Joint Order Dispatch and Repositioning for Urban Vehicle Sharing Systems Via Robust Optimization
Yiran Zhao,Guiyun Fan,Haiming Jin,Wenze Ma,Baoxiang He,Xinbing Wang
DOI: https://doi.org/10.1109/icdcs51616.2021.00069
2021-01-01
Abstract:Nowadays, the urban vehicle sharing (UVS) service offered by online platforms such as Car2Go and Zipcar has become an attractive choice for short-term mobility demands of city residents. Such way of shared mobility also demonstrates its potential in addressing various urban traffic problems (e.g., congestion, exhaust gas emission). However, the power of UVS systems could not be fully unleashed, unless the demand-supply gaps, incurred by the uneven spatial-temporal distributions of vehicle sharing orders and available vehicles, are effectively bridged. In this paper, we propose to resolve such problem by performing joint order dispatch and repositioning, which eventually maximizes the UVS platform's long-term cumulative profit. Specifically, we have addressed the various arising challenges, including capturing the interactive effect between order dispatch and repositioning, incorporating properly vehicles' fuel levels in our model, and dealing with stochastic demand with partially known statistical information. Technically, the problem is formulated as a stochastic dynamic program, and by integrating a series of methods, involving distributionally robust optimization, primal-dual transformation, as well as second order conic programming, we finally yield long-term optimal decisions. Furthermore, we validate our proposed methods using a large-scale real-world UVS dataset with over 300 vehicles and 300 thousand orders spanning overall 18 months.