卷烟配送中心选址优化研究与应用

为提高卷烟物流配送效率，节约配送里程和时间，合理选择配送中心修建数量及地址，从而整体降低配送成本，利用多视角聚类算法结合节点重要度排序理论，构建了卷烟物流配送中心选址优化模型。针对客户点之间不同类型的连接关系，构造双层特征网络，使用共识的配送区域划分矩阵来提取不同性质数据矩阵的共同聚类划分，再用不同的低维系数矩阵表示不同性质数据的异质性，利用KKT原则和乘法更新原则进行矩阵更新迭代，使用hard-cluster方法进行聚类，实现配送区域划分，同时，引入非负矩阵分解的稳定性误差用于选取最优的配送区域个数划分方案；最后使用介数中心性原理，实现基于节点重要度的配送中心选址，以实现配送区域划分的均匀性，降低物流成本，提升物流效率。在云南省保山市进行仿真实验，结果表明:（1）利用本文方法实现了卷烟物流配送中心的选址和对应配送区域的划分，经过优化后，配送总里程减少了6089.097 km，等同于配送时间减少了165.6 h，所有配送中心对应的服务区域较均衡，不存在长距离的跨配送中心送货存在；（2）该地区卷烟销量及配送需求与聚类划分结果基本一致。（3）优化后的单位卷烟物流配送成本显著降低，配送效率也得到了较大程度的提升。因此，使用该方法来解决卷烟配送中心选址和配送区域优化问题是有效且可行的。 

Research and application of tobacco distribution center location optimization

To improve the efficiency of tobacco logistics distribution, save distribution mileage and time, reasonably determine the number and address of distribution centers, so as to reduce the overall distribution cost, a location optimization model of tobacco logistics distribution center was established by using a multi-view clustering algorithm and node importance ranking theory. According to the different types of connection relationship between customer points, a two-layer feature network was constructed. The common clustering partition of data matrices with different properties was extracted by using the common distribution area partition matrix. The heterogeneity of data with different properties was represented by different low dimensional coefficient matrices. The matrix updating iteration was carried out by using the KKT principle and multiplicative updating principle, and the hard cluster method was used for clustering. At the same time, the stability error of nonnegative matrix factorization was introduced to determine the division scheme for optimal distribution area number; Finally, using the principle of betweenness centrality, the location of distribution center based on node importance was realized to achieve the uniformity of distribution area division, reduce logistics cost and improve logistics efficiency. The simulation experiment in the Baoshan City of Yunnan Province shows that: 1) the location of tobacco logistics distribution center and the division of corresponding distribution area can be realized by using the proposed method. After optimization, the total distribution mileage is reduced by 6089.097 km, the distribution time is reduced by 165.6 hours, and the corresponding service areas of all distribution centers are relatively balanced, with no long-distance cross-distribution center deliveries; 2) The results of tobacco sales and distribution demand in this region are consistent with the clustering results; 3) The optimized unit tobacco logistics distribution cost is significantly reduced, and the distribution efficiency has been greatly improved. Therefore, the proposed method is effective and feasible to realize site selection of tobacco distribution center and optimization of tobacco distribution area.