insert additional trains and allocate time slots for urgent maintenance. The results showed that the proposed MILP approach in many cases generates feasible, good solutions rather fast. However, proving optimality was in several cases time-consuming, especially for larger problems. Thus, there is a need to investigate and develop strategies to improve the computational performance. In this paper, we present results from a study, where a number of valid inequalities has been selected and applied to the MILP model with the aim to reduce the computation time. The experimental evaluation of the selected valid inequalities showed that although they can provide a slight improvement with respect to computation time, they are also weakening the LP relaxation of the model.</p>