Accelerating logic-based Benders decomposition for railway rescheduling by exploiting similarities in delays

Abstract

The operation of a railway system is subject to unpredictable delays or disruptions. Operators control the railway system to minimize losses in performance. Real-time rescheduling is the adaptation of a railway schedule to any unforeseen delay or disturbance and recovers an optimal system state. In this work we propose the extension of an existing Benders decomposition scheme used so far for timetabling, to the case of railway rescheduling. We show how to increase its computational speed by a factor 2, by considering libraries of Benders cuts computed for other instances, to be reused in the solution. We show how including extra cuts has to balance a speedup potential, with a general slowdown due to optimization problems of increased sizes. We show that, if delays in an instance of rescheduling are in fact unknown, but come from a known statistical distribution, we can use a similarity measure to identify a-priori the most promising libraries of Benders cuts, which lead to speedups up to 20%.