Vehicle routing for customized on-demand bus services

This study investigates a variant of the Vehicle Routing Problem (VRP) for customized on-demand bus service platforms. In this problem, the platform plans customized bus routes upon receiving a batch of orders released by passengers and informs the passengers of the planned pick-up and drop-off locations. The related decision process takes into account some passenger-side time window-related requirements, walking limits, the availability and capacities of various types of buses. A mixed-integer linear programming model of this new VRP variant with floating targets (passengers) is formulated. To solve the model efficiently, a solution method is developed that combines the branch-and-bound and column generation algorithms and also includes embedded acceleration techniques such as the multi-labeling algorithm. Experiments based on real data from Dalian, China are conducted to validate the effectiveness of the proposed model and efficiency of the algorithm; the small-scale experimental results demonstrate our algorithm can obtain optimal results in the majority of instances. Additionally, sensitivity analysis is conducted, and model extensions are investigated, to provide customized bus service platform operators with potentially useful managerial insights; for example, a platform need not establish as many candidate stops as possible, a wide range of walking distance may not bring early arrival at destinations for customers, more mini-buses should be deployed than large buses in our real-world case. Moreover, the rolling horizon-based context and zoning strategies are also investigated by extending our proposed methodology.

Automated summary

This study investigates a variant of the Vehicle Routing Problem (VRP) for customized on-demand bus service platforms. A mixed-integer linear programming model is formulated and a solution method combining branch-and-bound and column generation algorithms is developed. Real data experiments validate the effectiveness of the proposed model and efficiency of the algorithm, demonstrating optimal results in the majority of instances. Sensitivity analysis, model extensions, as well as rolling horizon-based context and zoning strategies are also investigated.