WO2025050573A1 - Method and apparatus for determining main coverage cell of 5g private network - Google Patents

Abstract

The present application relates to the field of wireless communication, and proposes a method and apparatus for determining a main coverage cell of a 5G private network. The method comprises: acquiring neighbor cell measurement statistical data and neighbor cell switching statistical data of a first region, and on the basis of the neighbor cell measurement statistical data and the neighbor cell switching statistical data, obtaining a cell association graph; extracting sample feature variables of a main coverage cell of a sample scenario, and mapping the sample feature variables to the cell association graph to generate corresponding mapping data; inputting the mapping data into a graph neural network GraphSAGE model to obtain features of respective sample nodes, on the basis of the features of the respective sample nodes and labels of the respective sample nodes, training a logistic regression model, and after training is finished, obtaining a 5G private network cell identification model; and extracting feature variables of a scenario cell to be identified, and inputting the feature variables into the 5G private network cell identification model to obtain a prediction result.

Description

A method and device for determining the primary coverage cell of a 5G private network

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority to Chinese patent application No. 202311159469.5 filed in China on September 7, 2023, the entire contents of which are incorporated herein by reference.

Technical Field

The present application relates to the field of wireless communications, and in particular to a method and device for determining a primary coverage cell of a 5G private network.

Background Art

With the development of technologies such as big data, the Internet of Things, and the Industrial Internet, vertical industries are increasingly requiring higher network transmission rates, network latency, and security. The fifth-generation mobile communication technology (5th Generation Mobile Communication Technology, 5G) public network cannot meet the needs of enterprises in terms of security and other aspects, which is driving more and more companies to target 5G private networks.

Based on the principle of "network follows industry and network is built on demand", China Mobile has launched three modes of 5G private network: "Premium, Exclusive, and Premium", and released a number of 5G private network capabilities such as edge computing, super uplink, and network services. It helps industry customers quickly build a customized exclusive network that is safe, reliable, stable in performance, and has visible services, meeting customer needs for data not leaving the site, ultra-low latency, and ultra-large bandwidth, and realizing operational scenarios such as high-definition video backhaul and remote control.

The "Premium" product solution is of ordinary level and uses the resources of China Mobile's main network. Among them, wireless, transmission, and bearer are consistent with China Mobile's main network. The "Exclusive" product solution also uses existing cells, but will reserve special private network resources in the cells. The "Premium" product solution will add new cells and exclusively use the cell resources. The "Premium and Exclusive" solutions require the main coverage cells in the customer scenario to carry private network services, and the "Exclusive" solution requires understanding the main coverage cells in the customer scenario to plan new exclusive sites. Therefore, the output of the specific 5G private network "Premium, Exclusive, and Exclusive" model solutions is based on the precise survey of the 5G cells that cover the user's needs.

The methods for determining the primary coverage area of 5G private networks in related technologies include: on-site testing, which has high accuracy but requires test engineers to conduct full-scale testing on site, which has high time and economic costs. The basic data judgment method does not utilize the existing precise positioning big data resources and related intelligent algorithms for identification, and its accuracy is too low, resulting in poor accuracy of the solution. The method of modeling and determining the main coverage from the perspective of a single cell ignores the network structure, which will reduce the integrity and accuracy of the main coverage cell of the private network.

Summary of the invention

In response to the above problems, a method and device for determining the main coverage cell of a 5G private network are proposed.

The first aspect of the present application proposes a method for determining a primary coverage cell of a 5G private network, comprising:

Obtaining neighboring cell measurement statistics and neighboring cell handover statistics of the first area, and obtaining a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics;

Extracting sample characteristic variables of the main coverage cell of the sample scenario, mapping the sample characteristic variables to the cell association map, and generating corresponding mapping data;

The mapping data is input into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and a logistic regression model is trained according to the characteristics of each sample node and the label of each sample node. After the training, a 5G private network cell identification model is obtained;

Extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction result.

Optionally, obtaining a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics includes:

Obtaining a correlation coefficient between any two cells in the first area according to the neighboring cell measurement statistical data, and constructing an initial cell correlation map according to the correlation coefficient;

The average daily number of switching between any two cells in the first area is obtained according to the neighboring cell switching statistics, and the redundant edges in the initial cell association map are trimmed according to the average daily number of switching and preset trimming regulations to obtain the cell association map.

Optionally, obtaining a correlation coefficient between any two cells in the first area according to the neighboring cell measurement statistical data, and constructing an initial cell correlation map according to the correlation coefficient includes:

For the i cell and the j cell in the first area, the correlation coefficient is:
H _i,j =(X _i +X _j )/Y _i +Y _j )

Wherein, _Xi is the neighboring cell measurement statistics of cell i as the serving cell measured to cell j The number of sampling points, _Yi is the total number of sampling points of cell i as the serving cell, _Xj is the number of sampling points of cell j as the serving cell measured to cell i, and _Yj is the total number of sampling points of cell j as the serving cell;

If the relationship coefficient H _i,j between the i cell and the j cell >a, then there is a connection between the i cell and the j cell, and the i cell and the j cell are connected by an edge, and the association coefficient H _i,j is used as the characteristic value of the edge connecting the i cell and the j cell, where a is the preset association coefficient threshold.

Optionally, obtaining the daily average number of handovers between any two cells in the first area according to the neighboring cell handover statistics, and pruning redundant edges in the initial cell association map according to the daily average number of handovers and a preset clipping rule to obtain the cell association map includes:

For cell a and cell b in the first area, obtaining the daily average number of handovers of each connected edge of cell a and cell b according to the neighboring cell handover statistics;

The connected edges whose daily average switching times are lower than the preset switching times are taken as pre-pruned edges;

For the pre-pruned edge A1 _a,b , if the pre-pruned edge A1 _a,b is not the last edge between cells a and b, and pruning the pre-pruned edge A1 _a,b will not cause the split of the initial cell association graph, then prune the pre-pruned edge A1 _a,b ;

Repeat the above steps to successively remove the redundant edges between the remaining cells to obtain the cell association graph.

Optionally, the extracting of sample characteristic variables of the main coverage cell of the sample scenario includes:

Extracting basic information features of the main coverage cell of the sample scenario, wherein the basic information features include coverage distance, coverage angle, antenna height, longitude, latitude, overlap area, frequency band and location;

The Minimization of Drive-Test (MDT) coverage features of the main coverage cell of the sample scenario are extracted, wherein the MDT coverage features include coverage signal strength, number of coverage sampling points per grid and average coverage area of the scenario.

Optionally, mapping the sample characteristic variables to the cell association map to generate corresponding mapping data includes:

The sample feature variables are mapped to the nodes and connected edges of the cell association graph to obtain corresponding sample nodes and sample connected edges.

Optionally, the step of inputting the mapping data into a graph neural network GraphSAGE model to obtain features of each sample node includes:

Each sample node uses its surrounding K neighbor nodes to sample information until each sample node All points contain adjacent node features;

Through the mean aggregation function, for each sample node, the characteristics of each sample node and the neighboring nodes are aggregated one by one to obtain the characteristics of each sample node.

Optionally, after inputting the characteristic variable into the 5G private network cell identification model to obtain a prediction result, the method further includes:

Compare the prediction result with the preset prediction threshold to determine whether the scene cell to be identified is the main coverage cell of the 5G private network.

Optionally, comparing the prediction result with a preset prediction threshold to determine whether the scene cell to be identified is a 5G private network primary coverage cell includes:

If the probability of the prediction result is not less than the preset prediction threshold, it is determined that the scene cell to be identified is a 5G private network primary coverage cell;

Otherwise, it is determined that the scene cell to be identified is a non-5G private network main coverage cell.

The second aspect of the present application proposes a 5G private network primary coverage cell determination device, comprising:

An association map construction module, configured to obtain neighboring cell measurement statistics and neighboring cell handover statistics of the first area, and obtain a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics;

A feature extraction module, used to extract sample feature variables of the main coverage cell of the sample scenario, map the sample feature variables to the cell association map, and generate corresponding mapping data;

A training module is used to input the mapping data into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and train a logistic regression model according to the characteristics of each sample node and the label of each sample node. After the training, a 5G private network cell identification model is obtained;

The prediction module is used to extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction results.

The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

By proposing a method for identifying association graphs between communication cells, building a wireless communication graph network, using graph structure to reflect the network structure, and using cell physical coverage characteristics and MDT minimized drive test data as the analysis basis, and using massive user drive test data as the analysis basis, the accuracy is higher and time and economic costs are saved. A 5G private network cell identification model based on a two-layer stack of graph neural network and logistic regression is constructed. In view of the particularity of 5G private network coverage scenarios, inter-cell association features are introduced to greatly improve the identification The comprehensiveness and accuracy of the data ensure the quality of service (QOS) for private network users.

Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and/or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

FIG1 is a flow chart of a method for determining a primary coverage cell of a 5G private network according to an embodiment of the present application;

Figure 2 is a block diagram of a 5G private network main coverage cell determination device according to an embodiment of the present application.

DETAILED DESCRIPTION

The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

There are some differences between the identification of private network main coverage cells and general main coverage cells. Private networks require ensuring the quality of service (QOS) of private network users, and need to consider user perception of the entire service process, while the identification of general main coverage cells mainly considers the coverage accuracy of a single cell. Due to this difference, the identification of private network main coverage cells needs to further consider the wireless network structure. When a private network user occupies private network cell A, A and B are strongly associated cells on the association map, and B is a non-private network cell. The private network user may switch to B briefly due to various reasons, such as signal reflection and signal fluctuation. Since B does not reserve resources for private network users, user perception will decrease.

FIG1 is a flow chart of a method for determining a primary coverage cell of a 5G private network according to an embodiment of the present application, including:

Step 101: Obtain neighboring cell measurement statistics and neighboring cell handover statistics of a first area, and obtain a cell association map according to the neighboring cell measurement statistics and neighboring cell handover statistics.

In the embodiment of the present application, the first area can be the total network data of a city, or it can be The total network data of the area is not restricted in this application, but the amount of data cannot be too small, otherwise there will be great limitations in the subsequent training process.

In addition, neighbor cell handover statistics refer to: in traditional network optimization involving neighbor cell optimization, the coverage range is greatly different from the original planned data due to construction or basic data collection errors. It is generally discovered through road testing or abnormal handover success rate, but this method is inefficient and relies on the mechanical method of comparing each cell one by one. There is currently no effective way to combine the handover report with the basic data. The general format of the report in the neighbor cell handover statistics is information such as the number of handovers from the source cell to the target cell and the number of successes. Various tools can import the report into a graphical tool for analysis.

Neighboring cell measurement statistics means that in a wireless network, the terminal first needs to measure the signal level or signal quality of the cell to perform cell selection or cell reselection. When in the RRC_Connected state, the user terminal (User Equipment, UE) needs to report the measurement results to the 5G base station (gNodeB, gNB), that is, the UE continuously sends measurement reports to the base station at certain periodic intervals. The measurement reports include the measurement results of the serving cell and the measurement results of the neighboring cells.

In the embodiment of the present application, a correlation coefficient between any two cells in the first area is obtained based on neighboring cell measurement statistical data, and an initial cell correlation map is constructed based on the correlation coefficient.

Specifically, for the i cell and the j cell in the first area, the correlation coefficient is:
H _i,j =(X _i +X _j )/Y _i +Y _j )

Wherein, _Xi is the number of sampling points of the i cell as the serving cell measured to the j cell in the neighboring cell measurement statistics, _Yi is the total number of sampling points of the i cell as the serving cell, _Xj is the number of sampling points of the j cell as the serving cell measured to the i cell, and _Yj is the total number of sampling points of the j cell as the serving cell;

If the relationship coefficient H _{i,j between cell i and cell j} >a, then there is a connection between cell i and cell j. Cell i and cell j are connected by an edge, and the association coefficient H _i,j is used as the characteristic value of the edge connecting cell i and cell j, where a is the preset association coefficient threshold.

In the embodiment of the present application, there is no specific limitation on the value of the preset correlation coefficient threshold.

In an embodiment of the present application, the daily average number of switching between any two cells in the first area is obtained based on the neighboring cell switching statistics, and the redundant edges in the initial cell association map are trimmed based on the daily average number of switching and preset clipping regulations to obtain the cell association map.

Specifically, for cell a and cell b in the first area, the daily average number of handovers of each connected edge of cell a and cell b is obtained according to the neighboring cell handover statistics;

The connected edges whose daily average switching times are lower than the preset switching times are taken as pre-pruned edges;

For the pre-pruned edge A1 _a,b , if the pre-pruned edge A1 _a,b is not the last edge between cells a and b, and pruning the pre-pruned edge A1 _a,b will not cause the initial cell association graph to split, then prune the pre-pruned edge A1 _a,b ;

Repeat the above steps and cut off the redundant edges between the remaining cells in turn to obtain the cell association graph.

It should be noted that the priority order of edge cutting is that the smaller the number of handovers between cell a and cell b, the higher the priority.

Step 102: extract sample characteristic variables of the main coverage cell of the sample scenario, map the sample characteristic variables to the cell association map, and generate corresponding mapping data.

In the embodiment of the present application, the feature variables include basic information features and minimization of drive-test (MDT) coverage features.

It should be noted that the basic information features include coverage distance, coverage angle, antenna height, longitude, latitude, overlapping area, frequency band and location, and the Minimization of Drive-Test (MDT) coverage features include coverage signal strength, number of coverage sampling points per grid and average coverage area of the scene.

For example, Table 1 is used to introduce the sample characteristic variables in detail.

Table 1

In the embodiment of the present application, the sample scene main coverage cell is an existing manually selected scene main coverage cell, such as an existing school or hospital scene main coverage cell.

Since private network users have high requirements for the QOS of the communication service process, it is necessary to integrate and analyze the characteristics of the communication service cell and its surrounding related cells to ensure that the private network main coverage cell selected by the model can reflect the network structure of the existing network.

In a possible embodiment, a school in the main coverage area of the sample scenario is taken as an example.

First, the spatial information field of the school and the information of the main coverage area are obtained. The cells within this area and the cells within 500 meters around this area are obtained through the spatial information field, and the characteristic variables of these cells are extracted;

After each cell has features, it is associated with the second column of main coverage cells to determine whether these cells are main coverage or non-main coverage, forming training data containing positive and negative samples.

In the embodiment of the present application, it is necessary to fuse the sample feature variables with the cell association map features.

Specifically, the sample feature variables are mapped to the nodes and connected edges of the cell association graph to obtain the corresponding sample nodes and sample connected edges.

Step 103, input the mapping data into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and train a logistic regression model based on the characteristics of each sample node and the label of each sample node. After the training is completed, a 5G private network cell identification model is obtained.

In the related technologies, the Graph Neural Networks (GNN) model is a framework that has emerged in recent years to directly learn graph structure data using deep learning. It extracts and discovers features and patterns in graph structure data to meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. The GraphSAGE algorithm is optimized compared to the traditional GNN graph neural network: First, the collection algorithm was optimized, which greatly improved the modeling efficiency. Second, the neighbor aggregation method was optimized to improve the accuracy of the aggregated features. Logistic regression, also known as logistic regression analysis, is a generalized linear regression analysis model. It belongs to supervised learning in machine learning. Its derivation process and calculation method are similar to the regression process, but in fact it is mainly used to solve binary classification problems. It is one of the most commonly used and most stable binary classification models.

In the embodiment of the present application, after the mapping data is obtained, GraphSAGE sampling and aggregation are performed on the cells of the associated graph.

Specifically, information is sampled from each sample node using its surrounding K neighbor nodes until each sample node contains adjacent node features, where K≤2.

Subsequently, since the cell node location already contains the characteristics of the adjacent nodes, the characteristics of the node and the adjacent nodes are further converged for each node one by one.

Specifically, the mean aggregation function is used for each sample node to aggregate the features of each sample node and its neighboring nodes one by one, to obtain the features of each sample node, and to train a logistic regression model in combination with the labels of each sample node.

It should be noted that GraphSAGE can choose a variety of mean aggregation functions, sum aggregation functions, convolution aggregation functions, long short-term memory (LSTM) aggregation functions, etc. This application uses the sum aggregation function.

It can be understood that model training is an iterative process, which is carried out by continuously adjusting the network parameters of the model until the overall loss function value of the model is less than the preset value, or the overall loss function value of the model no longer changes or changes slowly, the model converges, and a trained model is obtained.

Optionally, the training may be considered finished when a preset number of training times is reached.

Optionally, the training may be considered finished when a preset training time is reached.

Step 104: extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction results.

In an embodiment of the present application, as shown in step 102, basic information features of the scene cell to be identified and minimization of drive test (MDT) coverage features are extracted. The basic information features include coverage distance, coverage angle, antenna height, longitude, latitude, overlapping area, frequency band and position. The MDT coverage features include coverage signal strength, number of coverage sampling points per grid and average coverage area of the scene.

In the embodiment of the present application, after obtaining the prediction result, it is also used to compare the prediction result with the preset prediction threshold. The values are compared to determine whether the scene cell to be identified is the main coverage cell of the 5G private network.

Specifically, if the probability of the prediction result is not less than the preset prediction threshold, the scene cell to be identified is determined to be the 5G private network main coverage cell; otherwise, the scene cell to be identified is determined to be a non-5G private network main coverage cell.

In this way, the main coverage cell of the 5G private network is identified.

This application proposes a method for identifying association graphs between communication cells, constructs a wireless communication graph network, reflects the network structure with a graph structure, takes the physical coverage characteristics of the cell and the MDT minimized drive test data as the analysis basis, and uses massive user drive test data as the analysis basis, which has higher accuracy and saves time and economic costs. It constructs a 5G private network cell identification model based on a two-layer stack of graph neural network and logistic regression, introduces association features between cells in view of the particularity of 5G private network coverage scenarios, greatly improves the comprehensiveness and accuracy of identification, and ensures the service quality QOS of private network users.

FIG2 is a block diagram of a 5G private network primary coverage cell determination device 200 according to an embodiment of the present application, including:

The association map construction module 210 is used to obtain the neighboring cell measurement statistics and the neighboring cell handover statistics of the first area, and obtain the cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics;

The feature extraction module 220 is used to extract sample feature variables of the main coverage cell of the sample scene, map the sample feature variables to the cell association map, and generate corresponding mapping data;

The training module 230 is used to input the mapping data into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and train the logistic regression model according to the characteristics of each sample node and the label of each sample node. After the training, a 5G private network cell identification model is obtained;

The prediction module 240 is used to extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction results.

Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of this disclosure can be achieved, and this document is not limited here.

The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims (10)

A method for determining a primary coverage cell of a 5G private network, comprising: Obtaining neighboring cell measurement statistics and neighboring cell handover statistics of the first area, and obtaining a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics; Extracting sample characteristic variables of the main coverage cell of the sample scenario, mapping the sample characteristic variables to the cell association map, and generating corresponding mapping data; The mapping data is input into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and a logistic regression model is trained according to the characteristics of each sample node and the label of each sample node. After the training, a 5G private network cell identification model is obtained; Extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction result. The method according to claim 1, wherein the obtaining a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics comprises: Obtaining a correlation coefficient between any two cells in the first area according to the neighboring cell measurement statistical data, and constructing an initial cell correlation map according to the correlation coefficient; The average daily number of switching between any two cells in the first area is obtained according to the neighboring cell switching statistics, and the redundant edges in the initial cell association map are trimmed according to the average daily number of switching and preset trimming regulations to obtain the cell association map. The method according to claim 2, wherein obtaining a correlation coefficient between any two cells in the first area according to the neighboring cell measurement statistical data, and constructing an initial cell correlation map according to the correlation coefficient, comprises: For the i cell and the j cell in the first area, the correlation coefficient is:
H _i,j =(X _i +X _j )/Y _i +Y _j ) Wherein, _Xi is the number of sampling points in the neighboring cell measurement statistics where the i cell is measured as a serving cell to the j cell, _Yi is the total number of sampling points where the i cell is used as a serving cell, _Xj is the number of sampling points where the j cell is measured as a serving cell to the i cell, and _Yj is the total number of sampling points where the j cell is used as a serving cell; If the relationship coefficient H _{i,j between cell i and cell j} >a, then there is a connection between cell i and cell j. Use an edge to connect cell i and cell j, and use the association coefficient H _i,j as the edge connecting cell i and cell j. The characteristic value of , where a is the preset correlation coefficient threshold. The method according to claim 2, wherein the obtaining the daily average number of handovers between any two cells in the first area according to the neighboring cell handover statistics, and pruning redundant edges in the initial cell association map according to the daily average number of handovers and a preset clipping rule to obtain the cell association map comprises: For cell a and cell b in the first area, obtaining the daily average number of handovers of each connected edge of cell a and cell b according to the neighboring cell handover statistics; The connected edges whose daily average switching times are lower than the preset switching times are taken as pre-pruned edges; For the pre-pruned edge A1 _a,b , if the pre-pruned edge A1 _a,b is not the last edge between cells a and b, and pruning the pre-pruned edge A1 _a,b will not cause the split of the initial cell association graph, then prune the pre-pruned edge A1 _a,b ; Repeat the above steps to successively remove the redundant edges between the remaining cells to obtain the cell association graph. The method according to claim 1, wherein the extracting the sample characteristic variables of the main coverage cell of the sample scene comprises: Extracting basic information features of the main coverage cell of the sample scenario, wherein the basic information features include coverage distance, coverage angle, antenna height, longitude, latitude, overlap area, frequency band and location; Extract the minimization of drive tests (MDT) coverage features of the primary coverage cell in the sample scenario, wherein the MDT coverage features include coverage signal strength, number of coverage sampling points per grid and average coverage area of the scenario. The method according to claim 1, wherein mapping the sample feature variables to the cell association map to generate corresponding mapping data comprises: The sample feature variables are mapped to the nodes and connected edges of the cell association graph to obtain corresponding sample nodes and sample connected edges. The method according to claim 1, wherein the step of inputting the mapping data into a graph neural network GraphSAGE model to obtain features of each sample node comprises: For each sample node, use the K neighboring nodes around it to sample information until each sample node contains the characteristics of the adjacent nodes; Through the mean aggregation function, each sample node and its neighboring nodes are integrated one by one. The features are aggregated to obtain the features of each sample node. The method according to claim 1, wherein, after inputting the characteristic variable into the 5G private network cell identification model to obtain a prediction result, it also includes: Compare the prediction result with the preset prediction threshold to determine whether the scene cell to be identified is the main coverage cell of the 5G private network. The method according to claim 8, wherein comparing the prediction result with a preset prediction threshold to determine whether the scene cell to be identified is a 5G private network primary coverage cell includes: If the probability of the prediction result is not less than the preset prediction threshold, it is determined that the scene cell to be identified is a 5G private network primary coverage cell; Otherwise, it is determined that the scene cell to be identified is a non-5G private network main coverage cell. A 5G private network primary coverage cell determination device, comprising: An association map construction module, configured to obtain neighboring cell measurement statistics and neighboring cell handover statistics of the first area, and obtain a cell association map according to the neighboring cell measurement statistics and the neighboring cell handover statistics; A feature extraction module, used to extract sample feature variables of the main coverage cell of the sample scenario, map the sample feature variables to the cell association map, and generate corresponding mapping data; A training module is used to input the mapping data into the graph neural network GraphSAGE model to obtain the characteristics of each sample node, and train a logistic regression model according to the characteristics of each sample node and the label of each sample node. After the training, a 5G private network cell identification model is obtained; The prediction module is used to extract the characteristic variables of the scene cell to be identified, input the characteristic variables into the 5G private network cell identification model, and obtain the prediction results.