A logistics distribution center site selection method and system based on HDBSCAN

Abstract

The present invention relates to a method and system for selecting a logistics distribution center based on HDBSCAN. The method comprises the following steps: 1) adjusting the sub-cluster control capability of HDBSCAN by using cluster splitting and cluster merging algorithms, so that the distribution volume of distribution centers at the same level is balanced, and using the center of gravity method to determine the distribution center for each distribution range; 2) realizing the balanced site selection layout of multi-level distribution centers by using this method iteratively. The present invention utilizes the characteristics of the HDBSCAN algorithm that it has clustering stability and can process data with different local densities, and applies it to the site selection of logistics distribution centers. In view of the problem that the HDBSCAN algorithm lacks the ability to balance the size of sub-clusters, and the hierarchical clustering tree generated in the clustering process cannot directly correspond to the sub-cluster center level, a corresponding improvement method is proposed, and finally a logistics distribution center site selection scheme that takes into account the hierarchy and the balanced distribution volume of distribution centers at each level is established.

Description

Technical Field

[0001] The invention belongs to the field of geographic information technology, and in particular relates to a method and system for selecting a logistics distribution center site based on HDBSCAN. Background Art

[0002] With the rapid development of economic globalization and network technology, the logistics and express delivery industry has rapidly developed into an important emerging pillar industry of the national economy and a new economic growth point. The development of the logistics industry is not only related to the social quality of life of residents, but also affects the development of social economy. It is also an indicator of a country's comprehensive national strength and degree of modernization. In recent years, while the logistics industry has achieved rapid development, transportation costs and warehousing and inventory costs have increased rapidly, accounting for more than half of the total cost, which has seriously hindered the improvement of corporate competitiveness. This has become an urgent problem to be solved in the development of enterprises. Therefore, how to effectively reduce logistics distribution costs and improve logistics operation efficiency has become a hot issue in the field of logistics distribution.

[0003] The logistics distribution center is an important node in the entire logistics distribution network system. It covers most of the basic logistics operations and is the most concentrated embodiment of logistics activities. Therefore, the reasonable location of the logistics distribution center can not only effectively reduce logistics costs and improve service quality, but also has important significance for improving the utilization efficiency of logistics resources.

[0004] At present, the site selection methods of logistics distribution centers can be mainly divided into two types: qualitative and quantitative. Among them, qualitative methods generally have the problem of excessive subjectivity in the site selection process, while quantitative methods can be attributed to solving an NP-hard mathematical problem, which is difficult to calculate and cannot guarantee that a solution can be found. In addition, most of these methods do not take into account the multi-level nature of the site selection of logistics distribution centers and the balanced characteristics of the logistics distribution volume of the same-level distribution centers, resulting in the calculated logistics distribution centers being difficult to meet the actual application needs. In response to these problems, the present invention intends to use an improved HDBSCAN algorithm to study the site selection of logistics distribution centers to solve the problems of many subjective judgments in the existing site selection process, the difficulty in calculating the optimal location, and the inability to take into account the multi-level balanced distribution of logistics distribution centers, thereby providing technical support for optimizing logistics site selection. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for selecting a logistics distribution center site based on HDBSCAN.

[0006] The technical solution of the present invention to solve the above technical problems is as follows:

[0007] A logistics distribution center site selection method based on HDBSCAN includes the following steps:

[0008] Step 1: Obtain the distribution information of the recipient address of a region, where the distribution information of the recipient address includes multiple data points, and each data point is a recipient address information;

[0009] Step 2: Use the HDBSCAN algorithm to divide all data points into multiple non-overlapping sub-clusters;

[0010] Step 3: Use cluster splitting and cluster merging algorithms to balance the sub-clusters and obtain the final sub-clusters;

[0011] Step 4: Assuming that the transportation cost is proportional to the transportation distance, use the transportation distance instead of the transportation cost, use the centroid method to obtain the coordinates of all data points in each sub-cluster and take the average value to get the sub-cluster center, and use all sub-cluster centers as the distribution center location to obtain the distribution center location result at this level;

[0012] Step 5: Determine whether there is a higher-level site selection requirement. If so, go to step 1 and replace the delivery address distribution information of the region in step 1 with the distribution center site selection result obtained in step 4 to calculate a higher-level site selection result. If not, the site selection ends.

[0013] Furthermore, the step 2 specifically includes the following steps:

[0014] Step 2.1, calculate the reachable distance between data points pairwise, take the data points as vertices, and the weight of the edge is equal to the mutual reachable distance of the vertices, and construct a distance weighted graph;

[0015] Step 2.2, use the prim algorithm to calculate the minimum spanning tree of the distance-weighted graph, sort the edges of the tree in ascending order by distance, create a new subcluster for each edge through union-find, and build a subcluster hierarchy;

[0016] Step 2.3, compress the cluster hierarchy according to the minimum subcluster size, traverse the hierarchy, delete all edges from the minimum spanning tree in descending order of weight, each time an edge is deleted, the subcluster is split, and it is determined whether the data points in the new connected component created by deleting the edge are less than those in the minimum subcluster. If so, the new connected component is declared as a false connected component, and the false component is marked as noise; and the subcluster is adjusted: if all components generated after the subcluster deletes the edge are false, then delete the subcluster; if there are both false components and true components in the components generated after the subcluster deletes the edge, then retain its original subcluster label, that is, retain the subcluster before removing the edge; if the components generated after the subcluster deletes the edge are not false, then assign a new subcluster label to each component, that is, successfully split the original subcluster into a new subcluster;

[0017] Step 2.4: Traverse the tree in reverse topological sorting order to determine whether the stability of each subcluster's parent cluster is greater than the sum of its child nodes. If so, declare the subcluster as a selected subcluster and merge all its descendants. After reaching the root node, all selected subclusters are obtained.

[0018] Furthermore, the step 3 specifically includes the following steps:

[0019] Traverse all selected subclusters, and according to the preset capacity value v, merge the subclusters with a size smaller than v / 2 with the nearest subcluster, and split the subclusters with a size larger than v. Traverse multiple times until the subclusters do not change, and the remaining subclusters are the final subclusters.

[0020] The beneficial effects of the present invention are as follows: the present invention proposes corresponding improvements to address the problem that the HDBSCAN algorithm lacks the ability to balance the size of sub-clusters, and that the hierarchical clustering tree generated during the clustering process cannot directly correspond to the hierarchy of the sub-cluster center, so as to achieve a logistics distribution center site selection plan that takes into account the hierarchy and the balanced distribution volume of distribution centers at each level. The main advantages of the present invention are as follows: 1) By using cluster splitting and cluster merging algorithms to adjust the sub-cluster control ability of HDBSCAN, the distribution volume of distribution centers at the same level is balanced, and the center of gravity method is used to determine the distribution center for each distribution range; 2) By iteratively using this method, a balanced site selection layout of multi-level distribution centers is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0022] Figure 2 This is a schematic diagram of the clustering effect obtained by the method of the present invention; DETAILED DESCRIPTION

[0023] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0024] After years of development, most logistics companies have adopted a multi-level sorting and distribution system: express parcels are delivered from the highest-level regional transit warehouse to the first-level regional sorting center, the second-level regional sorting center, the terminal express station, and finally to the demand point. It can be seen from the architecture of the logistics network and the modern logistics distribution process that in the process of logistics distribution, the goods are transported from the highest-level regional transit warehouse to the sorting center, and then transported to the terminal express station after sorting, and finally delivered to the customer base. Goods are configured with agent factories in each region as much as possible. The goods required by customers are produced by the factory and transported to the transit warehouse of the corresponding region. After the goods are distributed according to customer needs, the express delivery goes through the sorting centers of provinces, cities and districts and finally arrives at the express station or express cabinet in the customer's residential area; and in the process of logistics express operation and distribution, there will be no hierarchical transportation across logistics distribution centers.

[0025] This paper mainly aims at the needs of logistics distribution center site selection problem and proposes a logistics distribution center site selection method based on HDBSCAN. Figure 1 As shown, the following steps are included:

[0026] 1. Select the location of the lowest level (kth level) distribution center

[0027] 1. Dataset acquisition

[0028] The validation dataset uses the official dataset provided by the 2018 Global Operational Optimization Challenge, which includes serial number, longitude and latitude information, package information, and merchant delivery time window information.

[0029] 2. Use the improved HDBSCAN algorithm to divide the distribution range

[0030] Construct a data-weighted graph in which data points are vertices and the weight of the edge between any two points is equal to the mutually reachable distance between these points. Then, construct a minimum spanning tree of the distance-weighted graph, sort the edges of the minimum spanning tree in ascending order by distance, and then iterate to create a new subcluster for each edge by union-find. Compress the cluster hierarchy according to the minimum subcluster size, traverse the hierarchy, and judge each time the subcluster is split, and only the splits that meet the minimum subcluster size are retained. Finally, extract stable clusters from the compressed tree and traverse the tree in reverse topological sorting order. If the stability of the parent-child cluster is greater than the sum of its child nodes, declare the subcluster as the selected subcluster and merge all its descendants. After reaching the root node, use the currently selected cluster as the clustering result.

[0031] The subclusters are adjusted according to the input parameter capacity v. After the HDBSCAN algorithm extracts stable clusters from the compressed tree, all generated subclusters are traversed. According to the given capacity value v, subclusters with a size smaller than v / 2 are merged with the nearest subcluster, and subclusters with a size larger than v are split. Since both the cluster merging and cluster splitting algorithms may generate new subclusters whose subcluster sizes are not in the interval [v / 2,v], it is necessary to traverse multiple times to obtain a balanced clustering result.

[0032] 3. Calculate the logistics distribution center using the center of gravity method

[0033] The transportation distance is used instead of the transportation cost, so the transportation cost is Take the partial derivatives of x and y, and find the point where the partial derivative is 0. The extreme value theorem shows that the transportation cost is the minimum at this time. The transportation cost is minimized when the average value of the coordinates of the data points in each subcluster is taken, which is the center of the required subcluster, that is, the address of the logistics distribution center. Use the centroid method to determine the logistics distribution center corresponding to each subcluster, that is, the distribution range.

[0034] 2. Select a location for a higher-level distribution center

[0035] The site selection results of the kth layer are used as the input data for the site selection of a higher layer (k-1th layer). The iterative method is used to change the input and repeat the two steps of dividing the radiation range and selecting the location of the distribution center within the radiation range to obtain the site selection plan for the multi-level logistics distribution center from the bottom up.

[0036] 3. Experimental Results

[0037] The experiment was conducted using the official data set provided by the Global Operational Optimization Challenge. Only the longitude and latitude information was taken and these 1091 data were used as the addresses of the underlying logistics distribution centers (express cabinets or express stations). The multi-level logistics distribution center site selection method based on HDBSCAN proposed in the present invention was used for site selection. In this case, it is necessary to decide on two levels of distribution centers. According to the comparison of the carrying capacity of the first-level logistics distribution center and the second-level logistics distribution center, i.e., the terminal sorting center, and the terminal express station or express cabinet, when selecting the site of the second-level logistics center (second-level regional sorting center), the capacity value is set to 120, and when selecting the site of the first-level logistics distribution center (first-level regional sorting center), the capacity value is set to 3. Finally, a decision plan for 9 second-level regional sorting centers and 4 first-level regional sorting centers was obtained. Since it is relatively simple to select four first-level logistics distribution centers based on nine second-level logistics distribution centers in the second layer, the main analysis is whether the site selection of the second-level logistics distribution center is reasonable. The distribution range division of the second-level logistics distribution center, i.e., the first clustering result of the algorithm iteration, is as follows:

[0038] Table 1 Algorithm results

[0039]

[0040] From Table 1, we can see that the results obtained by the method of the present invention show that except for one sub-cluster with too many points, the other sub-clusters are relatively balanced. The size of the sub-cluster represents the number of demand points within the distribution range of the distribution center, which is proportional to the distribution volume of the distribution center. Therefore, the distribution volume of the logistics distribution center generated by the method of the present invention is relatively balanced. The distribution efficiency of the logistics distribution network is high after it is put into use, and the distribution resources of each logistics node will not be wasted. The address of each data point and its cluster are visualized, and the results are as follows Figure 2 shown.

[0041] The experimental results show that the multi-level logistics distribution center site selection scheme generated by the HDBSCAN clustering algorithm proposed in the present invention has a good ability to control the distribution volume balance of logistics distribution centers at the same level, and can solve the multi-level site selection problem of logistics distribution centers.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A logistics distribution center site selection method based on HDBSCAN, characterized in that: The following steps are involved: Step 1: Obtain the distribution information of the recipient address of a region, where the distribution information of the recipient address includes multiple data points, and each data point is a piece of recipient address information; Step 2: Use the HDBSCAN algorithm to divide all data points into multiple non-overlapping sub-clusters; The step 2 specifically includes the following steps: Step 2.1, calculate the reachable distance between data points pairwise, take the data points as vertices, and the weight of the edge is equal to the mutual reachable distance of the vertices, and construct a distance weighted graph; Step 2.2, use the prim algorithm to calculate the minimum spanning tree of the distance-weighted graph, sort the edges of the tree in ascending order by distance, create a new subcluster for each edge through union-find, and build a subcluster hierarchy; Step 2.3, compress the cluster hierarchy according to the minimum subcluster size, traverse the hierarchy, delete all edges from the minimum spanning tree in descending order of weight, each time an edge is deleted, the subcluster is split, and it is determined whether the data points in the new connected component created by deleting the edge are less than those in the minimum subcluster. If so, the new connected component is declared as a false connected component, and the false component is marked as noise; and the subcluster is adjusted: if all components generated after the subcluster deletes the edge are false, then delete the subcluster; if there are both false components and true components in the components generated after the subcluster deletes the edge, then retain its original subcluster label, that is, retain the subcluster before removing the edge; if the components generated after the subcluster deletes the edge are not false, then assign a new subcluster label to each component, that is, successfully split the original subcluster into a new subcluster; Step 2.4, traverse the tree in reverse topological sorting order to determine whether the stability of each sub-cluster's parent cluster is greater than the sum of its child nodes. If so, declare the sub-cluster as a selected sub-cluster and merge all its descendants. After reaching the root node, all selected sub-clusters are obtained; Step 3: Use cluster splitting and cluster merging algorithms to balance the sub-clusters and obtain the final sub-clusters; Step 4: Assuming that the transportation cost is proportional to the transportation distance, use the transportation distance instead of the transportation cost, use the centroid method to obtain the coordinates of all data points in each sub-cluster and take the average value to get the sub-cluster center, and use all sub-cluster centers as the distribution center location to obtain the distribution center location result at this level; Step 5: Determine whether there is a higher-level site selection requirement. If so, go to step 1 and replace the delivery address distribution information of the region in step 1 with the distribution center site selection result obtained in step 4 to calculate a higher-level site selection result. If not, the site selection ends.

2. The method for selecting a logistics distribution center based on HDBSCAN according to claim 1 is characterized in that: The step 3 specifically includes the following steps: Traverse all selected subclusters, and according to the preset capacity value v, merge the subclusters with a size smaller than v / 2 with the nearest subcluster, and split the subclusters with a size larger than v. Traverse multiple times until the subclusters do not change, and the remaining subclusters are the final subclusters.

3. A logistics distribution center site selection system based on HDBSCAN, characterized in that: It includes an information acquisition module, a subcluster division module, a subcluster adjustment module, a distribution center location selection module and a judgment module. The information acquisition module is used to obtain the distribution information of the recipient address of an area. The recipient address distribution information contains multiple data points, and each data point is a recipient address information; the subcluster division module is used to use the HDBSCAN algorithm to divide all data points into multiple non-overlapping subclusters, and the HDBSCAN algorithm includes the following steps: Step 2.1, calculate the reachable distance between data points in pairs, take the data points as vertices, and the weight of the edge is equal to the mutual reachable distance of the vertices to construct a distance weighted graph; Step 2.2, use the prim algorithm to calculate the minimum spanning tree of the distance weighted graph, sort the edges of the tree in ascending order by distance, create a new subcluster for each edge by union-find, and construct a subcluster hierarchy; Step 2.3, compress the cluster hierarchy according to the minimum subcluster size, traverse the hierarchy, and select the subcluster from the minimum spanning tree according to the minimum subcluster size. Delete all edges in descending order of weight. Each time an edge is deleted, a subcluster is split. Determine whether the data points in the new connected component created by deleting the edge are less than those in the smallest subcluster. If so, declare the new connected component as a false connected component and mark the false component as noise. Perform subcluster adjustment: If all components generated after deleting the edge of the subcluster are false, delete the subcluster. If there are both false and true components in the components generated after deleting the edge of the subcluster, retain its original subcluster label, that is, retain the subcluster before removing the edge. If none of the components generated after deleting the edge of the subcluster are false, assign a new subcluster label to each component, that is, successfully split the original subcluster into a new subcluster. Step 2.4: Traverse the tree in reverse topological sorting order to determine whether the stability of the parent-child cluster of each subcluster is greater than the sum of its child nodes. If so, declare the subcluster as a selected subcluster and merge all its descendants. After reaching the root node, all selected subclusters are obtained. The subcluster adjustment module is used to use cluster splitting and cluster merging algorithms to balance the subclusters and obtain the final subclusters; the distribution center location selection module is used to use transportation distance instead of transportation cost, use the center of gravity method to obtain the coordinates of all data points in each subcluster and take the average value to obtain the subcluster center, and use all subcluster centers as distribution center locations to obtain the distribution center location selection results at this level; the judgment module is used to judge whether there is a higher level location selection demand after the distribution center location selection at this level is completed. If so, the distribution center location selection results at this level will be used to replace the regional delivery address distribution information in the information acquisition module to calculate the higher level location selection results, otherwise the location selection ends.

Images