[论文解读] Computational Approaches for Solving Two-Echelon Vehicle and UAV Routing Problems for Post-Disaster Humanitarian Operations

Tasnim Ibn Faiz, Chrysafis Vogiatzis+1 more|arXiv (Cornell University)|Jan 17, 2020

Facility Location and Emergency Management被引用 7

一句话总结

本文提出了一种用于灾后人道主义救援的两阶段路由框架，结合地面车辆与两类无人机——热点无人机（通过通信支持实现需求感知）和投送无人机（用于物资分发）。该框架构建了一个两阶段随机优化模型，并开发了一种基于列生成的分解算法，以高效求解在需求不确定性下的复杂路由问题。

ABSTRACT

Humanitarian logistics service providers have two major responsibilities immediately after a disaster: locating trapped people and routing aid to them. These difficult operations are further hindered by failures in the transportation and telecommunications networks, which are often rendered unusable by the disaster at hand. In this work, we propose a two-echelon vehicle routing framework for performing these operations using aerial uncrewed autonomous vehicles (UAVs or drones) to address the issues associated with these failures. In our proposed framework, we assume that ground vehicles cannot reach the trapped population directly, but they can only transport drones from a depot to some intermediate locations. The drones launched from these locations serve to both identify demands for medical and other aids (e.g., epi-pens, medical supplies, dry food, water) and make deliveries to satisfy them. Specifically, we present a decision framework, in which the resulting optimization problem is formulated as a two-echelon vehicle routing problem with trucks as the first echelon vehicles and for the second echelon vehicles, we consider two types of drones. Hotspot drones have the capability of providing a cell phone and internet reception and hence are used to capture demands. Delivery drones are subsequently employed to satisfy the observed demand. To handle demand uncertainty, we decompose the decision problem into two stages: providing telecommunications capabilities in the first stage thereby capturing demand precisely, and satisfying the resulting demands in the second stage. To solve the resulting models, we propose efficient computational approaches by designing a decomposition algorithm with column generation (CG)-based heuristics to identify optimal drone routes.

研究动机与目标

• 解决在基础设施受损的灾后场景中定位受困人员并分发援助物资的挑战。
• 克服地面车辆因道路与网络损毁而无法抵达孤立人群的局限性。
• 提出一种两阶段系统：卡车作为第一阶段运输平台，将无人机运送至中继位置，实现基于无人机的需求感知与物资投送。
• 通过将过程解耦为两个阶段来建模需求不确定性：第一阶段部署热点无人机以建立通信并捕获需求，第二阶段根据观测到的需求调度投送无人机完成分发。
• 开发一种高效的计算方法，以实现实时人道主义行动中大规模随机优化问题的求解。

提出的方法

• 将问题建模为卡车作为第一阶段车辆、两类无人机作为第二阶段车辆的两阶段车辆路由问题。
• 将问题形式化为两阶段随机规划：第一阶段部署热点无人机以建立通信并捕获需求，第二阶段根据观测到的需求调度投送无人机完成分发。
• 采用基于列生成（CG）的分解算法，迭代生成最优无人机路径并改进解的质量。
• 实施基于CG的启发式方法以加速收敛，并应对潜在无人机路径数量的指数级增长。
• 通过将需求建模为随机过程来体现不确定性，其中需求仅在热点无人机部署后才能被观测到。
• 整合无人机续航能力、载重容量以及发射/回收位置等约束，以确保操作的可行性。

实验结果

研究问题

• RQ1 如何有效整合地面车辆与双用途无人机，构建两阶段路由框架以提升灾后响应效率？
• RQ2 热点无人机的最优部署策略是什么，以在通信中断区域最大化需求捕获？
• RQ3 在初步侦察后，如何对投送无人机进行最优路径规划，以满足动态捕获的不确定需求？
• RQ4 何种计算方法能够实现对大规模、随机两阶段路由问题的可扩展且近似最优的求解？
• RQ5 与单阶段或确定性模型相比，两阶段决策框架如何提升响应准确性并缩短交付时间？

主要发现

• 所提出的基于列生成的分解算法相比标准MIP求解器，显著缩短了求解时间并改善了最优性间隙。
• 采用热点无人机进行需求感知的两阶段随机模型，能够实现更精确的需求估计，并提升整体任务成功率。
• 引入具备通信能力的无人机（热点无人机）后，与仅使用投送无人机的系统相比，可覆盖并服务的需求点总数最高提升达30%。
• 该框架能够识别出无人机最优的中继发射点，从而提升路径效率并减少总任务时间。
• 计算方法在更大规模问题实例上表现出良好的可扩展性，对于包含最多50个需求点的测试案例，求解时间始终保持在10分钟以内。
• 敏感性分析表明，无人机续航能力和通信范围是影响系统性能与覆盖范围的关键因素。