Data-driven optimization for ride-sourcing vehicle dispatching and relocation under demand and travel time uncertainty

• Published in: Transportation research. Part C, Emerging technologies Vol. 178; p. 105217
• Main Authors: Huang, Yunping, Zheng, Nan, Huang, Zheng, Liang, Enming, Hsu, Shu-Chien, Zhong, Renxin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2025
• Subjects: Approximate dynamic programming; Dispatching and relocation; Time-dependent travel time; Uncertain demand; Approximate dynamic programming; Uncertain demand; Time-dependent travel time; Dispatching and relocation
• ISSN: 0968-090X
• DOI: 10.1016/j.trc.2025.105217

The imbalance between vehicle supply and on-demand customers has been a long-standing challenge for central ride-sourcing platforms. The current literature usually bases the design of dispatching and relocation strategies on the assumption of time-invariant traffic (speed) to reduce the dimension of the problem. However, uncertain demand and dynamic travel time subject to traffic congestion can significantly affect optimal solutions. To meet these challenges, this article adopts a regional-level formulation and estimates network-level travel time with functional data analysis that captures the dynamics and stochasticity of travel time. We then propose a multi-stage decision model to address the dispatching and relocation decisions of a fleet of vehicles for a centralized platform. Furthermore, we formulate the problem as a stochastic programming model to account for spatial–temporal uncertainties in customer demand. We develop an Approximate Dynamic Programming (ADP) based approach to solve multi-stage decisions efficiently. To evaluate the effectiveness of our algorithm, we use a simulator based on New York City (NYC) yellow taxi data and the Manhattan road network. Numerical studies demonstrate that the incorporation of dynamic travel time is beneficial to improve system profit compared to using mean historical travel times. The ADP can significantly improve the total system profit compared to several popular decision practices. •Model network-based congestion and estimate time-dependent travel times using FDA•Develop a multi-stage decision model for dispatching and relocation optimization•Data-driven approximate dynamic programming algorithm to approximate value function•Demonstrate the benefits of incorporating time-dependent travel time