基于乘客订单取消预测的网约共享出行平台派单优化研究

To mitigate the high incidence of order cancellations by passengers in ride-sourcing systems and enhance platform operational efficiency, this study adopts a data-driven approach within the theoretical framework of “predict then optimize”. By integrating predictive methodologies from data science with operations research optimization techniques, this research analyzes the complex dispatching challenges in ride-sourcing systems, specifically considering passenger order cancellations. The study reveals that: i) The predictive-then-optimization framework effectively simplifies the platform’s dispatching optimization problem with consideration of passenger’s order cancellations into a linear programming model, which significantly improving the solvability of the model and reducing the difficulty of theoretical analysis; ii) Employing real data, machine learning models can effectively predict whether passengers will cancel orders, thus avoiding the limitations of assumptions inherent in mathematical modeling of passenger decision-making processes; iii) Compared to dispatching models that do not consider passenger’s order cancellation behavior, the model proposed in this paper can effectively improve the revenue of the ride-sharing platform. Numerical experiments indicate that as the supply-demand ratio (drivers/passengers) increases, the solutions of the dispatching strategies that consider passenger’s order cancellation behavior and those that do not gradually converge; compared to orders with short or long travel distances, orders with medium travel distances contribute more significantly to the platform’s revenue; compared to cost-priority and profit-priority strategies, the dispatching strategy that accounts for passenger’s order cancellation behavior achieves higher revenue and can effectively reduce the total waiting time of passengers, among other benefits. This paper provides a modeling approach and solution method for the optimization of ride-sharing platform’s dispatching considering passenger’s order cancellation behavior, offering theoretical reference for the improvement of dispatching strategies.