Online-ordered meal logistics services (OMLSs) that accept online bookings and make vehicle plans to deliver meals from restaurants to customers have recently emerged. Customers have the option to cancel orders that are not delivered by appointment times, leading to significant financial, reputational, and customer losses for the OMLS providers. This study aims to make an appropriate vehicle plan for OMLS providers to minimize the expected total cost under the uncertainty of order cancellations. The problem is formulated as a two-stage stochastic programming model, and sample average approximation equivalent problems are generated using Monte Carlo simulation. To solve the equivalent problems, a parallel adaptive large neighborhood search (pALNS) with statistical guarantees is developed. Experiment results show that the vehicle plan derived from the ALNS is much better than the solution found by Gurobi within 10,800 s, with an average improvement of 14.90%. Additionally, the pALNS provides better statistical bounds in a shorter time compared to both the ALNS and the unsynchronized pALNS. Analytical experiments reveal that earlier cancellations lead to more severe consequences, offering valuable insights for OMLS providers to implement proactive measures to retain “urgent” customers.

中文翻译：

---

预约时间存在餐点取餐和送餐问题，订单取消存在不确定性


最近出现了在线订购的餐食物流服务 （OMLS），它接受在线预订并制定车辆计划，将餐点从餐厅送到客户手中。客户可以选择取消未在预约时间交付的订单，这将给 OMLS 提供商带来重大的财务、声誉和客户损失。本研究旨在为 OMLS 提供商制定适当的车辆计划，以在订单取消的不确定性下最大限度地降低预期总成本。该问题被表述为两阶段随机规划模型，并使用蒙特卡洛仿真生成样本平均近似等效问题。为了解决等效问题，开发了一种具有统计保证的并行自适应大邻域搜索 （pALNS）。实验结果表明，在 10,800 s 内，从 ALNS 得出的车辆计划远优于 Gurobi 找到的解，平均提高了 14.90%。此外，与 ALNS 和未同步的 pALNS 相比，pALNS 在更短的时间内提供了更好的统计边界。分析实验表明，提前取消会导致更严重的后果，为 OMLS 提供商实施主动措施来留住“紧急”客户提供了宝贵的见解。