A communication cell division method, apparatus and device

Abstract

The application provides a communication cell division method, device and equipment, wherein the communication cell division method comprises the following steps: obtaining a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set; obtaining a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; obtaining a target feature vector according to the normalized Laplacian matrix; and dividing the to-be-divided set according to the target feature vector; the to-be-divided set comprises at least two to-be-divided communication cells; and the reference matrix comprises an adjacency weight matrix or a distance matrix. The scheme can divide a large number of communication cells into multiple sub-areas (i.e. multiple cell subsets) at one time, simplifies the division process, improves the division efficiency, more reasonably extracts the feature correlation between the communication cells, improves the division accuracy and the stability of the division result, and solves the problems of low division efficiency and poor stability of the division result in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a communication cell division method, device and equipment. BACKGROUND

[0002] Many current methods based on graph neural networks have been successfully applied to communication cells; however, with the continuous growth of the number of communication cells, algorithms based on stochastic gradient descent (SGD) face two major problems: 1) the computational cost increases exponentially with the increase in the number of network layers; 2) maintaining the embedding of the entire graph and each node in memory requires a large amount of space. Based on this, it is crucial to divide a large number of communication cells in order to fully utilize the existing network efficiency and improve the memory and computational efficiency of the model. However, current schemes for dividing communication cells mainly include the following two types:

[0003] The first scheme is the traditional manual division method: the division is performed according to administrative regions, homogeneity, regional population, and other principles, which constrains the number, size, and functional properties of the communication cells.

[0004] The second scheme is a communication cell division method based on bisection spectral clustering: assuming that the undirected (connected) graph G of the communication cells can be divided into two parts, the non-zero eigenvectors of the Laplacian matrix L of the graph G can be used to generate two subsets of nodes, thereby achieving the bisection of the graph G. If it is desired to divide the graph into multiple subgraphs, the method needs to be iteratively used.

[0005] However, the above communication cell division schemes have the following defects:

[0006] 1. Some communication cells are difficult to directly divide under different rule constraints, requiring manual division by human experts, which is time-consuming and difficult to balance in terms of precision and workload.

[0007] 2. The bisection spectral clustering algorithm has low computational efficiency and unstable results.

[0008] As can be seen from the above, the existing communication cell division schemes have the problems of low division efficiency and poor stability of the division results. SUMMARY

[0009] The purpose of the present application is to provide a communication cell division method, device and equipment to solve the problem of low division efficiency and poor stability of the division results in the existing communication cell division schemes.

[0010] To solve the above technical problems, the present application provides a communication cell division method, which comprises:

[0011] obtaining a reference matrix of each of the communication cells according to distances between each two of the communication cells in the to-be-partitioned set;

[0012] obtaining a normalized Laplacian matrix corresponding to the to-be-partitioned set according to the reference matrix;

[0013] obtaining a target feature vector according to the normalized Laplacian matrix;

[0014] performing partition on the to-be-partitioned set according to the target feature vector;

[0015] The to-be-partitioned set includes at least two to-be-partitioned communication cells.

[0016] The reference matrix includes an adjacency weight matrix or a distance matrix.

[0017] Optionally, in a case where the reference matrix is an adjacency weight matrix, the obtaining of the reference matrix of each of the communication cells according to distances between each two of the communication cells in the to-be-partitioned set includes:

[0018] obtaining a weight value between the first communication cell and each other communication cell according to a distance between the first communication cell and each other communication cell;

[0019] obtaining an adjacency weight matrix of the first communication cell according to the weight value;

[0020] The first communication cell is any communication cell in the to-be-partitioned set, and the other communication cell is any communication cell in the to-be-partitioned set except the first communication cell.

[0021] Optionally, the obtaining of the reference matrix of each of the communication cells according to distances between each two of the communication cells in the to-be-partitioned set further includes:

[0022] obtaining an average traffic of the first communication cell in a first time period;

[0023] obtaining a traffic transfer amount between the first communication cell and each other communication cell according to the average traffic;

[0024] The obtaining of the adjacency weight matrix of the first communication cell according to the weight value includes:

[0025] obtaining the adjacency weight matrix of the first communication cell according to the weight value and the traffic transfer amount.

[0026] Optionally, the obtaining of the normalized Laplacian matrix corresponding to the to-be-partitioned set according to the reference matrix includes:

[0027] obtaining degrees corresponding to each communication cell;

[0028] obtaining a degree matrix corresponding to the to-be-partitioned set according to the degrees;

[0029] obtaining a normalized Laplacian matrix corresponding to the to-be-partitioned set according to the reference matrix and the degree matrix;

[0030] The degree represents a sum of weights of all edges adjacent to the communication cell.

[0031] Optionally, the obtaining of the target feature vector according to the normalized Laplacian matrix comprises:

[0032] performing feature decomposition on the normalized Laplacian matrix to obtain corresponding feature values;

[0033] obtaining a target feature vector according to the feature values.

[0034] Optionally, the obtaining of the target feature vector according to the feature values comprises:

[0035] obtaining at least two first non-zero feature values with smaller values from the feature values;

[0036] obtaining corresponding feature vectors from the Laplacian matrix as the target feature vector according to the at least two first non-zero feature values;

[0037] The maximum value of the at least two first non-zero feature values is smaller than the minimum value of other non-zero feature values.

[0038] The other non-zero feature values are non-zero feature values other than the first non-zero feature values in the feature values.

[0039] Optionally, the partitioning of the to-be-partitioned set according to the target feature vector comprises:

[0040] performing clustering on the target feature vector to obtain a clustering result;

[0041] partitioning the to-be-partitioned set according to the clustering result.

[0042] Optionally, the performing of clustering on the target feature vector to obtain a clustering result comprises:

[0043] constructing a feature matrix according to the target feature vector;

[0044] performing clustering on the feature matrix to obtain a clustering result.

[0045] Optionally, before the partitioning of the to-be-partitioned set according to the target feature vector, the method further comprises:

[0046] constructing a target undirected graph according to the to-be-partitioned set;

[0047] The partitioning the to-be-partitioned set according to the target feature vector comprises:

[0048] The target undirected graph is partitioned into at least two subgraphs that are not connected with each other according to the target feature vector;

[0049] At least two cell subsets corresponding to the at least two subgraphs are obtained;

[0050] The at least two cell subsets are not intersected with each other, and a union set of the at least two cell subsets is equal to the to-be-partitioned set.

[0051] Optionally, after the to-be-partitioned set is partitioned according to the target feature vector, the method further comprises:

[0052] In a case where a number of cells contained in a first cell subset obtained by the partitioning is less than a lower limit value, the first cell subset is combined with a target cell subset; the target cell subset is a cell subset in which a center communication cell contained in the cell subset is closest to a center communication cell of the first cell subset among cell subsets obtained by the partitioning;

[0053] In a case where a number of cells contained in a second cell subset obtained by the partitioning is greater than an upper limit value, the to-be-partitioned set is updated according to the second cell subset, and the method returns to the step of obtaining the reference matrix of each communication cell according to distances between each two communication cells in the to-be-partitioned set;

[0054] In a case where a number of cells contained in all cell subsets obtained by the partitioning satisfies a constraint condition, and a difference value between cell subsets corresponding to the to-be-partitioned set is less than a first threshold value, the number of target feature vectors obtained is updated, and the method returns to the step of obtaining the target feature vector according to the normalized Laplacian matrix;

[0055] The constraint condition comprises that the number of cells is greater than or equal to the lower limit value and less than or equal to the upper limit value.

[0056] Embodiments of the application further provide a communication cell partitioning device, which comprises:

[0057] A first obtaining module is configured to obtain a reference matrix of each communication cell according to distances between each two communication cells in a to-be-partitioned set;

[0058] A second obtaining module is configured to obtain a normalized Laplacian matrix corresponding to the to-be-partitioned set according to the reference matrix;

[0059] a third obtaining module, configured to obtain a target feature vector according to the normalized Laplacian matrix;

[0060] a first dividing module, configured to divide the to-be-divided set according to the target feature vector;

[0061] wherein the to-be-divided set comprises at least two to-be-divided communication cells;

[0062] the reference matrix comprises an adjacency weight matrix or a distance matrix.

[0063] Optionally, in the case where the reference matrix is an adjacency weight matrix, the reference matrix of each of the communication cells is obtained according to the distance between each two communication cells in the to-be-divided set, comprising:

[0064] a weight value between the first communication cell and each other communication cell is obtained according to the distance between the first communication cell and each other communication cell;

[0065] an adjacency weight matrix of the first communication cell is obtained according to the weight value;

[0066] wherein the first communication cell is any communication cell in the to-be-divided set, and the other communication cell is any communication cell in the to-be-divided set except the first communication cell.

[0067] Optionally, the reference matrix of each of the communication cells is obtained according to the distance between each two communication cells in the to-be-divided set, further comprising:

[0068] an average traffic of the first communication cell in a first time period is obtained;

[0069] a traffic transfer amount between the first communication cell and each other communication cell is obtained according to the average traffic;

[0070] wherein the adjacency weight matrix of the first communication cell is obtained according to the weight value, comprising:

[0071] the adjacency weight matrix of the first communication cell is obtained according to the weight value and the traffic transfer amount.

[0072] Optionally, the normalized Laplacian matrix corresponding to the to-be-divided set is obtained according to the reference matrix, comprising:

[0073] a degree corresponding to each communication cell is obtained;

[0074] a degree matrix corresponding to the to-be-divided set is obtained according to the degrees;

[0075] According to the reference matrix and the degree matrix, a normalized Laplacian matrix corresponding to the to-be-partitioned set is obtained;

[0076] The degree represents a sum of weights of all edges adjacent to the communication cell.

[0077] Optionally, the obtaining of the target feature vector according to the normalized Laplacian matrix comprises:

[0078] performing eigen decomposition on the normalized Laplacian matrix to obtain corresponding eigenvalues;

[0079] The target feature vector is obtained according to the eigenvalues.

[0080] Optionally, the obtaining of the target feature vector according to the eigenvalues comprises:

[0081] at least two first non-zero eigenvalues with smaller values are obtained from the eigenvalues;

[0082] According to the at least two first non-zero eigenvalues, corresponding feature vectors are obtained from the Laplacian matrix as the target feature vector.

[0083] The maximum value of the at least two first non-zero eigenvalues is smaller than the minimum value of other non-zero eigenvalues.

[0084] The other non-zero eigenvalues are non-zero eigenvalues other than the first non-zero eigenvalues in the eigenvalues.

[0085] Optionally, the partitioning of the to-be-partitioned set according to the target feature vector comprises:

[0086] performing clustering on the target feature vector to obtain a clustering result;

[0087] The to-be-partitioned set is partitioned according to the clustering result.

[0088] Optionally, the performing of clustering on the target feature vector to obtain a clustering result comprises:

[0089] a feature matrix is constructed according to the target feature vector;

[0090] the feature matrix is clustered to obtain a clustering result.

[0091] Optionally, the method further comprises:

[0092] a first construction module is configured to construct a target undirected graph according to the to-be-partitioned set before the to-be-partitioned set is partitioned according to the target feature vector;

[0093] The dividing the to-be-divided set according to the target feature vector comprises:

[0094] The target undirected graph is divided into at least two subgraphs without connection according to the target feature vector;

[0095] At least two cell subsets corresponding to the at least two subgraphs are obtained;

[0096] The at least two cell subsets are mutually exclusive, and the union of the at least two cell subsets is equal to the to-be-divided set.

[0097] Optionally, the method further comprises:

[0098] The first processing module is configured to, after the to-be-divided set is divided according to the target feature vector, in a case where the number of cells contained in a first cell subset obtained by the division is less than a lower limit value, merge the first cell subset with a target cell subset; the target cell subset is a cell subset with the smallest distance to the center communication cell of the first cell subset among the cell subsets obtained by the division;

[0099] After the to-be-divided set is divided according to the target feature vector, in a case where the number of cells contained in a second cell subset obtained by the division is greater than an upper limit value, the to-be-divided set is updated according to the second cell subset, and the step of obtaining the reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set is executed again;

[0100] After the to-be-divided set is divided according to the target feature vector, in a case where the number of cells contained in all cell subsets obtained by the division satisfies a constraint condition, and the difference value between the cell subsets corresponding to the to-be-divided set is less than a first threshold value, the number of target feature vectors obtained is updated, and the step of obtaining the target feature vector according to the normalized Laplacian matrix is executed again;

[0101] The constraint condition comprises: the number of cells is greater than or equal to the lower limit value and less than or equal to the upper limit value.

[0102] Embodiments of the application also provide a communication cell division device, comprising a processor and a transceiver;

[0103] The processor is configured to obtain a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set;

[0104] The reference matrix is used to obtain a normalized Laplacian matrix corresponding to the to-be-divided set;

[0105] Based on the normalized Laplacian matrix, the target feature vector is obtained;

[0106] The set to be divided is divided according to the target feature vector;

[0107] The set to be divided includes at least two communication cells to be divided;

[0108] The reference matrix includes either an adjacency weight matrix or a distance matrix.

[0109] Optionally, when the reference matrix is ​​an adjacency weight matrix, obtaining the reference matrix for each of the communication cells based on the distance between every two communication cells in the set to be divided includes:

[0110] Based on the distance between the first communication cell and each other communication cell, obtain the weight value between the first communication cell and each other communication cell;

[0111] Based on the weight values, the adjacency weight matrix of the first communication cell is obtained;

[0112] Wherein, the first communication cell is any communication cell in the set to be divided, and the other communication cells are any communication cells in the set to be divided other than the first communication cell.

[0113] Optionally, obtaining the reference matrix for each communication cell based on the distance between every two communication cells in the set to be divided further includes:

[0114] Obtain the average traffic of the first communication cell during the first time period;

[0115] Based on the average traffic, the traffic transfer amount between the first communication cell and each other communication cell is obtained;

[0116] The step of obtaining the adjacency weight matrix of the first communication cell based on the weight values ​​includes:

[0117] Based on the weight values ​​and traffic transfer volume, the adjacency weight matrix of the first communication cell is obtained.

[0118] Optionally, obtaining the normalized Laplace matrix corresponding to the set to be partitioned based on the reference matrix includes:

[0119] Obtain the degree corresponding to each communication cell;

[0120] Based on the stated degree, the degree matrix corresponding to the set to be partitioned is obtained;

[0121] Based on the reference matrix and the degree matrix, the normalized Laplace matrix corresponding to the set to be partitioned is obtained;

[0122] wherein the degree represents the sum of weights of all edges adjacent to the communication cell.

[0123] Optionally, the obtaining the target feature vector according to the normalized Laplacian matrix comprises:

[0124] performing eigen decomposition on the normalized Laplacian matrix to obtain corresponding eigenvalues;

[0125] obtaining the target feature vector according to the eigenvalues.

[0126] Optionally, the obtaining the target feature vector according to the eigenvalues comprises:

[0127] obtaining at least two first non-zero eigenvalues with smaller values from the eigenvalues;

[0128] obtaining corresponding feature vectors from the Laplacian matrix according to the at least two first non-zero eigenvalues as the target feature vector;

[0129] wherein a maximum value in the at least two first non-zero eigenvalues is smaller than a minimum value in other non-zero eigenvalues;

[0130] the other non-zero eigenvalues are non-zero eigenvalues in the eigenvalues except the first non-zero eigenvalues.

[0131] Optionally, the dividing the to-be-divided set according to the target feature vector comprises:

[0132] performing clustering on the target feature vector to obtain a clustering result;

[0133] dividing the to-be-divided set according to the clustering result.

[0134] Optionally, the performing clustering on the target feature vector to obtain a clustering result comprises:

[0135] constructing a feature matrix according to the target feature vector;

[0136] performing clustering on the feature matrix to obtain a clustering result.

[0137] Optionally, the processor is further configured to:

[0138] before the dividing the to-be-divided set according to the target feature vector, constructing a target undirected graph according to the to-be-divided set;

[0139] the dividing the to-be-divided set according to the target feature vector comprises:

[0140] partition the target undirected graph into at least two subgraphs without connection with each other according to the target feature vector;

[0141] obtain at least two cell subsets corresponding to the at least two subgraphs according to the at least two subgraphs;

[0142] wherein each of the cell subsets has no intersection with each other, and the union of the at least two cell subsets is equal to the to-be-partitioned set.

[0143] Optionally, the processor is further configured to:

[0144] after partitioning the to-be-partitioned set according to the target feature vector, in a case where the number of cells contained in a first cell subset obtained by the partitioning is less than a lower limit value, merging the first cell subset with a target cell subset; the target cell subset is a cell subset having a minimum distance to the center communication cell of the first cell subset among the cell subsets obtained by the partitioning;

[0145] after partitioning the to-be-partitioned set according to the target feature vector, in a case where the number of cells contained in a second cell subset obtained by the partitioning is greater than an upper limit value, updating the to-be-partitioned set according to the second cell subset, and returning to execute the obtaining of the reference matrix of each communication cell according to the distance between each two communication cells in the to-be-partitioned set;

[0146] after partitioning the to-be-partitioned set according to the target feature vector, in a case where the number of cells contained in all cell subsets obtained by the partitioning satisfies a constraint condition, and the difference value between the cell subsets corresponding to the to-be-partitioned set is less than a first threshold value, updating the number of obtaining of the target feature vector, and returning to execute the obtaining of the target feature vector according to the normalized Laplacian matrix;

[0147] wherein the constraint condition comprises: the number of cells is greater than or equal to the lower limit value, and less than or equal to the upper limit value.

[0148] The embodiment of the present application also provides a communication cell partitioning device, comprising a memory, a processor and a program stored in the memory and executable on the processor; the processor implements the communication cell partitioning method when executing the program.

[0149] The embodiment of the present application also provides a readable storage medium, which stores a program, and the program is executable on a processor to implement the steps in the communication cell partitioning method.

[0150] The beneficial effects of the above technical solutions of the present application are as follows:

[0151] In the scheme, the communication cell division method obtains a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set; obtains a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; obtains a target feature vector according to the normalized Laplacian matrix; and divides the to-be-divided set according to the target feature vector; wherein the to-be-divided set comprises at least two to-be-divided communication cells; the reference matrix comprises an adjacency weight matrix or a distance matrix; a large number of communication cells can be divided into a plurality of sub-regions (i.e. a plurality of cell subsets) at one time, the division process is simplified, the division efficiency is improved, the feature correlation between the communication cells is more reasonably extracted, the division accuracy and the stability of the division result are improved, and the problems of low division efficiency and poor stability of the division result of the existing communication cell division scheme are well solved. BRIEF DESCRIPTION OF DRAWINGS

[0152] Figure 1 A communication cell division method flowchart of an embodiment of the present application;

[0153] Figure 2 A communication cell division method specific implementation flowchart of an embodiment of the present application;

[0154] Figure 3 A communication cell division device structure diagram of an embodiment of the present application;

[0155] Figure 4 A communication cell division device structure diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0156] In order to make the technical problems, technical solutions and advantages of the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0157] The present application provides a communication cell division method to solve the problems of low division efficiency and poor stability of the division result of the existing communication cell division scheme, as shown in Figure 1 The method comprises the following steps:

[0158] Step 11: obtaining a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set;

[0159] Step 12: obtaining a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix;

[0160] Step 13: obtaining a target feature vector according to the normalized Laplacian matrix;

[0161] Step 14: dividing the to-be-divided set according to the target feature vector; wherein the to-be-divided set comprises at least two to-be-divided communication cells; and the reference matrix comprises an adjacency weight matrix or a distance matrix.

[0162] The "dividing the to-be-divided set" can obtain at least two cell subsets; and the "distance matrix" can be a matrix composed of the above-mentioned distances.

[0163] The communication cell division method provided by the embodiment of the present application can obtain the reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set; obtain the normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; obtain the target feature vector according to the normalized Laplacian matrix; and divide the to-be-divided set according to the target feature vector; wherein the to-be-divided set comprises at least two to-be-divided communication cells; and the reference matrix comprises an adjacency weight matrix or a distance matrix. The method can divide a large number of communication cells into multiple subsets (i.e. multiple cell subsets) at one time, simplifies the division process, improves the division efficiency, more reasonably extracts the feature correlation between the communication cells, improves the division precision and the stability of the division result, and solves the problems of low division efficiency and poor stability of the division result in the prior art.

[0164] In the embodiment of the present application, when the reference matrix is an adjacency weight matrix, the reference matrix of each communication cell is obtained according to the distance between each two communication cells in the to-be-divided set, comprising: obtaining the weight value between the first communication cell and each other communication cell according to the distance between the first communication cell and each other communication cell; and obtaining the adjacency weight matrix of the first communication cell according to the weight value; wherein the first communication cell is any communication cell in the to-be-divided set, and the other communication cell is any communication cell in the to-be-divided set except the first communication cell.

[0165] In this way, the adjacency weight matrix can be accurately obtained. The number of other communication cells is at least one, and the first communication cell and the other communication cells together constitute the to-be-divided set (which can also be understood as the union equaling the to-be-divided set).

[0166] Further, the reference matrix of each communication cell is obtained according to the distance between each two communication cells in the to-be-divided set, and further comprising: obtaining the average traffic of the first communication cell in a first time period; obtaining the traffic transfer amount between the first communication cell and each other communication cell according to the average traffic; wherein the adjacent weight matrix of the first communication cell is obtained according to the weight value and the traffic transfer amount.

[0167] In this way, the spatial dynamic correlation between the communication cells can be better captured.

[0168] In the embodiment of the application, the reference matrix is obtained according to the degree of each communication cell, and the degree matrix corresponding to the to-be-divided set is obtained according to the degree; the normalized Laplacian matrix corresponding to the to-be-divided set is obtained according to the reference matrix and the degree matrix; wherein the degree represents the sum of the weights of all edges adjacent to the communication cell.

[0169] In this way, the normalized Laplacian matrix can be accurately obtained. The "edge" in the normalized Laplacian matrix can represent the connection relationship between the communication cells.

[0170] In the embodiment of the application, the target feature vector is obtained according to the normalized Laplacian matrix, comprising: performing eigenvalue decomposition on the normalized Laplacian matrix to obtain a corresponding eigenvalue; and obtaining the target feature vector according to the eigenvalue.

[0171] In this way, the target feature vector can be accurately obtained.

[0172] The target feature vector is obtained according to the eigenvalue, comprising: obtaining at least two first non-zero eigenvalues with smaller values from the eigenvalues; and obtaining a corresponding feature vector from the Laplacian matrix as the target feature vector according to the at least two first non-zero eigenvalues; wherein the maximum value of the at least two first non-zero eigenvalues is smaller than the minimum value of other non-zero eigenvalues; the other non-zero eigenvalues are non-zero eigenvalues in the eigenvalues except the first non-zero eigenvalues.

[0173] In this way, the target feature vector can be obtained.

[0174] In the embodiment of the application, the to-be-divided set is divided according to the target feature vector, comprising: clustering the target feature vector to obtain a clustering result; and dividing the to-be-divided set according to the clustering result.

[0175] In this way, the to-be-divided set can be better divided.

[0176] The clustering of the target feature vectors comprises: constructing a feature matrix according to the target feature vectors; and clustering the feature matrix to obtain the clustering result.

[0177] In this way, the clustering result can be accurately obtained.

[0178] Further, before the dividing of the to-be-divided set according to the target feature vectors, the method further comprises: constructing a target undirected graph according to the to-be-divided set; and the dividing of the to-be-divided set according to the target feature vectors comprises: dividing the target undirected graph into at least two subgraphs that are not connected to each other according to the target feature vectors; and obtaining at least two cell subsets corresponding to the at least two subgraphs according to the at least two subgraphs; wherein each of the cell subsets is disjoint with each other, and a union set of the at least two cell subsets is equal to the to-be-divided set.

[0179] In this way, the dividing result can be made to be as optimal as possible, that is, the difference between communication cells in a cell subset is minimized, and the difference between communication cells in different cell subsets is maximized. Specifically, one cell subset can be obtained according to one subgraph.

[0180] Further, after the dividing of the to-be-divided set according to the target feature vectors, the method further comprises: in a case where a number of cells contained in a first cell subset obtained by the dividing is less than a lower limit value, merging the first cell subset with a target cell subset; the target cell subset is a cell subset in which a central communication cell contained in a cell subset obtained by the dividing is closest to a central communication cell of the first cell subset; in a case where a number of cells contained in a second cell subset obtained by the dividing is greater than an upper limit value, updating the to-be-divided set according to the second cell subset, and returning to execute the obtaining of the reference matrix of each of the communication cells according to the distance between each two communication cells in the to-be-divided set; in a case where a number of cells contained in all cell subsets obtained by the dividing satisfies a constraint condition, and a difference value between cell subsets corresponding to the to-be-divided set is less than a first threshold value, updating a number of target feature vectors obtained, and returning to execute the obtaining of the target feature vectors according to the normalized Laplacian matrix; wherein the constraint condition comprises that the number of cells is greater than or equal to the lower limit value and less than or equal to the upper limit value.

[0181] In this way, a dividing result with high stability can be finally obtained. After the merging of the first cell subset and the target cell subset, it can be further judged whether a number of cells contained in the target cell subset satisfies the constraint condition; and similar operations are performed according to whether the constraint condition is satisfied, that is, the merged target cell subset is compared with the first cell subset or the second cell subset.

[0182] In addition, the "updating the to-be-divided set according to the second subset of cells" can be understood as taking the subset of cells as the updated to-be-divided set; the "returning to performing the obtaining of the reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set" can be understood as a loop; that is, the subset of cells is divided by dividing the to-be-divided set.

[0183] The "all the subset of cells obtained by the division contains a number of cells satisfying the constraint condition" can also be adjusted to "at least a first proportion of the subset of cells obtained by the division contains a number of cells satisfying the constraint condition", which is not limited herein; the "difference value between the subset of cells" can be specifically equal to the variance B between the subset of cells divided by the variance A in the subset of cells (that is, the difference value is calculated by evaluating the inter-class variance and the intra-class variance, and the ideal classification condition is that the inter-class variance is maximum and the intra-class variance is minimum); but it is not limited thereto. The "updating the number of obtained target feature vectors" can be specifically increasing the number of obtained target feature vectors, such as adding 1; or it can be understood that the number of first non-zero feature values is updated. The "returning to performing the obtaining of the target feature vector according to the normalized Laplacian matrix" can be understood as: re-executing (and based thereon, re-executing the operation after the operation) the operation, which is a loop. Specifically, the "returning to performing the obtaining of the target feature vector according to the normalized Laplacian matrix" can include: obtaining the target feature vector according to the updated number of obtained target feature vectors and the normalized Laplacian matrix (and based thereon, re-executing the operation after the operation of obtaining the target feature vector).

[0184] In addition, the operation performed under the condition that the difference value between the subset of cells is less than the first threshold value can further ensure that the division result reaches the minimum difference of communication cells in the sub-area and the maximum difference of communication cells between the sub-areas; and solves the problems of the original scheme, such as low traffic similarity, multiple redundant nodes in the sub-area (corresponding to the subset of cells), incomplete feature extraction, and the like.

[0185] In the embodiment of the application, the final division result (that is, the final subset of cells obtained by the division) can also be output under the condition that the number of cells contained in the at least first proportion of the subset of cells satisfies the constraint condition, and the difference value between the subset of cells corresponding to the to-be-divided set is greater than or equal to the first threshold value.

[0186] The "first proportion" is the ratio of the number of the subset of cells containing a number of cells satisfying the constraint condition to the number of all the subset of cells. Specifically, the value of the first proportion can be greater than or equal to 0.5 and less than or equal to 1.

[0187] The communication cell division method provided by the embodiment of the present application is exemplified below, and a reference matrix is taken as an example of an adjacency weight matrix.

[0188] To solve the above technical problems, the embodiment of the present application provides a communication cell division method, which can be specifically implemented as a massive communication cell division method based on spectral clustering and a (traffic) migration model, involving: constructing an adjacency weight matrix through the distance between communication cell nodes (i.e., communication cells) and a traffic migration model; establishing a normalized Laplacian matrix; solving the eigenvalues and eigenvectors of the Laplacian matrix; selecting the first k non-zero eigenvectors (corresponding to the above target eigenvectors) and using a k-means clustering algorithm to cluster into K classes (corresponding to the above cell subsets); thereby accurately and efficiently dividing massive communication cells under the condition of known relevant communication cell data, while solving the problems of low traffic similarity in the prior art, multiple redundant nodes in the subsets (corresponding to the above cell subsets), and incomplete feature extraction.

[0189] Specifically, the scheme provided by the embodiment of the present application can be as shown in Figure 2 and includes the following operations:

[0190] 1. Set a subset division node number threshold (including the above lower limit value and upper limit value); input the topology structure and traffic data of the communication cells (corresponding to the communication cells in the above to-be-divided set);

[0191] Subsequently, the node positions of the communication cells can be obtained according to the topology structure, and then the distances between the communication cells can be obtained; as for the connection relationship between the communication cells, it can be obtained according to the topology structure or according to the above distance, which is not limited here; in addition, the average traffic of the communication cells in a T time period (corresponding to the above first time period) can be obtained according to the traffic data. The T time period can be an empirical value, which is not limited here.

[0192] 2. Communication cell graph (corresponding to the above target undirected graph) construction.

[0193] Specifically, it is assumed that the distance between the communication cell i (corresponding to the above first communication cell) and the communication cell j (corresponding to the above other communication cell) is d ij All to-be-divided communication cells (corresponding to the above to-be-divided set) can be defined as an undirected graph (corresponding to the above target undirected graph) G=(V,E), which is in the form of a matrix; V is a finite node set, and |V|=N, N is the number of nodes (i.e., the total number of communication cells) in the undirected graph, and a node represents a communication cell; E is a set of edges, and an "edge" represents the connection relationship between nodes. Subsequently, the communication cells without connection relationship can be divided according to the connection relationship. Specifically, d ijThe Euclidean distance calculated by latitude and longitude, but not limited thereto.

[0194] 3. Calculate the adjacency weight matrix between communication cell nodes, including: calculating the weight (corresponding to the weight value described above) and traffic transfer amount between communication cells. Specifically, it can include:

[0195] (1) Specifically, the weight value w between communication cell i and communication cell j (also known as adjacent nodes) in the undirected graph can be calculated by using the Gaussian kernel function ij , as follows:

[0196]

[0197] That is, in the case of i≠j and , w ij =0.

[0198] The exp in the formula represents the exponential function; where σ and ∈ (greater than 0 and less than 1) are threshold values (specifically, empirical values, and σ is not equal to 0, ∈ is greater than 0 and less than 1), which can be used to control the sparsity of the adjacency weight matrix.

[0199] (2) At the same time, in order to better capture the spatial dynamic correlation between nodes (i.e., communication cells), the potential traffic migration value (i.e., traffic transfer amount) between communication cells can be specifically calculated for use, where the potential traffic migration value can be proportional to the current traffic between two communication cells (i.e., communication cell i and communication cell j), and inversely proportional to the distance between the current two communication cells, but not limited thereto; Specifically, it is assumed that: the average traffic of each (communication) cell in the T period in the undirected graph G is known (for example, the average traffic of communication cell i in the T period is , and the average traffic of communication cell j in the T period is ), the potential traffic migration value can be calculated as follows:

[0200]

[0201] That is, in the case of w ij ≠0, otherwise

[0202] (3) Finally, for each communication cell to be partitioned, the following formula is used to construct its adjacency weight matrix (subsequently, the weight matrix A corresponding to the undirected graph can be constructed according to ):

[0203] where the scalar ρ is used to adjust The proportion in the whole adjacent weight matrix (specifically, p can be an empirical value, greater than 0 and less than 1).

[0204] 4. A Laplace matrix corresponding to the cell is established (corresponding to the normalized Laplace matrix corresponding to the above-mentioned set to be divided), and eigenvalues of the Laplace matrix are decomposed, k smallest non-zero eigenvalues (corresponding to the first non-zero eigenvalue) and corresponding eigenvectors (corresponding to the target eigenvector) are obtained. Specifically, it includes:

[0205] (1) Calculate the normalized Laplace matrix. After obtaining the adjacent weight matrix, the degree d of the node i (i.e. communication cell i) can be defined first i : the sum of the weights of all edges adjacent to the node i, that is where n is the number of neighbors (i.e. adjacent nodes) of node i (also understood as the number of communication cells j). d i is constructed as the diagonal element of the degree matrix D as follows:

[0206]

[0207] where N is the number of nodes in the undirected graph, and the Laplace matrix is defined as L=D-A in the present scheme, and the Laplace matrix is normalized , thereby obtaining the normalized Laplace matrix L sym . A in which is the weight matrix corresponding to the undirected graph constructed according to .

[0208] (2) Solve the eigenvalues and eigenvectors of the normalized Laplace matrix. After obtaining the normalized Laplace matrix, the eigenvalues x of the Laplace matrix are solved in the present scheme {x1, x2,..., x N}, the eigenvectors λ1, λ2,..., λ k corresponding to the first k non-zero eigenvalues (corresponding to the first non-zero eigenvalue) of the smallest eigenvalues of the Laplace matrix are selected, and a new characteristic matrix P={λ1, λ2,..., λ k} is constructed. Wherein k is greater than or equal to 2 and less than or equal to N.

[0209] 5. The K-means clustering algorithm is used to cluster the eigenvectors into K classes (corresponding to the clustering result). Specifically, it includes:

[0210] After obtaining the new feature matrix P, the scheme can use the K-means clustering method to cluster P, and then cut the undirected graph G (consisting of the communication cells to be divided) into K subgraphs (corresponding to the at least two subgraphs mentioned above, where K can be self-defined) that have no connection with each other according to the clustering results. The node set of the K subgraphs (corresponding to the at least two cell subsets mentioned above) is A1, A2,..., A k , and satisfies (that is, there is no intersection between each of the cell subsets), and A1∪A2∪...∪A k = V (corresponding to the union of the at least two cell subsets being equal to the set to be divided).

[0211] Further, after obtaining the classification of all communication cells to be divided (corresponding to the subgraphs), each type of cell can be renumbered according to the classification.

[0212] 6. Spectral clustering test: after dividing the undirected graph G, it is determined whether the number of nodes in each cluster (that is, each subgraph) and the difference between clusters meet the threshold value (that is, it is determined whether the number of nodes in each cell subset meets the above constraint condition; and whether the difference between cell subsets is greater than or equal to the first threshold value). If not, continue to divide; if yes, output the final divided sub-area node data (corresponding to the final division result). Specifically, it includes:

[0213] After obtaining K cluster clustering, the scheme can first determine whether the number of communication cells in the divided sub-area meets the constraint condition M min ≤|A i |≤M max , where |A i | is the total number of all communication cells in the current sub-area (corresponding to the number of cells mentioned above), M min , M max are the minimum and maximum number of communication cells allowed in the sub-area (corresponding to the lower limit value and the upper limit value mentioned above, respectively).

[0214] If the number of nodes in the sub-area is lower than M min , it is merged into the nearest sub-area (corresponding to the merging of the first cell subset and the target cell subset; in addition, after merging, it can be determined again whether the number of nodes in the merged sub-area meets the constraint condition);

[0215] If the number of nodes in the sub-area is higher than M maxThen, for this sub-region (corresponding to the second cell subset mentioned above), the communication cell map construction is re-implemented (that is, the cell set of this sub-region is used as a new set to be divided, and the above operation is performed; corresponding to the above, the set to be divided is updated according to the second cell subset, and the process of obtaining the reference matrix of each communication cell based on the distance between every two communication cells in the set to be divided is returned).

[0216] Once the number of cells within all sub-regions meets the constraints, the `calinski_harabaz_score` interface in sklearn can be used to calculate the variance A within each sub-region and the variance B between sub-regions, obtaining the significance of differences between each sub-region (specifically, the comprehensive difference value B / A, corresponding to the difference value between the aforementioned cell subsets). If the difference is low (corresponding to the difference value between the aforementioned cell subsets being less than the first threshold), then the process of "solving the eigenvalues ​​and eigenvectors of the normalized Laplacian matrix" is returned, and the value of k is increased, for example, by 1 (which can also be understood as increasing the number of eigenvectors obtained, corresponding to the number of target eigenvectors obtained in the above update, and returning to execute the step of obtaining the target eigenvector based on the normalized Laplacian matrix). Here, `calinski_harabaz_score` calculates the result for all the cell subsets obtained from the division, and this result is used as the basis for judging the difference.

[0217] If the number of nodes in each cluster and the difference between clusters meet the threshold, the final partitioning result is output (i.e., the final partitioning of sub-region node data, which includes: node data that identifies relevant information about the sub-region, that is, establishing a connection between the sub-region and the communication cell, such as labeling the communication cell according to the sub-region).

[0218] Here are some examples of application scenarios for this solution:

[0219] When performing operations such as prediction and fault detection on a massive number of cells in a communication network, the system is difficult to support simultaneous operation on tens of thousands of cells. In this case, the massive number of cells can be divided into sub-cells according to this solution. After the division, the traffic similarity and correlation within the cells in the sub-cells are high. When performing subsequent operations such as prediction and fault detection, the information loss caused by cell division can be minimized, thus avoiding poor accuracy of the prediction results.

[0220] As shown above, this solution involves a multi-path clustering method and a gravity-based traffic migration model, which can divide multiple sub-regions at once, improve the division efficiency, and extract potential feature associations between communication cells more reasonably, thereby improving the division accuracy.

[0221] In summary, compared with existing technologies, this solution has the following main advantages:

[0222] (1) can divide multiple areas at one time, simplify the division process, and the division result has high stability.

[0223] (2) only the distance and traffic data between communication cells are needed, other prior knowledge is not needed, and the scheme is simple to implement.

[0224] (3) the division result can achieve the minimum difference of communication cells in a sub-area and the maximum difference of communication cells between sub-areas.

[0225] The embodiment of the application also provides a communication cell division device, as shown in the accompanying drawings, comprising: Figure 3

[0226] The first acquisition module 31 is used for acquiring a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set;

[0227] The second acquisition module 32 is used for acquiring a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix;

[0228] The third acquisition module 33 is used for acquiring a target feature vector according to the normalized Laplacian matrix;

[0229] The first division module 34 is used for dividing the to-be-divided set according to the target feature vector;

[0230] The to-be-divided set comprises at least two to-be-divided communication cells.

[0231] The reference matrix comprises an adjacency weight matrix or a distance matrix.

[0232] The communication cell division device provided by the embodiment of the application can acquire a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set, acquire a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix, acquire a target feature vector according to the normalized Laplacian matrix, and divide the to-be-divided set according to the target feature vector, wherein the to-be-divided set comprises at least two to-be-divided communication cells, and the reference matrix comprises an adjacency weight matrix or a distance matrix. A large number of communication cells can be divided into multiple sub-areas (i.e. multiple cell subsets) at one time, the division process is simplified, the division efficiency is improved, the feature correlation between communication cells is more reasonably extracted, the division precision and the stability of the division result are improved, and the problems of low division efficiency and poor stability of the division result in the prior art are well solved.

[0233] ​In the case that the reference matrix is an adjacency weight matrix, the reference matrix of each of the communication cells is obtained according to distances between each two of the communication cells in the to-be-divided set, comprising: weight values between the first communication cell and each other communication cell are obtained according to distances between the first communication cell and each other communication cell; and the adjacency weight matrix of the first communication cell is obtained according to the weight values; wherein the first communication cell is any communication cell in the to-be-divided set, and the other communication cell is any communication cell in the to-be-divided set except the first communication cell.

[0234] Further, the reference matrix of each of the communication cells is obtained according to distances between each two of the communication cells in the to-be-divided set, further comprising: average traffic of the first communication cell in a first time period is obtained; and traffic transfer amounts between the first communication cell and each other communication cell are obtained according to the average traffic; wherein the adjacency weight matrix of the first communication cell is obtained according to the weight values, comprising: the adjacency weight matrix of the first communication cell is obtained according to the weight values and the traffic transfer amounts.

[0235] In the embodiment of the application, the normalized Laplacian matrix corresponding to the to-be-divided set is obtained according to the reference matrix, comprising: degrees corresponding to each of the communication cells are obtained; a degree matrix corresponding to the to-be-divided set is obtained according to the degrees; and the normalized Laplacian matrix corresponding to the to-be-divided set is obtained according to the reference matrix and the degree matrix; wherein the degree represents a sum of weights of all edges adjacent to the communication cell.

[0236] The target feature vector is obtained according to the normalized Laplacian matrix, comprising: the normalized Laplacian matrix is subjected to eigenvalue decomposition to obtain corresponding eigenvalues; and the target feature vector is obtained according to the eigenvalues.

[0237] Specifically, the target feature vector is obtained according to the eigenvalues, comprising: at least two first non-zero eigenvalues with small values are obtained from the eigenvalues; and the corresponding feature vector from the Laplacian matrix is obtained as the target feature vector according to the at least two first non-zero eigenvalues; wherein a maximum value of the at least two first non-zero eigenvalues is smaller than a minimum value of other non-zero eigenvalues; the other non-zero eigenvalues are non-zero eigenvalues in the eigenvalues except the first non-zero eigenvalues.

[0238] In the embodiment of the application, the to-be-divided set is divided according to the target feature vector, comprising: clustering is performed on the target feature vector to obtain a clustering result; and the to-be-divided set is divided according to the clustering result.

[0239] The clustering of the target feature vector comprises: constructing a feature matrix according to the target feature vector; and clustering the feature matrix to obtain the clustering result.

[0240] Further, the communication cell partitioning apparatus further comprises a first constructing module configured to, before partitioning the to-be-partitioned set according to the target feature vector, construct a target undirected graph according to the to-be-partitioned set; and the partitioning the to-be-partitioned set according to the target feature vector comprises: partitioning the target undirected graph into at least two subgraphs that are not connected with each other according to the target feature vector; and obtaining at least two cell subsets corresponding to the at least two subgraphs according to the at least two subgraphs; wherein each of the cell subsets is disjoint with each other, and a union set of the at least two cell subsets is equal to the to-be-partitioned set.

[0241] Further, the communication cell partitioning apparatus further comprises a first processing module configured to, after partitioning the to-be-partitioned set according to the target feature vector, in a case that a number of cells contained in a first cell subset obtained by the partitioning is less than a lower limit value, merging the first cell subset with a target cell subset; the target cell subset is a cell subset in which a center communication cell contained in a cell subset obtained by the partitioning has a minimum distance with a center communication cell of the first cell subset; in a case that a number of cells contained in a second cell subset obtained by the partitioning is greater than an upper limit value, updating the to-be-partitioned set according to the second cell subset, and returning to execute the obtaining of the reference matrix of each of the communication cells according to the distance between each two communication cells in the to-be-partitioned set; in a case that a number of cells contained in all cell subsets obtained by the partitioning satisfies a constraint condition, and a difference value between cell subsets corresponding to the to-be-partitioned set is less than a first threshold value, updating a number of target feature vectors obtained, and returning to execute the obtaining of the target feature vector according to the normalized Laplacian matrix; wherein the constraint condition comprises that the number of cells is greater than or equal to the lower limit value, and less than or equal to the upper limit value.

[0242] The implementation embodiments of the communication cell partitioning method are applicable to the embodiments of the communication cell partitioning apparatus, and can achieve the same technical effects.

[0243] The embodiment of the application further provides a communication cell partitioning device, as shown in the accompanying drawings, comprising a processor 41 and a transceiver 42. Figure 4

[0244] ​The processor 41 is configured to obtain a reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set;

[0245] The reference matrix is used to obtain a normalized Laplacian matrix corresponding to the to-be-divided set;

[0246] The normalized Laplacian matrix is used to obtain a target feature vector;

[0247] The target feature vector is used to divide the to-be-divided set;

[0248] The to-be-divided set includes at least two to-be-divided communication cells.

[0249] The reference matrix includes an adjacency weight matrix or a distance matrix.

[0250] The transceiver 42 is capable of communicating with the processor 41.

[0251] The communication cell division device provided by the embodiment of the application can obtain a reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set, obtain a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix, obtain a target feature vector according to the normalized Laplacian matrix, and divide the to-be-divided set according to the target feature vector. The to-be-divided set includes at least two to-be-divided communication cells. The reference matrix includes an adjacency weight matrix or a distance matrix. The device can divide a large number of communication cells into multiple subareas (i.e., multiple cell subsets) at one time, simplifies the division process, improves the division efficiency, more reasonably extracts the feature correlation between the communication cells, improves the division precision and the stability of the division result, and solves the problems of low division efficiency and poor stability of the division result in the prior art.

[0252] When the reference matrix is an adjacency weight matrix, the reference matrix of each communication cell is obtained according to the distance between each two communication cells in the to-be-divided set, including obtaining a weight value between a first communication cell and each other communication cell according to the distance between the first communication cell and each other communication cell, and obtaining an adjacency weight matrix of the first communication cell according to the weight value. The first communication cell is any communication cell in the to-be-divided set, and the other communication cell is any communication cell in the to-be-divided set except the first communication cell.

[0253] Further, the reference matrix of each communication cell is obtained according to the distance between each two communication cells in the to-be-divided set, and further includes: obtaining the average traffic of the first communication cell in a first time period; obtaining the traffic transfer amount between the first communication cell and each other communication cell according to the average traffic; and wherein the adjacent weight matrix of the first communication cell is obtained according to the weight value, including: obtaining the adjacent weight matrix of the first communication cell according to the weight value and the traffic transfer amount.

[0254] In the embodiment of the application, the normalized Laplacian matrix corresponding to the to-be-divided set is obtained according to the reference matrix, including: obtaining the degree corresponding to each communication cell; obtaining the degree matrix corresponding to the to-be-divided set according to the degree; and obtaining the normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix and the degree matrix; wherein the degree represents the sum of the weights of all edges adjacent to the communication cell.

[0255] The target feature vector is obtained according to the normalized Laplacian matrix, including: performing feature decomposition on the normalized Laplacian matrix to obtain corresponding eigenvalues; and obtaining the target feature vector according to the eigenvalues.

[0256] Specifically, the target feature vector is obtained according to the eigenvalues, including: obtaining at least two first non-zero eigenvalues with smaller values from the eigenvalues; and obtaining the corresponding feature vector from the Laplacian matrix as the target feature vector according to the at least two first non-zero eigenvalues; wherein the maximum value of the at least two first non-zero eigenvalues is smaller than the minimum value of other non-zero eigenvalues; and the other non-zero eigenvalues are non-zero eigenvalues in the eigenvalues except the first non-zero eigenvalues.

[0257] In the embodiment of the application, the to-be-divided set is divided according to the target feature vector, including: clustering the target feature vector to obtain a clustering result; and dividing the to-be-divided set according to the clustering result.

[0258] The target feature vector is clustered to obtain a clustering result, including: constructing a feature matrix according to the target feature vector; and clustering the feature matrix to obtain the clustering result.

[0259] Further, the processor is further configured to: before performing the dividing on the to-be-divided set according to the target feature vector, construct a target undirected graph according to the to-be-divided set; and the performing the dividing on the to-be-divided set according to the target feature vector comprises: dividing the target undirected graph into at least two subgraphs that are not connected with each other according to the target feature vector; and obtaining at least two cell subsets corresponding to the at least two subgraphs according to the at least two subgraphs; wherein each of the cell subsets is disjoint with each other, and a union set of the at least two cell subsets is equal to the to-be-divided set.

[0260] Further, the processor is further configured to: in a case where a number of cells included in a first cell subset obtained after the dividing on the to-be-divided set according to the target feature vector is less than a lower limit value, merging the first cell subset with a target cell subset; the target cell subset is a cell subset in which a center communication cell included in a cell subset obtained after the dividing is closest to a center communication cell of the first cell subset; in a case where a number of cells included in a second cell subset obtained after the dividing on the to-be-divided set according to the target feature vector is greater than an upper limit value, updating the to-be-divided set according to the second cell subset, and returning to perform the obtaining the reference matrix of each of the communication cells according to the distance between each two communication cells in the to-be-divided set; and in a case where a number of cells included in all cell subsets obtained after the dividing on the to-be-divided set according to the target feature vector satisfies a constraint condition, and a difference value between cell subsets corresponding to the to-be-divided set is less than a first threshold value, updating a number of target feature vectors obtained, and returning to perform the obtaining the target feature vector according to the normalized Laplacian matrix; wherein the constraint condition comprises: the number of cells is greater than or equal to the lower limit value, and less than or equal to the upper limit value.

[0261] The implementation embodiments of the communication cell dividing method described above are applicable to the embodiments of the communication cell dividing device, and can achieve the same technical effects.

[0262] The embodiment of the application further provides a communication cell dividing device, which comprises a memory, a processor and a program stored in the memory and executable on the processor; and the processor implements the communication cell dividing method described above when executing the program.

[0263] The implementation embodiments of the communication cell dividing method described above are applicable to the embodiments of the communication cell dividing device, and can achieve the same technical effects.

[0264] The embodiment of the application further provides a readable storage medium, which stores a program, and the program is executable on a processor to implement the steps in the communication cell dividing method described above.

[0265] The implementation embodiments of the communication cell partitioning method described above are applicable to the embodiments of the readable storage medium, and the same technical effects can be achieved.

[0266] It should be noted that many of the functional units described in this specification are referred to as modules to more particularly emphasize their implementation independence.

[0267] In the embodiments of the present application, the modules can be implemented in software, so as to be executed by various types of processors. For example, an identified executable code module can include one or more physical or logical blocks of computer instructions. For example, it can be structured as an object, a procedure, or a function. However, the executable code of the identified module need not be physically located together, but can include different instructions stored in different locations which, when joined logically together, comprise the module and achieve the stated purpose for the module.

[0268] In fact, the executable code module can be a single instruction, or many instructions, and can even be distributed over several different code segments, among different programs, and across several memory devices. Similarly, operational data can be identified within the module, and can be expressed in any data structure or form, and organized according to any suitable method. The operational data can be collected as a single data set, or distributed over different locations including over different storage devices, and can exist even only as electronic signals on a system or network.

[0269] When the modules can be implemented in software, the modules implemented in software can be built into corresponding hardware circuitry by those skilled in the art without considering the cost, including conventional very large scale integration (VLSI) circuitry or gate array, and existing semiconductors such as logic chips, transistors, and other discrete elements. The modules can also be implemented in programmable hardware devices, such as field programmable gate array, programmable array logic, programmable logic device, and the like.

[0270] The above is the preferred embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method of dividing a communication cell, characterized by, The method comprises: obtaining a reference matrix of each communication cell according to the distance between each two communication cells in a to-be-divided set; obtaining a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; obtaining a target feature vector according to the normalized Laplacian matrix; dividing the to-be-divided set according to the target feature vector, so that the difference value between cell subsets corresponding to the to-be-divided set is greater than or equal to a first threshold value, and the difference value between cell subsets is equal to the variance between cell subsets divided by the variance within cell subsets; wherein the to-be-divided set comprises at least two to-be-divided communication cells; the reference matrix comprises an adjacency weight matrix; the adjacency weight matrix is determined according to a weight value and a traffic transfer amount between each two communication cells; the weight value is calculated according to the distance by using a Gaussian kernel function, and the Gaussian kernel function comprises a threshold value for controlling the sparsity of the adjacency weight matrix; the traffic transfer amount is calculated according to the average traffic of the corresponding communication cell in a first time period and the distance.

2. The method of claim 1, wherein, In the case where the reference matrix is an adjacency weight matrix, the method comprises: obtaining a weight value between a first communication cell and each other communication cell according to the distance between the first communication cell and each other communication cell; obtaining an adjacency weight matrix of the first communication cell according to the weight value; wherein the first communication cell is any communication cell in the to-be-divided set, and the other communication cell is any communication cell in the to-be-divided set except the first communication cell.

3. The method of claim 2, wherein, The method further comprises: obtaining the average traffic of the first communication cell in a first time period; obtaining a traffic transfer amount between the first communication cell and each other communication cell according to the average traffic; wherein the method of obtaining an adjacency weight matrix of the first communication cell according to the weight value comprises: obtaining an adjacency weight matrix of the first communication cell according to the weight value and the traffic transfer amount.

4. The method of claim 1, wherein, The method further comprises: obtaining a degree corresponding to each communication cell; obtaining a degree matrix corresponding to the to-be-divided set according to the degrees; obtaining a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix and the degree matrix; wherein the degree represents the sum of the weights of all edges adjacent to the communication cell.

5. The method of claim 1, wherein, The method further comprises: performing eigenvalue decomposition on the normalized Laplacian matrix to obtain a corresponding eigenvalue; obtaining a target feature vector according to the eigenvalue.

6. The method of claim 5, wherein, The method of obtaining a target feature vector according to the eigenvalue comprises: obtaining at least two first non-zero eigenvalues with smaller values from the eigenvalues; According to the at least two first non-zero eigenvalues, a corresponding eigenvector is obtained from the Laplacian matrix as a target eigenvector; Wherein, the maximum value in the at least two first non-zero eigenvalues is less than the minimum value in other non-zero eigenvalues; The other non-zero eigenvalues are non-zero eigenvalues other than the first non-zero eigenvalues.

7. The method of claim 1, wherein, The division of the to-be-divided set according to the target eigenvector comprises: Clustering the target eigenvector to obtain a clustering result; According to the clustering result, the to-be-divided set is divided.

8. The method of claim 7, wherein, The clustering of the target eigenvector to obtain a clustering result comprises: According to the target eigenvector, a feature matrix is constructed; The feature matrix is clustered to obtain a clustering result.

9. The method of claim 1, wherein, Before the to-be-divided set is divided according to the target eigenvector, it further comprises: According to the to-be-divided set, a target undirected graph is constructed; According to the target eigenvector, the target undirected graph is divided into at least two subgraphs that have no connection with each other; According to the at least two subgraphs, at least two cell subsets corresponding to the at least two subgraphs are obtained; Wherein, each of the cell subsets has no intersection, and the union of the at least two cell subsets is equal to the to-be-divided set. After the to-be-divided set is divided according to the target eigenvector, it further comprises:

10. The method of claim 1, wherein, After the to-be-divided set is divided according to the target eigenvector, it further comprises: In the case that the number of cells contained in the first cell subset obtained by division is less than the lower limit value, the first cell subset is merged with a target cell subset; the target cell subset is the cell subset with the smallest distance between the central communication cell and the central communication cell of the first cell subset among the cell subsets obtained by division; In the case that the number of cells contained in the second cell subset obtained by division is greater than the upper limit value, the to-be-divided set is updated according to the second cell subset, and the reference matrix of each communication cell is obtained according to the distance between each two communication cells in the to-be-divided set is executed again; In the case that the number of cells contained in all cell subsets obtained by division meets the constraint condition, and the difference value between the cell subsets corresponding to the to-be-divided set is less than the first threshold value, the number of target eigenvectors obtained is updated, and the target eigenvector is obtained according to the normalized Laplacian matrix is executed again. Wherein, the constraint condition comprises: the number of cells is greater than or equal to the lower limit value, and less than or equal to the upper limit value. Comprise:

11. A communication cell partitioning apparatus characterized by comprising: The first obtaining module is configured to obtain a reference matrix of each communication cell according to the distance between each two communication cells in the to-be-divided set; The second obtaining module is configured to obtain a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; The third obtaining module is configured to obtain a target eigenvector according to the normalized Laplacian matrix; ​ The first dividing module is configured to divide the to-be-divided set according to the target feature vector, so that an inter-cell subset difference value corresponding to the to-be-divided set is greater than or equal to a first threshold value, the inter-cell subset difference value being equal to a variance between cell subsets divided by a variance within the cell subsets; The to-be-divided set includes at least two to-be-divided communication cells; The reference matrix includes an adjacency weight matrix; The adjacency weight matrix is determined according to a weight value and a traffic transfer amount between each two communication cells; The weight value is calculated according to the distance based on a Gaussian kernel function, the Gaussian kernel function including a threshold value for controlling sparsity of the adjacency weight matrix; The traffic transfer amount is calculated according to an average traffic of a corresponding communication cell in a first time period and the distance.

12. A communication cell partitioning device, characterized by Comprise: A processor and a transceiver; The processor is configured to obtain a reference matrix of each communication cell in a to-be-divided set according to a distance between each two communication cells in the to-be-divided set; Obtain a normalized Laplacian matrix corresponding to the to-be-divided set according to the reference matrix; Obtain a target feature vector according to the normalized Laplacian matrix; Divide the to-be-divided set according to the target feature vector, so that an inter-cell subset difference value corresponding to the to-be-divided set is greater than or equal to a first threshold value, the inter-cell subset difference value being equal to a variance between cell subsets divided by a variance within the cell subsets; The to-be-divided set includes at least two to-be-divided communication cells; The reference matrix includes an adjacency weight matrix; The adjacency weight matrix is determined according to a weight value and a traffic transfer amount between each two communication cells; The weight value is calculated according to the distance based on a Gaussian kernel function, the Gaussian kernel function including a threshold value for controlling sparsity of the adjacency weight matrix; The traffic transfer amount is calculated according to an average traffic of a corresponding communication cell in a first time period and the distance.

13. A communication cell partitioning device comprising a memory, a processor, and a program stored on the memory and executable on the processor; characterized in that, The processor implements the communication cell division method of any one of claims 1-10 when executing the program.

14. A readable storage medium, having stored thereon a program, characterized in that, The program is executed by the processor to implement the steps in the communication cell division method of any one of claims 1-10. The program is executed by the processor to implement the steps in the communication cell division method of any one of claims 1-10.

Images