Dynamic Discretization Discovery for the Multidepot Vehicle Scheduling Problem with Trip Shifting

The solution of the multidepot vehicle scheduling problem (MDVSP) can often be improved substantially by incorporating trip shifting (TS) as a model feature. By allowing departure times to deviate a few minutes from the original timetable, new combinations of trips may be carried out by the same vehicle, thus leading to more efficient scheduling. However, explicit modeling of each potential trip shift quickly causes the problem to get prohibitively large for current solvers such that researchers and practitioners are obligated to resort to heuristic methods to solve large instances. In this paper, we develop a dynamic discretization discovery algorithm that guarantees an optimal continuous-time solution to the MDVSP-TS without explicit consideration of all trip shifts. It does so by iteratively solving and refining the problem on a partially time-expanded network until the solution can be converted to a feasible vehicle schedule on the fully time-expanded network. Computational results demonstrate that this algorithm outperforms both the explicit modeling approach and a branch-and-price algorithm by a wide margin and is able to solve the MDVSP-TS for real-life instances with close to 4,000 trips even when many departure time deviations are considered.