Optimizing large on-demand transportation systems through stochastic conic programming

Abstract

On-demand transportation systems (OTS) are increasingly popular worldwide. Prior literature has studied how to control vehicle fleet in queueing-networks to rebalance excess supply or demand in OTS. This aggregated setting models the stochastic demand process and decompose large-scale networks for which product-form equilibrium distributions exist. However, such an approach is unsatisfactory in terms of computational complexity for its dependence on vehicle numbers. This paper presents a stochastic conic programming approach that obtains the near-optimal vehicle repositioning controls with endogenous demand with mild computational complexity and high fidelity. This global framework covers most existing queueing-network-based OTS models in the literature. Leveraging this approach, we explore day-to-day vehicle repositioning problems for on-demand vehicle operations in New York City. These results support the potential for providing a more accessible and sustainable on-demand mobility service, which is of particular significance as multimodal transport continues to emerge.