Neural approximate dynamic programming for the ultra-fast order dispatching problem
研究了超快速订单调度问题，提出结合近似动态规划和深度强化学习的NeurADP方法，通过神经网络处理高维动态，显著提升配送效率，为同日配送平台提供管理启示。
Same-Day Delivery (SDD) services aim to maximize the fulfillment of online orders while minimizing delivery delays, but are beset by operational challenges, particularly the uncertainty in online order arrivals. Our work focuses on the ultra-fast Order Dispatching Problem (ODP), which involves dynamically matching and dispatching orders to couriers from a centralized warehouse, with the goal of completing each delivery within a strict time window. We introduce important extensions to the ultra-fast ODP, including order batching and explicit courier assignments, to better reflect real-world dispatching operations and improve delivery efficiency. As a solution method, we focus on NeurADP, a methodology that combines Approximate Dynamic Programming (ADP) and Deep Reinforcement Learning (DRL); our work constitutes the first application of NeurADP outside of the ride-pool matching problem. NeurADP is particularly well-suited to ultra-fast ODP, as it addresses complex one-to-many matching and routing intricacies through a neural network-based function approximation that captures high-dimensional problem dynamics. Numerical results demonstrate that the inclusion of order batching and courier queues enhances the efficiency of delivery operations and that NeurADP significantly outperforms benchmark policies. Extensive sensitivity analysis confirms the robustness of NeurADP across a variety of operational scenarios and provides managerial insights for ultra-fast delivery platforms.