A method for selecting a logistics center based on a hybrid clustering algorithm

Abstract

The application discloses a method for selecting a logistics center based on a hybrid clustering algorithm, which comprises the following steps: step 1, combining the process of finding density extreme points of a MeanShift algorithm with the process of distributing non-density extreme points of a DPC algorithm to control the number of clustering; step 2, performing data cleaning on logistics center data; step 3, obtaining a real road site distance data matrix according to application data and by means of a third-party map platform; step 4, combining the real road site distance data matrix A according to step 1; step 5, establishing a joint model of clustering density extreme points and maximum express quantity data, and maximizing the express quantity within a threshold T and minimizing the sum of distances from sites to the density extreme points by means of a maximization algorithm; and step 6, fusing the processed MS-DPC algorithm with the joint model, and finally obtaining a prediction result of the address of the logistics center by means of model training and testing. The application solves the misclassification problem that may occur in the current hybrid clustering algorithm.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of site selection problems and data mining, specifically involving a method for logistics center site selection based on a hybrid clustering algorithm. Background Technology

[0002] The extremely high time sensitivity required for fresh produce and cold chain delivery has placed higher standards on the logistics and distribution industry, prompting major logistics and distribution companies to continuously improve their delivery routes and methods. Logistics centers are crucial hubs in the entire process of parcel delivery, playing a vital role in connecting upstream and downstream processes. A suitable logistics center can minimize economic and time costs during transshipment, transportation, and delivery, thereby improving efficiency.

[0003] Logistics center site selection differs from other traditional site selection problems. Based on the logistics architecture, it is generally divided into four levels of distribution: the first level is the smart warehouse, the largest storage center, storing almost all transportable goods; the second level is the provincial / municipal logistics center, responsible for transshipment and transportation of parcels within the province / city; the third level is the regional distribution stations, responsible for scheduling all parcels within their region; and the fourth level is the grassroots logistics stations, where couriers handle short-distance deliveries. However, establishing logistics distribution centers requires huge investments, and different decision-makers may have different outcomes. Subjective decision-making can lead to significant resource waste. Clustering in data mining is an important decision-making aid. By mining the coordinates of existing stations and applying clustering algorithms, with distance as the main influencing factor, it uncovers inherent implicit relationships, allowing for the construction of a reasonable number of logistics centers. This reduces path delivery time and distance costs, improves platform efficiency and profits, increases user satisfaction, and further alleviates urban traffic congestion.

[0004] In recent years, site selection has been a subject of much research, resulting in numerous proposed methods. These include heuristic algorithms such as the propagation algorithm, ant colony algorithm, and particle swarm optimization, as well as methods like the centroid method, p-median method, and p-center method. However, these methods can only perform calculations based on theoretical distances and cannot consider changes in external factors in real-world conditions. Later, research began to consider constructing joint models that capture external factors for calculation, such as the K-means++ method and the Meanshift+K-means hybrid clustering method. These methods divide the city's coordinates and incorporate external factors for site selection. However, most of these models rely on Euclidean distance calculations and do not overcome the inherent flaws of the Meanshift and K-means algorithms, such as the potential for misclassification. Furthermore, most studies predict new sites based on existing sites, without considering the replanning of all sites in a city, thus failing to identify substantively unreasonable sites. Summary of the Invention

[0005] To address the key issues that remain unresolved, this invention proposes a method for logistics center site selection based on a hybrid clustering algorithm. The aim is to resolve the imbalance in demand distribution in cities, the errors in Euclidean distance in the model, and improve the accuracy of site selection.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for logistics center site selection based on a hybrid clustering algorithm includes the following steps:

[0008] Step 1: Propose a density- and distance-based hybrid clustering application method (MS-DPC), which combines the process of finding density extrema in the MeanShift algorithm with the process of allocating non-density extrema in the DPC algorithm to control the number of clusters;

[0009] Step 2: Preprocess coordinate data and clean the data of the logistics center, mainly retaining the attributes of location information, parcel saturation, and throughput.

[0010] Step 3: Based on the application data, using a third-party map platform, obtain an n×n data matrix of actual road site distances, denoted as A=(a ij ) m×n Where n represents the number of logistics centers, a ij This represents the actual road network distance between station i and station j.

[0011] Step 4: Based on the method proposed in Step 1, and combining the n×n data matrix A of actual road site distances, replace the Euclidean distance in the Gaussian kernel function to obtain ND cluster density extreme points, i.e., Ψ={dep1,dep2,…,dep ND}. Where dep(p) represents the p-th density extremum point;

[0012] Step 5: Establish a joint model of cluster density extreme points and maximum parcel quantity data. By using a maximization algorithm, maximize the parcel quantity within a threshold T while minimizing the sum of distances from the station to the density extreme point. This ensures that MS-DPC considers not only the parcel quantity factor but also the actual path conditions, rather than just geographical proximity.

[0013] Step 6: Fuse the processed MS-DPC algorithm with the joint model, and obtain the final prediction result of the logistics center address through model training and testing.

[0014] As a preferred approach, the hierarchical clustering method used in step 1 includes the following:

[0015] Step 1-1: Given a d-dimensional space R containing n points dThe dataset X = {x1, x2, ... x} n Let x be a randomly selected point. Consider a d-dimensional sphere centered at x. Assume there are m points inside the sphere besides x, with x as the starting point. i Let m vectors be created with the endpoint as the endpoint. Adding these m vectors together yields the Mean shift vector. The basic form of the Mean shift vector is then: Where M represents a vector, x i S represents the position coordinates of a point. k S is a high-dimensional spherical region with radius r, h is a parameter that can be adjusted according to the density of the dataset, and k indicates that k points fall into S. k Within a region, the Meanshift vector is the sum of the vectors from each point within the region to the origin;

[0016] Steps 1-2: [Regarding...] center (t+1) =center (t) +shift (t) center (t) Let the Mean shift vector at time t be the origin. Then, construct a high-dimensional spherical region with radius r, using the endpoint of the Mean shift vector at time t as the origin. Repeat the above steps to obtain a new Mean shift vector, shift. (t) The sum of the two vectors gives the Meanshift vector at time t+1, with the endpoint being center. (t+1) After multiple iterations, the Mean Shift algorithm can converge to the point where the probability density is maximum; let this point be the density extremum point, dep.

[0017] Steps 1-3: Add the Gaussian kernel function to the basic Mean shift vector, then the original vector form is transformed into... in It is a Gaussian kernel function, which is affected by bandwidth h and distance. The smaller the bandwidth h and the closer the distance, the larger the function value.

[0018] Steps 1-4: Replace the Euclidean distance in the Gaussian kernel function with adaptive distance. Utilize the density reachability concept of the DBSCAN algorithm: if sample X is within the neighborhood of core sample Y, then Y to X is density-reachable; if sample Y is within the neighborhood of core sample Z, then sample X and sample Z are density-reachable. Specifically, the adaptive distance is: if two points are density-reachable, the adaptive distance is equal to the Euclidean distance; if two points are density-inaccessible, the adaptive distance is an exponential multiple of the Euclidean distance.

[0019] Steps 1-5: Repeat steps 1-2 and 1-3 until no new density extrema appear, and obtain the density extrema set dep(p);

[0020] Steps 1-6: For all non-density extrema x in the dataset X, calculate the distance L from point x to each density extrema dep(p). x→dep(p) ;

[0021] Steps 1-7: Using the classification idea of ​​the DPC algorithm, classify point x into L. x→dep(p) The minimum density extremum point dep, and all non-density extremum points x classified as that point are clustered into one class, denoted as C. p The output clustering result Φ = {C1, C2, ... C} p};

[0022] Preferably, step 2, the data preprocessing stage, includes the following:

[0023] Step 2-1: Clean the coordinate data and parcel quantity data, remove irrelevant attributes from the data, and mainly retain the location information, parcel saturation, and throughput. Remove null values ​​and fill in missing values. It can be represented as D = (longitude, latitude, saturation, throughput, spare), where longitude and latitude are latitude and longitude information, saturation and throughput are saturation and throughput respectively, and spare is other optional attributes.

[0024] Preferably, step 3, obtaining the data matrix, includes the following:

[0025] Step 3-1: Based on the coordinate data D processed in Step 2-1, which is randomly distributed throughout the city, use a third-party map platform to obtain an n×n data matrix of real road site distances, denoted as A=(a ij ) m×n , where a ij This represents the actual road network distance between station i and station j.

[0026] As a preferred embodiment, step 4, which involves obtaining the density extremum point set based on the data matrix, includes the following:

[0027] Step 4-1: Since this problem involves actual distances, change the Euclidean distance of the Gaussian kernel function in Step 1-4 to the actual road network distance processed in Step 3-1. This makes the calculation of the correlation between logistics stations more realistic;

[0028] Step 4-2: Randomly select x from the dataset i Perform steps 1-5 until no new density extrema appear, obtaining the density extrema set Ψ = {dep1, dep2, ..., dep}. p}, where p≤n;

[0029] As a preferred embodiment, step 5, which establishes a joint model of cluster density extreme points and express delivery quantity data, includes the following steps:

[0030] Step 5-1: Based on steps 1-7, obtain the clustering result Φ = {C1, C2, ... C}. p By determining the number of logistics stations m in the area, the number of density extreme points is adjusted to make p = m;

[0031] Step 5-2: Based on the saturation of parcels at the logistics station and the adjusted m density extreme points, use the maximization algorithm to reclassify D = (longitude, latitude, saturation, throughput, spare) to obtain Φ = {C1, C2, ... C}. m}, making it classified as C m The sum of the parcel throughput of the number of stations is less than the parcel saturation of the logistics center, and the sum of the actual road network distances from all stations to the logistics center to which they belong is minimized. This is denoted as .

[0032] Preferably, step 6, which integrates the predictions of the algorithm and the model, includes the following:

[0033] Step 6-1: Combine the processed MS-DPC algorithm with the joint model, and through model training and testing, finally obtain the prediction result of the logistics center address;

[0034] Step 6-2: Train the model multiple times, using AC, PR, and RE as accuracy evaluation metrics to predict the accuracy of the number of different logistics centers.

[0035] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0036] To address the current problem of logistics site selection, this invention proposes a hybrid clustering algorithm based on Mean Shift and DPC algorithms, resolving potential misclassification issues in existing hybrid clustering algorithms. Furthermore, it employs a road network matrix, using actual road network distances instead of Euclidean distances in the Gaussian kernel function for offsetting density extrema. By incorporating constraints on the number of parcels at each logistics center, the algorithm achieves a more balanced distribution of logistics centers within the city, effectively improving prediction accuracy under real-world conditions and facilitating subsequent experimental research. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the logistics center site selection process based on a hybrid clustering algorithm according to the present invention.

[0038] Figure 2 This is the actual road network data matrix between some stations in Jinan City in this invention.

[0039] Figure 3 This is a schematic diagram of the MS-DPC hybrid clustering algorithm proposed in this invention.

[0040] Figure 4 This is a schematic diagram of the grassroots logistics stations in Jinan City in this invention.

[0041] Figure 5 This is a schematic diagram illustrating the prediction of the logistics center in Jinan City in this invention.

[0042] Figure 6 This is a schematic diagram showing the location of the logistics center in the clustering planning of this invention. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0044] like Figure 1 As shown, the technical route of this invention mainly includes six steps: proposing a hybrid clustering application method, preprocessing data, constructing a real road network data matrix, calculating the density extreme point set, constructing a joint model, and training and predicting the fusion algorithm and model.

[0045] This example uses Cainiao Station data from Gaode Maps for testing. The following sections will introduce this example from three aspects: constructing a real road network matrix, model building and training, and prediction results.

[0046] 1) Propose a hybrid clustering algorithm

[0047] First, based on the coordinate information data analysis of Cainiao Post Stations on Gaode Map, the size and function of the stations are similar. Based on the actual analysis, the actual road network distance between the two stations needs to be replaced by the adaptive distance of the hybrid clustering algorithm. This can prevent the straight-line distance between the two stations from being too close when performing the Mean shift algorithm drift. However, in reality, there are uncontrollable factors such as large rivers or inconvenient transportation between the stations.

[0048] This invention proposes a hybrid clustering application method based on density and distance (MS-DPC). The improvements of the MS-DPC algorithm are mainly reflected in two aspects: First, it combines the advantages of the Mean Shift algorithm and the DPC algorithm to overcome the problems of incorrect selection and misclassification of extreme points; second, it improves the Euclidean distance to an adaptive distance to prevent the problem of incorrect assignment, and uses the actual road network distance instead of the adaptive distance in the location selection of logistics centers.

[0049] Given a d-dimensional space R with n points d The dataset X = {x1, x2, ... x} n If a point x i If the local density of x is greater than the density of its surroundings, and it is far from other points with high local density, then x i Let dep be a density extremum point. By calculating all dep values, we obtain the density extremum point set Ψ = {dep1, dep2, ..., dep}. ND By adjusting ND, the remaining points are classified to obtain the final clustering result Φ={C1,C2,…C M The clustering process flowchart is as follows: Figure 3 As shown, the specific steps are as follows:

[0050] Step 1: Given a d-dimensional space R containing n points d The dataset X = {x1, x2, ... x} n}, then the basic form of the Meanshift vector is Where S k S is a high-dimensional spherical region with radius h, where k represents the number of points falling into S. k Within a region, the Mean shift vector is the sum of the vectors from each point within the region to the origin;

[0051] Step 2: For center (t+1) =center (t) +shift (t) Using the endpoint of the Mean shift vector as the origin, construct a high-dimensional sphere and repeat the above steps to obtain another Mean shift vector. By repeating this process, the Mean shift algorithm can converge to the point where the probability density is maximized; this point is the density extremum, dep.

[0052] Step 3: Add the Gaussian kernel function to the basic Mean Shift vector, then the original vector form is transformed into... in The Gaussian kernel function is affected by bandwidth h and distance; the smaller the bandwidth h and the closer the distance, the larger the function value.

[0053] Step 4: Change the Euclidean distance in the Gaussian kernel function to the adaptive distance. Specifically: if the density between two points is reachable, the adaptive distance is equal to the Euclidean distance; if the density between two points is not reachable, the adaptive distance is an exponential multiple of the Euclidean distance.

[0054] Step 5: Repeat steps 2 and 3 until no new density extrema appear, and obtain the density extrema set dep(p);

[0055] Step 6: For all points x in dataset X, calculate the distance L from point x to dep(p). x→dep(p) ;

[0056] Step 7: Using the classification idea of ​​the DPC algorithm, classify point x into L. x→dep(p) The minimum dep outputs the clustering result Φ = {C1, C2, ... C}. M};

[0057] 2) Model building and training

[0058] First, a logistics station has a maximum number of parcels it can receive. The average number of parcels received is determined by the historical number of parcels received. The sum of the maximum number of parcels received by logistics stations within the logistics center should be less than the maximum number of parcels received by the logistics center. Therefore, the maximum number of parcels received by a logistics station is used as a threshold condition to ensure that the logistics stations within the logistics center are controllable and that the number of parcels is within their own capacity.

[0059] The coordinate data and the number of express items are validated by deleting irrelevant attributes, removing null and outlier values, and filling in missing values. This can be represented as D = (longitude, latitude, saturation, throughput, spare), where longitude and latitude are latitude and longitude information, saturation and throughput are saturation and throughput respectively, and spare is other optional attributes.

[0060] Using a third-party map platform, obtain an n×n data matrix of real road site distances, denoted as A=(a ij ) m×n , where a ij Represents the actual road network distance between station i and station j, such as Figure 2 As shown. Since this problem involves real-world distances, the Euclidean distance of the Gaussian kernel function is changed to the processed real road network distance, making the calculation of the correlation between logistics stations more realistic.

[0061]

[0062] Randomly select x from the dataset i The MS-DPC algorithm is iterated until no new density extrema appear, resulting in a density extrema set Ψ = {dep1, dep2, ..., dep}. ND}, where ND≤n. By determining the number m of logistics stations in the area, the number of density extrema points is adjusted to make ND=m. By investigating the actual situation of the maximum parcel quantity threshold T of the logistics stations, based on the adjusted m density extrema points, a maximization algorithm is used to maximize the parcel quantity within the threshold T while minimizing the sum of the distances from the station to the density extrema point, θ. m =argmax(T,L,x i ), θ m This represents the number of sites x contained in each cluster within the m-th cluster. i .

[0063] 3) Prediction Results

[0064] The processed MS-DPC algorithm is fused with a joint model. Through model training and testing, the final prediction result of the logistics center address is obtained. First, accuracy (AC), precision (PR), and recall (RE) are used to evaluate the prediction performance, explained in detail below:

[0065]

[0066]

[0067]

[0068] In formula AC, a i These are correctly classified samples. K is the number of clusters, and n is the number of data points in the dataset. In formulas PR and RE, K is the number of clusters, and n is the number of data points in the dataset. i It was correctly assigned to class C. i The number of objects (1≤i≤K), b i It is incorrectly assigned to class C. i The number of objects, c i It should be in class C. i However, it does not contain the number of objects that were correctly assigned to it. For AC, PR, and RE, higher values ​​indicate better clustering quality, and a value of 1 indicates that the clustering result is completely correct. Furthermore, we use clustering time and the number of iterations to evaluate the efficiency of each algorithm.

[0069] This method conducts predictive experiments using coordinate data and parcel volume data of Cainiao Post stations in Jinan, and re-clusters and plans the locations of third-level logistics centers, such as... Figure 5 and Figure 6 As shown in the figure, the positions marked with × represent the predicted locations of logistics centers. Figure 5 The number of logistics centers is 9. Figure 6The number of logistics centers is 35. Cainiao stations in the city are generally scattered, but more densely packed in the urban area. Under the premise of ensuring that the maximum number of parcels in the logistics center does not exceed the threshold, the minimum distance sum of each area is reasonably allocated. As shown in the figure, the MS-DPC hybrid clustering algorithm can cluster suitable areas regardless of the size of K, and the prediction is relatively accurate and close to the actual situation.

[0070] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art should understand that certain modifications and variations can be made to the present invention without departing from the spirit and substance of the present invention, but these modifications and variations should be covered within the scope of protection of the present invention.

Claims

1. A method for logistics center site selection based on a hybrid clustering algorithm, characterized in that, Includes the following steps: Step 1: Propose a hybrid clustering application method based on density and distance, MS-DPC, which combines the process of finding density extrema in the Mean Shift algorithm with the process of allocating non-density extrema in the DPC algorithm to control the number of clusters; Step 2: Preprocess coordinate data and clean the data of the logistics center, mainly retaining the attributes of location information, parcel saturation, and throughput. Step 3: Based on the application data, using a third-party map platform, obtain an n×n data matrix of actual road site distances, denoted as A=(a ij ) m×n Where n represents the number of logistics centers, a ij This represents the actual road network distance between station i and station j. Step 4: Based on the method proposed in Step 1, and combining the n×n data matrix A of actual road site distances, replace the Euclidean distance in the Gaussian kernel function to obtain ND cluster density extreme points, i.e., Ψ={dep1,dep2,…,dep ND }; where dep(p) represents the p-th density extremum point; Step 5: Establish a joint model of cluster density extreme points and maximum number of parcels. By using a maximization algorithm, the number of parcels is maximized within the threshold T while the sum of the distances from the station to the density extreme point is minimized. This allows MS-DPC to consider not only the number of parcels but also the actual path in real-world conditions, rather than just geographical proximity. Step 6: Fuse the processed MS-DPC algorithm with the joint model, and obtain the final prediction result of the logistics center address through model training and testing.

2. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 1, characterized in that, Step 1 includes the following methods: Step 1-1: Given a d-dimensional space R containing n points d The dataset X = {x1, x2, ... x} n Let x be a randomly selected point. Consider a d-dimensional sphere centered at x. Assume there are m points inside the sphere besides x, with x as the starting point. i Let m vectors be created with the endpoint as the endpoint. Adding these m vectors together yields the Mean shift vector. The basic form of the Mean shift vector is then: Where M represents a vector, x i S represents the position coordinates of a point. k S is a high-dimensional spherical region with radius r, h is a parameter that can be adjusted according to the density of the dataset, and k indicates that k points fall into S. k Within a region, the Meanshift vector is the sum of the vectors from each point within the region to the origin; Steps 1-2: [Regarding...] center (t+1) =center (t) +shift (t) center (t) Let t be the Mean shift vector at time t. Taking the endpoint of the Mean shift vector at time t as the origin, construct a high-dimensional spherical region with radius r, and repeat the above steps to obtain a new Mean shift vector shift. (t) The sum of the two vectors gives the Mean shift vector at time t+1, with the endpoint being center. (t+1) After multiple iterations, the Mean Shift algorithm can converge to the point where the probability density is the highest. Let this point be the density extremum point, dep. Steps 1-3: Add the Gaussian kernel function to the basic Mean shift vector, then the original vector form is transformed into... in It is a Gaussian kernel function, which is affected by bandwidth h and distance. The smaller the bandwidth h and the closer the distance, the larger the function value. Steps 1-4: Change the Euclidean distance in the Gaussian kernel function to adaptive distance. Utilize the density reachability concept of the DBSCAN algorithm: if sample X is in the neighborhood of core sample Y, then Y to X is density-directly reachable; if sample Y is in the neighborhood of core sample Z, then sample X and sample Z are density-reachable. The concept of adaptive distance is as follows: if two points are density-reachable, then the adaptive distance is equal to the Euclidean distance; if two points are density-inaccessible, then the adaptive distance is an exponential multiple of the Euclidean distance. Steps 1-5: Repeat steps 1-2 and 1-3 until no new density extrema appear, and obtain the density extrema set dep(p); Steps 1-6: For all non-density extrema x in the dataset X, calculate the distance L from point x to each density extrema dep(p). x→dep(p) ; Steps 1-7: Using the classification idea of ​​the DPC algorithm, classify point x into L. x→dep(p) The minimum density extremum point dep, and all non-density extremum points x classified as that point are clustered into one class, denoted as C. p The output clustering result Φ = {C1, C2, ... C} p } 3. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 2, characterized in that, Step 2, the data preprocessing stage, includes the following: Step 2-1: Clean the coordinate data and parcel quantity data by deleting irrelevant attributes. The main attributes to retain are location information, parcel saturation, and throughput. Also, delete null values ​​and fill in missing values, represented as D = (longitude, latitude, saturation, throughput, spare), where longitude and latitude are latitude and longitude information, saturation and throughput are saturation and throughput respectively, and spare is other optional attributes.

4. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 3, characterized in that, Step 3, obtaining the data matrix, includes the following steps: Step 3-1: Based on the coordinate data D processed in Step 2-1, which is randomly distributed throughout the city, use a third-party map platform to obtain an n×n data matrix of real road site distances, denoted as A=(a ij ) m×n , where a ij This represents the actual road network distance between station i and station j.

5. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 4, characterized in that, Step 4, obtaining the density extremum point set based on the data matrix, includes the following steps: Step 4-1: Since this problem involves actual distances, change the Euclidean distance of the Gaussian kernel function in Step 1-4 to the actual road network distance processed in Step 3-1. This makes the calculation of the correlation between logistics stations more realistic; Step 4-2: Randomly select x from the dataset i Perform steps 1-5 until no new density extrema appear, obtaining the density extrema set Ψ = {dep1, dep2, ..., dep}. p }, where p≤n.

6. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 2, characterized in that, Step 5 involves establishing a joint model of cluster density extreme points and maximum parcel quantity data, which includes the following steps: Step 5-1: Based on steps 1-7, obtain the clustering result Φ = {C1, C2, ... C}. p By determining the number of logistics stations m in the area, the number of density extreme points is adjusted to make p = m; Step 5-2: Based on the saturation of parcels at the logistics station and the adjusted m density extreme points, use the maximization algorithm to reclassify D = (longitude, latitude, saturation, throughput, spare) to obtain Φ = {C1, C2, ... C}. m }, making it classified as C m The sum of the parcel throughput of the number of stations is less than the parcel saturation of the logistics center, and the sum of the actual road network distances from all stations to the logistics center to which they belong is minimized. This is denoted as .

7. The method for logistics center site selection based on a hybrid clustering algorithm according to claim 1, characterized in that, Step 6 includes the following: Step 6-1: Combine the processed MS-DPC algorithm with the joint model, and through model training and testing, finally obtain the prediction result of the logistics center address; Step 6-2: Train the model multiple times, using AC, PR, and RE as accuracy evaluation metrics to predict the accuracy of the number of different logistics centers.

Images