基于鲁棒优化与外部协同的动态清洁物流机制研究

Cleaning logistics refers to logistics activities collecting hazardous, environmentally polluting or recyclable waste. Under the slogan of sustainable development, a global research boom has been launched. This project focuses on the vehicle routing problem at an operational level. The existing research does not solve the following problems well: it is difficult to accurately measure the volume of waste (demand randomness), and the actual demand can only be updated after the vehicle arrives (information dynamics). Besides the waste can be collected in multiple rounds (split pickups). To this end, we first propose two methods, robust optimization and external collaboration, and then combine their respective advantages to design a hybrid strategy to improve the performance of logistics enterprises in uncertain environments. “Robust optimization” considers uncertainty when working out routing decisions to ensure optimality in the worst-case scenario. Our innovation is to improve the traditional robust optimization method using machine learning techniques. “External collaboration” refers to the mechanism that many transportation companies cooperate to fight against uncertainty. The innovation lies in the proposed two-layer decision-making mechanism that can realize collaborative vehicle routing based on partial public information. Finally, a hybrid strategy is realized by introducing regularization parameters to form closed-loop linkage between the two mechanisms above. Simulation results and trial runs at enterprises will be used to verify the effectiveness of the new mechanisms and to derive the best hybrid strategy and managerial implications in different uncertain environments.

清洁物流是指对具有危险性、环境污染性或循环利用价值的废弃物进行回收的物流活动，在可持续发展的口号下，掀起了全球范围内的研究热潮。本项目关注运作层面的车辆路径规划问题，现有研究未能很好解决以下难题：废弃物的货量难以准确预估（需求随机性），往往车辆到达后才能获知（信息动态性），且可拆成多次回收（可拆分性）。为此，首先提出鲁棒优化与外部协同两种方法，然后结合各自优势，设计一种混合策略，以提升企业在不确定环境下的效益。“鲁棒优化”方法在制定运输计划时预先考虑不确定性，以保证在最坏情况下的方案最优，创新点在于使用机器学习的方法改进了鲁棒优化方法。“外部协同”则是指多家运输企业通过协作共同对抗不确定性，创新在于提出的双层决策机制可以基于部分公开信息实现协同路径规划。最后通过引入正则化参数实现了两种策略的闭环联动。结合仿真实验和企业试运行，验证新机制的效果，并得出不同环境下的最佳混合策略与管理建议。

在国家大力倡导可持续发展的背景下，本项目重点聚焦废弃物回收的路径规划问题。该问题作为清洁物流的核心，却未得到足够的重视。有别于传统研究，本项目充分考虑了清洁物流中“随机性”、“动态性”和“可拆分性”三大特征，探讨一家清洁物流企业在制定运输计划时，如何提升应对不确定环境的能力。上述特征均来源于真实的企业调研，因此本项目的研究成果具有很强的现实意义和广阔的应用前景。本项目创新性地将机器学习的方法引入鲁棒优化：运用支持向量聚类，从无穷多的可能性中识别出高概率区域，从而构建一个精确、紧凑的不确定集。这种新的不确定集构建方法，将有效提升解的鲁棒性，避免过分保守主义带来的效率下降，并显著降低鲁棒优化模型的求解难度。同时，创新性地将废弃物的“可拆分性”引入鲁棒优化和外部协同两种策略：对于鲁棒优化，可以通过安排多辆车合作回收的方式共同缓冲风险；对于外部协同，既可以将整个订单外协，也可以将其中的一部分请求外协。该特性的引入，既反映了清洁物流行业的真实特点，也丰富了现有研究体系。最后，创新性地提出结合鲁棒优化和外部协同的混合策略：充分发挥各自优势，使得清洁物流企业更好地平衡内、外部资源，应对环境不确定性。提出混合策略的架构，并基于试运营或实验仿真的结果对联动参数进行迭代优化，得到最优的混合策略。结合实际企业案例，对混合策略的有效性进行验证。