Data-Driven Driver Dispatching System with Allocation Constraints and Operational Risk Management for a Ride-Sharing Platform

In this article, we develop and analyze a driver dispatching system for a control center that aims to minimize passengers' waiting time. The system imposes allocation constraints that ensure a minimum number of drivers in different regions to manage operational risk. The data-driven system is based on Rolling Time Horizon approach and utilizes knowledge learned from historical data. It incorporates a hybrid forecasting model and a heuristic algorithm to solve the off-line problem in each iteration. We show that the NP-hardness of the off-line problem lies in allocation constraints. We test the performance of the system with a simulation study based on actual past taxi order data. The result suggests that the system markedly decreases the average waiting time and saves planning time in comparison with the request-driven dispatching mode. The result also demonstrates that in nonextreme cases, the dispatching system finds an acceptable solution which approximately satisfies allocation constraints while guaranteeing a short increase in waiting time.