Data-driven optimization of bus schedules under uncertainties

The vehicle scheduling problem (VSP) is one of the sub-problems of public transport planning. It aims to minimize operational costs while assigning exactly one bus per timetabled trip and respecting the capacity of each depot. Public transport planning is subject to various endogenous and exogenous causes of uncertainty, notably affecting travel time and energy consumption. Despite the uncertainties involved, the VSP and its variants are usually solved deterministically to address trackability issues. However, considering deterministic travel time in the VSP can compromise schedule adherence, whereas considering deterministic energy consumption in the electric VSP (E-VSP) may lead to solutions with sub-optimal true costs (including recourse costs and the cost of ownership of battery electric buses).


In this thesis, we propose a methodology for measuring the reliability (or schedule adherence) of public transport, along with stochastic data driven mathematical models and branch-and-price algorithms for two variations of this problem, namely the multi-depot vehicle scheduling problem (MDVSP) and the E-VSP. The thesis is composed of four articles. In the first article, we predict the distribution of bus travel times to assess the reliability of vehicle schedules in terms of their tolerance to delays. Then, in the second article, we introduce a model for the reliable MDVSP with stochastic travel time, minimizing both operational costs and penalties associated with delays. The third and fourth articles address the E-VSP with a two-stage stochastic program with recourse for the electric MDVSP, considering stochastic travel time and energy consumption, and a stochastic model for the E-VSP with battery degradation, respectively.

Emplacement : Zoom