联系我们

论文查询结果	返回搜索

论文编号：	15300
作者编号：	2120233664
上传时间：	2025/6/4 17:42:25
中文题目：	考虑多交付选项的“卡车-无人车”联合 配送问题研究
英文题目：	Research on the Joint Delivery Problem of ‘Truck- Drone’ with Multiple Delivery Options
指导老师：	安利平
中文关键字：	自提柜；卡车-无人车联合配送问题；时间窗；启发式算法；强化学习算法
英文关键字：	Self-pickup Locker; Truck-Autonomous Vehicle Joint Delivery Problem; Time Windows; Heuristic Algorithm; Reinforcement Learning Algorithm
中文摘要：	摘要 近年来，随着绿色物流理念的深入推广和电动技术的快速发展，传统物流配送面临成本高昂、能耗较大及客户需求日趋多样化的挑战。为此，卡车与无人车联合配送模式成为研究热点，多交付选项（卡车配送、无人车配送、自提柜自提）为企业提供了更加灵活且高效的服务策略。然而，在城市配送网络中，客户节点、快递站点和自提柜存在显著差异的服务时间窗和操作约束，如何在满足客户交付偏好基础上科学规划配送路径、降低总成本并提高配送效率，是亟需解决的关键问题。 本文围绕上述问题，构建了考虑客户偏好与多交付选项的卡车-无人车联合配送混合整数规划模型，明确以最小化卡车与无人车行驶成本、快递站点卸货成本、自提柜使用费用为优化目标，全面考虑车辆容量约束、各类节点时间窗约束以及客户交付偏好的满足比例。随后，为有效求解该模型，设计了一种融合SISRs算法和Q-learning强化学习的混合启发式算法（Q-S算法）。SISRs算法通过特有的破坏-重建机制，以成串移除客户节点并优化重建路径的方式，有效拓展解空间；Q-learning则通过学习路径插入操作的长期收益，动态调整插入顺序和位置，显著降低路径成本并提高配送效率。 数值实验表明，所提出的模型与算法在不同规模和复杂场景下均能获得高质量解，并具有较好的求解效率和鲁棒性。实验验证初始解由贪心策略高效生成，结合Q-learning决策的移除和插入算子有效提升路径优化质量。本文研究成果为物流企业实现降本增效、推进绿色配送提供了有价值的理论支撑与决策参考。 关键词：自提柜；卡车-无人车联合配送问题；时间窗；启发式算法；强化学习算法
英文摘要：	Abstract In recent years, the widespread adoption of global green logistics concepts and advancements in electric vehicle technology have posed significant challenges to traditional logistics distribution systems, including high operational costs, excessive energy consumption, and increasingly diversified customer demands. Urban logistics, in particular, has witnessed the emergence of truck-autonomous vehicle (TAV) joint delivery systems as a prominent research focus. This innovative approach capitalizes on the complementary advantages of trucks and autonomous vehicles to achieve efficient, cost-effective, and environmentally sustainable last-mile delivery. By integrating multiple delivery options—including truck delivery, autonomous vehicle delivery, and self-pickup locker services—enterprises can flexibly accommodate heterogeneous customer preferences regarding delivery methods and time windows. However, the complex network structure involving customer nodes, satellite facilities, and self-pickup lockers presents operational challenges due to divergent service time windows and system constraints. Consequently, optimizing vehicle routing to minimize costs while ensuring customer satisfaction and operational efficiency remains a critical research problem. This study establishes a two-stage mixed-integer linear programming (MILP) model to mathematically formulate the TAV joint delivery problem, incorporating the interdependencies among distribution centers, satellite nodes, self-pickup lockers, and customer nodes, along with customer delivery preferences. The model aims to minimize total distribution costs, encompassing truck/autonomous vehicle operating costs, satellite node unloading costs, and locker usage fees, subject to constraints including vehicle capacity, battery range, time windows, and preference compliance. To address this NP-hard optimization challenge, we propose a novel hybrid metaheuristic algorithm that synergistically combines the destruction-reconstruction mechanism of the SISRs heuristic with Q-learning reinforcement learning. The SISRs component facilitates global exploration through iterative route removal and reconstruction, while the Q-learning module dynamically optimizes insertion sequences by evaluating long-term rewards of state-action pairs, thereby systematically reducing route costs and improving solution quality. Computational experiments demonstrate the model’s robustness and the algorithm’s superior performance across diverse problem scales and distribution scenarios. The greedy-based initial solution ensures constraint feasibility, while the Q-learning-guided removal operator strategically relaxes route constraints to enable subsequent optimization. During reconstruction, the Q-learning-enhanced insertion operator achieves cost reduction through incremental cost analysis and real-time Q-value updates. Empirical results confirm the algorithm’s effectiveness in minimizing total costs, reducing delivery durations, and enhancing system reliability, providing actionable insights for logistics enterprises pursuing sustainable and efficient distribution. Future research directions include incorporating dynamic demand fluctuations and real-time traffic data, as well as integrating other advanced optimization techniques to improve model precision and practical applicability. Keywords：Self-pickup Locker; Truck-Autonomous Vehicle Joint Delivery Problem; Time Windows; Heuristic Algorithm; Reinforcement Learning Algorithm
查看全文：	预览  下载(下载需要进行登录)