摘要
在绿色供应链网络中,因市场需求波动、环境政策变化、供应商生产能力不稳定以及自然灾害等不可预见事件等多方面复杂因素交织作用,导致多源不确定性,多主体间静态或动态采用单一方式求解设计,难以精准捕捉不同主体间在面对这些不确定性时的动态响应机制以及各不确定性因素间的复杂交互关系。为此,提出一种基于图神经网络(GNN)随机优化的不确定性绿色供应链网络多主体均衡设计方法。首先,将含制造商、配送中心与客户的三级网络建模为时序异构图,借助时序图注意力网络自适应提取节点与边在多维不确定性中的隐式关联特征,生成具有高表征精度的不确定性量化参数;进而,构建两阶段分布式鲁棒优化模型:第一阶段基于特征驱动实现配送中心选址与客户分配的静态优化,第二阶段面向不确定性场景动态调整物流分配,并引入列与约束生成算法高效求解该多目标均衡问题。实验表明,所提方法在10类不确定性场景中关联熵均值达0.95,验证了特征编码的有效性;经均衡设计后,运输成本与运输时间最大降幅分别为34.5%与45.4%,绿色度损失平均降低47.4%,显著提升了制造商、配送中心与客户间的利益协同水平,为高不确定性环境下的绿色供应链网络提供了兼具解释性、自适应性与全局均衡性的决策支持。
In the green supply chain network,due to complex factors such as market demand fluctuations,environmental policy changes,unstable supplier production capacity,and unforeseeable events such as natural disasters,multiple sources of uncertainty are intertwined.Static or dynamic solutions are designed in a single way among multiple entities,making it difficult to accurately capture the dynamic response mechanisms of different entities in the face of these uncertainties and the complex interaction relationships between various uncertain factors.Therefore,a multi-agent equilibrium design method for uncertain green supply chain networks based on graph neural network(GNN)random optimization is proposed.Firstly,a three-level network consisting of manufacturers,distribution centers,and customers is modeled as a temporal heterogeneous graph.With the help of a temporal graph attention network,implicit correlation features between nodes and edges in multidimensional uncertainty are adaptively extracted,generating uncertainty quantification parameters with high representation accuracy;Furthermore,a two-stage distributed robust optimization model is constructed:the first stage is based on feature driven static optimization of distribution center location and customer allocation,and the second stage dynamically adjusts logistics allocation for uncertain scenarios,and introduces column and constraint generation algorithms to efficiently solve the multi-objective equilibrium problem.The experiment shows that the proposed method achieves an average correlation entropy of 0.95 in 10 types of uncertain scenarios,verifying the effectiveness of feature encoding;After balanced design,the maximum reduction in transportation costs and transportation time was 34.5%and 45.4%,respectively,and the average reduction in greenness loss was 47.4%.This significantly improved the level of benefit coordination between manufacturers,distribution centers,and customers,providing explanatory,adaptive,and globally balanced decision support for green supply chain networks in high uncertainty environments.
作者
夏景
张睿云
吴晓畅
XIA Jing;ZHANG Rui-yun;WU Xiao-chang(China Academy of Industrial Internet,Beijing 100080,China)
出处
《机电产品开发与创新》
2026年第2期36-42,共7页
Development & Innovation of Machinery & Electrical Products
关键词
绿色供应链
多主体均衡
图神经网络
分布式鲁棒优化
不确定性
Green supply chain
Multi-agent equilibrium
Graph neural network
Distributed robust optimization
uncertainty
