Modeling methods, devices, equipment, media, and program products for community relationships
By introducing a graph neural network algorithm into the joint area optimization of wireless networks, a cell relationship representation model is constructed, which solves the problem of incomplete extraction of inter-cell influence relationships in existing technologies, and achieves more efficient and accurate optimization results, especially improving the accuracy in fault root cause localization.
Patent Information
- Application Number
- CN202510617801.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies cannot accurately extract the influence relationships between cells in the joint area optimization of wireless networks, resulting in low optimization accuracy and efficiency. Furthermore, existing methods fail to fully consider the mutual influence between cells and neighboring cells.
By introducing cell-level and neighbor-cell-level data, a relationship representation model is constructed using graph neural network algorithms. The graph neural network algorithm is used for multi-level iteration to calculate the influence relationship between cells, and to distinguish between central cells and edge cells, updating only the data of central cells.
It improves the accuracy and efficiency of joint area optimization, and can accurately extract the influence relationship between cells, especially significantly improving the accuracy in sub-tasks such as root cause localization.
Smart Images

Figure CN123123972A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of wireless network technology, and in particular to a method, apparatus, device, medium, and program product for modeling cell relationships. Background Technology
[0002] In wireless network operation and maintenance, joint area optimization is an important scenario. Each area contains multiple network elements, i.e., cells in the wireless network. The parameter settings of a current cell not only affect its own network performance but also the network performance of its neighboring cells. The indicators collected by the wireless network monitoring center include both indicators reflecting only the current cell's own status and indicators reflecting the mutual influence between the current cell and its neighboring cells. Therefore, when designing optimization schemes, it is necessary to consider not only the impact of the optimization scheme on the current cell itself but also the mutual influence between the current cell and its neighboring cells.
[0003] In existing joint area optimization scenarios, the common method is to use a unified representation model of a single cell for joint area optimization. However, the construction of this model only refers to indicators reflecting the state of each cell, and the referenced data is not comprehensive enough. It cannot accurately extract the influence relationship between cells, resulting in low optimization accuracy and efficiency. Summary of the Invention
[0004] In view of the above problems, this disclosure is made to provide a method, apparatus, device, medium and program product for modeling cell relationships.
[0005] According to one aspect of this disclosure, a method for modeling cell relationships is provided, comprising:
[0006] Based on the indicators representing cell status collected by the wireless network monitoring center, cell-level data and neighbor-cell-level data within the area are obtained. The cell-level data is used to represent the status of each cell in the area, and the neighbor-cell-level data is used to represent the influence relationship between each cell and its neighboring cells in the area.
[0007] Based on the cell-level data and the neighbor cell-level data, a graph neural network algorithm is used to obtain a relationship representation model that characterizes the influence relationship between cells within the area.
[0008] Furthermore, according to a cell relationship modeling method of one aspect of this disclosure, based on the cell-level data and the neighbor cell-level data, a graph neural network algorithm is used to obtain a relationship representation model for characterizing the influence relationship between cells within the area, including:
[0009] The cell-level data and neighbor cell-level data are preprocessed, including data segmentation and dimensionality compression.
[0010] The preprocessed cell-level data is used as nodes, and the preprocessed neighbor-cell-level data is used as edges, which are then input into a graph neural network algorithm.
[0011] Based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, calculate the nodes and edges of the Lth layer of the graph neural network algorithm, where L is a positive integer greater than or equal to 2; the nodes of the first layer of the graph neural network algorithm are preprocessed cell-level data, and the edges of the first layer are preprocessed neighbor-cell-level data;
[0012] When the Lth layer is the last layer of the graph neural network algorithm, a relational representation model for representing the influence relationship between cells within the area is obtained based on the nodes and edges of the Lth layer.
[0013] If the Lth layer is not the last layer of the graph neural network algorithm, update L = L + 1, and return to the steps of executing the nodes and edges of the (L-1)th layer based on the graph neural network algorithm, and calculating the nodes and edges of the Lth layer of the graph neural network algorithm, until the updated Lth layer is the last layer of the graph neural network algorithm.
[0014] Furthermore, according to a cell relationship modeling method of one aspect of this disclosure, the nodes and edges of the Lth layer of the graph neural network algorithm are calculated based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, including:
[0015] Based on each node of the (L-1)th layer and the linear matrix of the Lth layer of the graph neural network algorithm, the attention score representing the association relationship between any two adjacent nodes in the Lth layer is calculated.
[0016] Based on the attention score and the edge between the two adjacent nodes in the (L-1)th layer, calculate the edge between the target initial adjacent nodes corresponding to the two adjacent nodes in the Lth layer;
[0017] Using the edge between the target initial adjacent nodes in the Lth layer and the two adjacent nodes in the (L-1)th layer, calculate the target initial adjacent nodes in the Lth layer;
[0018] Obtain the cell type represented by the initial neighboring nodes of the target;
[0019] Based on the cell type represented by the target initial neighbor node, the final neighbor node of the Lth layer is determined from the target initial neighbor node and the two neighbor nodes; and the edge between the final neighbor node of the Lth layer is determined from the edge between the two neighbor nodes and the edge between the target initial neighbor node.
[0020] Furthermore, according to a cell relationship modeling method of one aspect of this disclosure, based on the cell type represented by the target initial neighbor node, the final neighbor node of the Lth layer is determined from the target initial neighbor node and the two neighbor nodes, including:
[0021] If the cell type represented by the target initial neighbor node is a central cell, the target initial neighbor node is taken as the final neighbor node of the Lth layer.
[0022] If the cell type represented by the initial neighboring node of the target is an edge cell, the two neighboring nodes are regarded as the final neighboring nodes of the Lth layer.
[0023] Furthermore, according to a cell relationship modeling method of one aspect of this disclosure, determining the edges between the final adjacent nodes in the Lth layer from the edges between the two adjacent nodes and the edges between the target initial adjacent nodes includes:
[0024] When the cell type represented by the target initial neighbor node is a central cell, the edge between the target initial neighbor node is taken as the edge between the final neighbor node in the Lth layer;
[0025] When the cell type represented by the initial neighboring node of the target is an edge cell, the edge between the two neighboring nodes is taken as the edge between the final neighboring nodes of the Lth layer.
[0026] Furthermore, according to the cell relationship modeling method of one aspect of this disclosure, obtaining the cell type represented by the target initial neighboring nodes includes:
[0027] Based on the number of signal handovers and coverage overlap between cells, obtain the n cells with the highest relationship to the cells represented by the initial adjacent nodes of the target.
[0028] The n cells with the highest relationship density are taken as the neighboring cells of the cell represented by the target initial neighboring node;
[0029] When all neighboring cells are within the coverage area of the area, the cell type represented by the target initial neighboring node is a central cell; the coverage area of the area is determined based on the signal coverage equipment installed within the area.
[0030] If not all neighboring cells are within the coverage area of the area, the target initial neighboring node represents the cell type as an edge cell.
[0031] Furthermore, according to the cell relationship modeling method of one aspect of this disclosure, the cell-level data and neighbor-cell-level data are preprocessed, including:
[0032] Based on the acquisition frequency and acquisition time range of the wireless network monitoring center, the cell-level data and the neighbor cell-level data are segmented according to a preset time length to obtain cell-level data and neighbor cell-level data in time block form;
[0033] Dimensional compression is performed on the neighboring cell-level data in the form of time blocks.
[0034] According to another aspect of this disclosure, a modeling apparatus for cell relationships is provided, comprising:
[0035] The acquisition module is used to acquire cell-level data and neighbor-cell-level data within a region based on indicators representing cell status collected by the wireless network monitoring center. The cell-level data is used to represent the status of each cell included in the region, and the neighbor-cell-level data is used to represent the influence relationship between each cell and its neighboring cells within the region.
[0036] The calculation module is used to obtain a relationship representation model that characterizes the influence relationship between cells within the area, based on the cell-level data and the neighbor cell-level data, using a graph neural network algorithm.
[0037] According to another aspect of this disclosure, a computer device is provided, including a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to implement the method of one aspect above.
[0038] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method of one aspect above.
[0039] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of the above-described aspect.
[0040] As will be described in detail below, a cell relationship modeling method, apparatus, device, medium, and program product according to embodiments of the present disclosure introduces cell-level data and neighbor-cell-level data, and uses a graph neural network algorithm to construct a relationship representation model that represents the influence relationship between cells within a region. It introduces the full range of neighbor-cell relationships, that is, it utilizes data on the mutual influence relationship between cells and their neighboring cells, which can accurately extract the influence relationship between cells. The relationship representation model constructed by the present disclosure can be used as an independent module in various sub-tasks in the joint region optimization scenario, which is beneficial to improving the accuracy and efficiency of optimization.
[0041] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description
[0042] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.
[0043] Figure 1 This is a flowchart illustrating a method for modeling cell relationships according to an embodiment of this disclosure.
[0044] Figure 2 This is another flowchart illustrating a method for modeling cell relationships according to an embodiment of the present disclosure.
[0045] Figure 3 This is a flowchart illustrating the construction of a classification model according to an embodiment of the present disclosure.
[0046] Figure 4 This is a schematic diagram of the structure of a cell relationship modeling apparatus according to an embodiment of the present disclosure.
[0047] Figure 5 This is a schematic diagram illustrating the structure of a computer device according to an embodiment of the present disclosure.
[0048] Figure 6 This is a schematic diagram illustrating a computer program product according to an embodiment of the present disclosure. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.
[0050] In wireless network operation and maintenance, joint area optimization is an important scenario. Each area contains multiple network elements, i.e., cells in the wireless network. The parameter settings of a current cell not only affect its own network performance but also the network performance of its neighboring cells. The indicators collected by the wireless network monitoring center include both indicators reflecting only the current cell's own status and indicators reflecting the mutual influence between the current cell and its neighboring cells. Therefore, when designing optimization schemes, it is necessary to consider not only the impact of the optimization scheme on the current cell itself but also the mutual influence between the current cell and its neighboring cells.
[0051] In existing joint area optimization scenarios, the common method is to use a unified representation model of a single cell for joint area optimization. However, the construction of this model only refers to indicators reflecting the state of each cell, such as KPI, MR (cell-level), engineering parameters, and alarms of the cell itself. It does not introduce the full range of neighbor cell relationships, that is, it does not utilize the indicators of mutual influence between the current cell and its neighbor cells. The referenced data is not comprehensive enough, and it is impossible to accurately extract the influence relationship between cells, resulting in low optimization accuracy and efficiency.
[0052] In addition, there is the use of human experience to observe the mutual influence of important indicators between cells and design a representation scheme that is applicable to multiple cells. Although the learning of the relationship between cells is introduced, the data is not comprehensive enough and the structure of the representation scheme is too simple to accurately extract the relationship between cells.
[0053] In addition, the division of the area is not precise enough. Since the neighboring areas of the community need to be included, and the neighboring areas have new neighboring areas, there will inevitably be situations where the neighboring areas are not in the current area. The existing technology ignores this situation, which leads to inaccurate learning of the influence relationship between communities.
[0054] The above description, with reference to the accompanying drawings, illustrates a method, apparatus, device, medium, and program product for modeling cell relationships according to embodiments of the present disclosure. By introducing cell-level data and neighbor-cell-level data, a relationship representation model characterizing the influence relationships between cells within a region is constructed using a graph neural network algorithm. The model incorporates all neighbor-cell relationships, utilizing data showing mutual influence between cells and their neighbors, thus accurately extracting the influence relationships between cells. The relationship representation model constructed in this disclosure can be used as an independent module in various sub-tasks within a joint region optimization scenario, which is beneficial for improving the accuracy and efficiency of optimization.
[0055] Cell data is stored separately as cell-level data and neighbor-cell-level data. Cell-level data includes metrics representing the cell itself, such as KPIs, MR (cell-level), parameter tuning (cell-level), and operating parameters. Neighbor-cell-level data includes MR (neighbor-cell-level) and parameter tuning (neighbor-cell-level). The neighbor-cell-level data is organized into a complete data representation of N*N cell relationships, arranged in pairs of cells. Furthermore, the neighbor-cell-level data is structured in time blocks, which facilitates interaction between graph neural network algorithms when learning the relationships between cells and their neighbors.
[0056] Using cell-level data as nodes and neighboring cell-level data as edges, multi-attribute nodes and multi-attribute edges are formed. Attention scores between cells (nodes) are calculated using a graph neural network algorithm, and the edges are updated. Then, the nodes are updated using the updated edges. Through multiple iterations, the representation learning process of the influence relationship between cells is completed, and finally, a relationship representation model is obtained.
[0057] Using signal coverage devices (such as physical points and multi-system combining platforms (POIs)) to divide areas improves the accuracy of area division. The concept of a central cell is introduced, classifying cells within an area into central cells and edge cells. A central cell is one whose neighboring cells are all within the current area, while an edge cell is one whose neighboring cells are only partially located within the current area. During each update, only the data of central cells is updated, excluding the data of edge cells. This approach considers cases where neighboring cells are not within the current area, resulting in a more accurate understanding of the influence relationships between cells.
[0058] To facilitate understanding of this embodiment, a cell relationship modeling method disclosed in this disclosure will first be described in detail. The execution entity of the cell relationship modeling method provided in this disclosure is generally a computer device with certain computing capabilities. This computer device may include, for example, a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, vehicle-mounted device, wearable device, etc. In some possible implementations, the cell relationship modeling method can be implemented by a processor calling computer-readable instructions stored in memory.
[0059] like Figure 1 The diagram shows a flowchart of a cell relationship modeling method provided in this embodiment of the present disclosure, the method including S101-S102:
[0060] S101: Based on the indicators representing the cell status collected by the wireless network monitoring center, obtain cell-level data and neighbor cell-level data within the area.
[0061] In this context, a "region" refers to any region within the joint region optimization scenario. The coverage area of a region can be determined based on signal coverage equipment, such as a multi-system combining platform (POI) or physical points, enabling more precise region delineation. Cell-level data characterizes the individual status of each cell within a region, including indicators such as PM, NRM cell parameters, and engineering parameters. Neighbor-level data characterizes the influence relationship between each cell and its neighboring cells within the region, including indicators such as NRM frequency, MR, NRM neighbor cell relationships, and neighbor cell parameter tuning records.
[0062] S102: Based on cell-level and neighbor-cell-level data, a graph neural network algorithm is used to obtain a relationship representation model for characterizing the influence relationship between cells within the area.
[0063] The graph neural network algorithm includes multiple hidden layers. Specifically, S102 includes the following steps 1-4:
[0064] Step 1: Preprocess the cell-level data and neighbor cell-level data.
[0065] Preprocessing includes data splitting and compression, specifically including:
[0066] Based on the acquisition frequency and acquisition time range of the wireless network monitoring center, cell-level data and neighbor-cell-level data are segmented according to the preset time length to obtain cell-level data and neighbor-cell-level data in time block form; the neighbor-cell-level data in time block form is then subjected to dimensional compression.
[0067] For example, the wireless network monitoring center collects data at a frequency of 15 minutes. For data collected over a period of one hour or one day, it is copied at a 15-minute granularity and aligned. For text-type data, one-hot encoding is used to convert it into a numerical representation. The entire long time series is divided into four time blocks to obtain data in time block format, which serves as the smallest input unit for the graph neural network algorithm.
[0068] Since the graph neural network algorithm involves traversing and stacking each dimension of the neighboring data during the computation process, if the original neighboring data has too many dimensions, it can easily cause the intermediate matrix to have too large a dimension. Therefore, dimensional compression can improve computational efficiency.
[0069] Step 2: Input the preprocessed cell-level data as nodes and the preprocessed neighbor-cell-level data as edges into the graph neural network algorithm.
[0070] Specifically, let's assume the node (cell-level data) is X, and the edge (neighbor-level data) is E:
[0071] The cell-level data X before preprocessing is (batch, cell, t, d1); where batch is the number of data entries input to the graph neural network algorithm at one time, cell is the number of cells in the area, t is the number of time entries in the sequence, d1 is the feature dimension of the cell-level data, and d1 is equal to the sum of the number of data type indicators and the length of the indicators after one-hot encoding of the text type.
[0072] Preprocessed cell-level data (nodes) X: (batch, t, cell, out_len), where out_len is the dimension of the output space.
[0073] The preprocessed neighbor-level data E: (batch, cell, neighbor, t, d2); where cell is the number of cells in the area, neighbor is the number of neighboring cells. In this embodiment, a square matrix is used, so cell = neighbor. t is the number of time entries in the sequence, and d2 is the feature dimension of the neighbor-level data. d2 is equal to the sum of the number of data type indicators and the length of the indicators after one-hot encoding of the text type.
[0074] The preprocessed neighbor-level data (edges) E: (batch,t,cell,neighbour,dc_2), where dc_2 is the feature dimension of the compressed neighbor-level data.
[0075] Step 3: Based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, calculate the nodes and edges of the Lth layer of the graph neural network algorithm.
[0076] In this context, the Lth layer is the hidden layer of the Lth layer, where L is a positive integer greater than or equal to 2. The nodes of the first layer of the graph neural network algorithm are preprocessed cell-level data, and the edges of the first layer are preprocessed neighbor-cell-level data.
[0077] Specifically, step 3 includes the following steps:
[0078] Step 3.1: Based on the nodes of the (L-1)th layer and the linear matrix of the Lth layer of the graph neural network algorithm, calculate the attention score that represents the relationship between any two adjacent nodes in the Lth layer.
[0079] Specifically, any two adjacent nodes X i and X j The formula for calculating the attention score between individuals is:
[0080]
[0081] in, For any two adjacent nodes X in the (L-1)th layer i and X j Attention score of the relationship between them For node X in layer L-1 i Features For node X in layer L-1 j The characteristic of is calculated using the following formula:
[0082] g L (X L-1 ) = X L-1 W L
[0083] Among them, W L Let X be a linear matrix of the Lth layer.L-1 Let L be any node in the Lth layer.
[0084] Node X i and X j The features are concatenated and multiplied by a parameterized weight vector α∈R. 2*out_len Then apply LeakyReLU to activate softmax.
[0085] It should be noted that, regardless of whether it is at level L-1 or level L, node X i and node X j The adjacency relationships between nodes remain unchanged, but the characteristics of the nodes will change. For example, in layer L-1, node X i The characteristic is A, and at level L, node X i The characteristic of has changed to B.
[0086] Step 3.2: Based on the attention score and the edge between two adjacent nodes in the (L-1)th layer, calculate the edge between the target initial adjacent nodes corresponding to the two adjacent nodes in the Lth layer.
[0087] For example, suppose two adjacent nodes in the (L-1)th layer are nodes X. i With node X j In the Lth layer, the initial neighboring node of the target corresponding to two adjacent nodes is node X. i 'With node X j '. With node X i For example, node X i With node X i Although they are the same node, they are located in different layers and have different characteristics. For ease of distinction, this embodiment classifies them as node X. i and node X i '.
[0088] Specifically, attention scores and node X of layer L-1 i With node X j The edge between Perform the Hadamard product and calculate the node X of the Lth layer. i 'With node X j The edge between 'E' L The calculation process is as follows:
[0089]
[0090] E L =α L
[0091] in, This represents the edge between node i' and node j' in the Lth layer.
[0092] Step 3.3: Using the edges between the initial adjacent nodes of the target in layer L and two adjacent nodes in layer L-1, calculate the initial adjacent nodes of the target in layer L;
[0093] Specifically, the edge E of the Lth layer is used. L Node X of layer L-1 i With node X j Calculate the node X in the Lth layer. i 'With node X j '.
[0094] Node X in layer L L (Node X) i 'or node X j ') The edge E of the Lth layer is calculated L and node X of layer L-1 L-1 (Node X) i With node X j We obtain the edge E by iterating through the feature channels p of each edge and calculating the edge E of feature channel p. L and node X of layer L-1 L-1 The product is calculated by multiplying the edge of the i-th cell with the node, traversing all cells, and summing all the results to obtain the final product. The calculation results of all feature channels are concatenated to form the node X of the L-th layer. L The shape is (batch, timestamp, cell, dc_2*out_len), and the calculation process is as follows:
[0095]
[0096] Where || is the concatenation operation, σ is the activation function, and X L For the node of layer L, E for different feature channels p L .
[0097] Step 3.4: Obtain the cell type represented by the initial neighboring nodes of the target.
[0098] Among them, the types of communities include central communities and peripheral communities. Central communities are communities whose neighboring communities are all located in the area, while peripheral communities are communities whose neighboring communities are not all located in the area.
[0099] Specifically, the method for determining the type of residential community is as follows:
[0100] Based on the number of signal handovers and coverage overlap between cells, the n cells with the highest relationship density with the cell represented by the target initial neighbor node are obtained; the n cells with the highest relationship density are taken as the neighbor cells of the cell represented by the target initial neighbor node; when all neighbor cells are within the coverage area of the area, the cell type represented by the target initial neighbor node is a central cell; when not all neighbor cells are within the coverage area of the area, the cell type represented by the target initial neighbor node is an edge cell.
[0101] The coverage area of a region is determined based on the signal coverage equipment installed within that region, such as physical points or points of interest (POIs). Assuming the initial neighboring nodes of the target represent cell A, and the n cells with the highest relationship density are the top-n neighboring cells, if cell A and its top-n neighboring cells are all located in the same region, then cell A is the central cell; otherwise, cell A is the edge cell. Optionally, a mask identifier distinguishes between central and edge cells at the cell dimension of the data, with central cells marked as 1 and edge cells marked as 0.
[0102] Step 3.5: Based on the cell type represented by the target initial neighbor node, determine the final neighbor node of the Lth layer from the target initial neighbor node and the two neighbor nodes; and determine the edge between the final neighbor nodes of the Lth layer from the edge between the two neighbor nodes and the edge between the target initial neighbor node.
[0103] Specifically, determining the final adjacent nodes of the Lth layer includes:
[0104] When the target initial neighbor node represents a central cell, the target initial neighbor node is taken as the final neighbor node of the Lth layer; when the target initial neighbor node represents a peripheral cell, the two neighbor nodes are taken as the final neighbor nodes of the Lth layer.
[0105] Specifically, determining the edges between the final adjacent nodes in the Lth layer includes:
[0106] When the target initial neighboring node represents a central cell, the edge between the target initial neighboring node is taken as the edge between the final neighboring nodes in the Lth layer; when the target initial neighboring node represents a peripheral cell, the edge between two neighboring nodes is taken as the edge between the final neighboring nodes in the Lth layer.
[0107] Optionally, the mask identifier of the central cell can be applied to the nodes and edges of the Lth layer, so that only the central cell is updated and the edge cells are not updated, as shown in the following formula:
[0108] X L =X L*C+X L-1 *(1-C)
[0109] E L =E L *C+E L-1 *(1-C)
[0110] Among them, X L E is the final neighboring node of the Lth layer. L Let C be the edge between the final adjacent nodes in the Lth layer, and C be the mask identifier of the central cell. C=1 indicates that the cell type represented by the node is the central cell, and C=0 indicates that the cell type represented by the node is the edge cell.
[0111] Step 4: In the case that the Lth layer is the last layer of the graph neural network algorithm, obtain a relational representation model based on the nodes and edges of the Lth layer to represent the influence relationship between cells within the area;
[0112] If the Lth layer is not the last layer of the graph neural network algorithm, update L = L + 1 and return to step 3 until the updated Lth layer is the last layer of the graph neural network algorithm.
[0113] like Figure 2 The diagram shown is another flowchart of a cell relationship modeling method provided in this disclosure, the method including S201-S210:
[0114] S201: Determine the coverage area of a region (e.g., region P) based on POI or physical point.
[0115] S202: For each cell in area P, determine the top-n neighboring cells with the highest relationship with each cell based on the neighbor cell configuration table, the number of signal handovers, and the coverage overlap range.
[0116] S203: Determine the community type for each community.
[0117] Cell types include central cells and edge cells. Cells whose top-n neighboring cells are all located in area P are marked as central cells, and cells whose top-n neighboring cells are not all located in area P are marked as edge cells.
[0118] S204: Process the indicators representing cell status collected by the wireless network monitoring center.
[0119] The indicators representing the cell status collected by the wireless network monitoring center are divided into cell-level data and neighbor-cell-level data. Cell-level data reflects the cell's own status, such as PM, NRM cell, and engineering parameters. Neighbor-cell-level data is the data between the cell and its neighboring cells, such as NRM frequency points, MR, NRM neighbor cell relationships, and neighbor cell parameter tuning records.
[0120] Traverse all cells in region P and compare them pairwise. If cell j is a neighbor of cell i, then Adj[i,j] = 1; otherwise, record it as 0. Finally, obtain the adjacency matrix.
[0121] For the cell dimension of the data, a mask identifier C is set to distinguish between central cells and edge cells. If it is a central cell, C=1, and if it is an edge cell, C=0.
[0122] S205: Preprocess cell-level data and neighbor cell-level data.
[0123] The preprocessing is structured processing, including data splitting and dimensional compression:
[0124] Based on the acquisition frequency and acquisition time range of the wireless network monitoring center, cell-level data and neighbor-cell-level data are segmented according to the preset time length to obtain cell-level data and neighbor-cell-level data in time block form; the neighbor-cell-level data in time block form is then subjected to dimensional compression.
[0125] For example, the wireless network monitoring center collects data at a frequency of 15 minutes. For data collected over a period of one hour or one day, it is copied at a 15-minute granularity and aligned. For text-type data, one-hot encoding is used to convert it into a numerical representation. The entire long time series is divided into four time blocks to obtain data in time block format, which serves as the smallest input unit for the graph neural network algorithm.
[0126] The definition of the smallest input unit for a graph neural network algorithm is as follows:
[0127] Node (X): Cell-level data X:(batch,cell,t,d1); where batch is the number of data entries input into the model at one time, cell is the number of cells in area P, t is the number of time entries in the sequence, and d1 is the feature dimension of the cell-level data, which is equal to the sum of the number of data type indicators and the length of the indicators after one-hot encoding of the text type.
[0128] Edge (E): Neighbor-level data E:(batch,cell,neighbour,t,d2); where cell is the number of cells, neighbor is the number of neighbor cells. In this embodiment, a square matrix is used, so cell = neighbor. t is the number of time entries in the sequence, and d2 is the feature dimension of the neighbor-level data, which is equal to the sum of the number of data type indicators and the length of the indicators after one-hot encoding of the text type.
[0129] Adjacency matrix (A): The adjacency matrix between cells, with the shape (cell, neighbor), denoted as A.
[0130] Mask identifier (C): Identifier for the central cell, shaped like a (cell), with the central cell denoted as 1 and the peripheral cells as 0. Further dimension expansion yields the corresponding mask matrix, which can be denoted as C=1 or C=0.
[0131] In addition, this embodiment introduces a graph neural network algorithm to represent the relationship between edges and nodes in the region. The overall model architecture of the graph neural network algorithm consists of 8 hidden layers with identical structures.
[0132] Each hidden layer consists of eight header graph attention layers, one linear layer, and a center cell mask mechanism. The input to each header graph attention layer is cell-level data from the same layer and neighbor-level data after compression. The outputs of the eight header graph attention layers are concatenated and then linearly transformed by the linear layer to restore the original dimensions before compression, which are then used as the input to the next hidden layer. Each hidden layer updates edges and nodes once. In this embodiment, the following steps use the Lth hidden layer as an example to illustrate the specific calculation process.
[0133] S206: Calculate the attention score between each pair of nodes in the Lth layer based on the features of the nodes in the L-1th layer.
[0134] First, the data dimensions of the nodes are transformed, and through linear transformation, the feature dimensions of the cell-level data are transformed into the dimensions of the output space, namely X:(batch,t,cell,out_len).
[0135] Any two adjacent nodes X i and X j The formula for calculating the attention score between individuals is:
[0136]
[0137] in, For any two adjacent nodes X in the (L-1)th layer i and X j Attention score of the relationship between them For node X in layer L-1 i Features For node X in layer L-1 j The characteristic of is calculated using the following formula:
[0138] g L (X L-1 ) = X L-1 W L
[0139] Among them, W L Let X be a linear matrix of the Lth layer. L-1Let L be any node in the Lth layer.
[0140] Node X i and X j The features are concatenated and multiplied by a parameterized weight vector α∈R. 2*out_len Then apply LeakyReLU to activate softmax.
[0141] It should be noted that, regardless of whether it is at level L-1 or level L, node X i and node X j The adjacency relationships between nodes remain unchanged, but the characteristics of the nodes will change. For example, in layer L-1, node X i The characteristic is A, and at level L, node X i The characteristic of has changed to B.
[0142] S207: Calculate the features of the edges in the Lth layer using attention scores.
[0143] For example, suppose two adjacent nodes in the (L-1)th layer are nodes X. i With node X j In the Lth layer, the initial neighboring node of the target corresponding to two adjacent nodes is node X. i 'With node X j '. With node X i For example, node X i With node X i Although they are the same node, they are located in different layers and have different characteristics. For ease of distinction, this embodiment classifies them as node X. i and node X i '.
[0144] Attention score and node X of layer L-1 i With node X j The edge between Perform the Hadamard product and calculate the node X of the Lth layer. i 'With node X j The edge between 'E' L The calculation process is as follows:
[0145]
[0146] E L =α L
[0147] in, This represents the edge between node i' and node j' in the Lth layer.
[0148] S208: Calculate the features of nodes in the Lth layer based on the features of edges in the Lth layer and the features of nodes in the (L-1)th layer.
[0149] Specifically, the edge E of the Lth layer is used. L Node X of layer L-1 i With node X j Calculate the node X in the Lth layer. i 'With node X j '.
[0150] Node X in layer L L (Node X) i 'or node X j ') The edge E of the Lth layer is calculated L and node X of layer L-1 L-1 (Node X) i With node X j We obtain the edge E by iterating through the feature channels p of each edge and calculating the edge E of feature channel p. L and node X of layer L-1 L-1 The product is calculated by multiplying the edge of the i-th cell with the node, traversing all cells, and summing all the results to obtain the final product. The calculation results of all feature channels are concatenated to form the node X of the L-th layer. L The shape is (batch, timestamp, cell, dc_2*out_len), and the calculation process is as follows:
[0151]
[0152] Where || is the concatenation operation, σ is the activation function, and X L For the node of layer L, E for different feature channels p L .
[0153] S209: Features of the nodes in the Lth layer obtained by splicing the attention layer of the multi-head graph.
[0154] S206-S208 are the calculation steps for one of the head graph attention layers. The calculation process for the 8 head graph attention layers is the same. All the nodes obtained from the 8 head graph attention layers of the Lth layer are concatenated together, with the shape X:(batch,timestamp,cell,8*dc_2*out_len). The dimension of the node is restored to the initial dimension d1 through a linear layer. Therefore, the shape of the node is restored to X:(batch,timestamp,cell,d1), which facilitates the iterative update of each layer.
[0155] S210: Based on the mask mechanism of the central cell, the final nodes and edges are obtained.
[0156] Applying the mask identifier of the central cell to the nodes and edges of the Lth layer ensures that only the central cell is updated, and the edge cells are not updated. This can be expressed as the following formula:
[0157] X L =X L *C+X L-1 *(1-C)
[0158] E L =E L *C+E L-1 *(1-C)
[0159] Among them, X L E is the final neighboring node of the Lth layer. L Let C be the edge between the final adjacent nodes in the Lth layer, and C be the mask identifier of the central cell. C=1 indicates that the cell type represented by the node is the central cell, and C=0 indicates that the cell type represented by the node is the edge cell.
[0160] S206-S210 above describes the message passing process for each hidden layer. By stacking 8 layers, the final nodes and edges can be obtained. After obtaining the final edges, a linear layer is used to restore the feature dimensions of the edges.
[0161] This disclosure provides an application example: the cell relationship modeling method can be integrated into the cell relationship calculation module for direct application. This embodiment takes the problem of fault root cause location in the joint area optimization scenario of wireless network as an example.
[0162] When a work order is issued in a community, root cause analysis is required to pinpoint the cause of the fault. Traditional methods only consider the indicators of the current community, while joint area optimization needs to consider the impact of other communities on the current community, thereby improving the accuracy of root cause analysis. Introducing a community relationship calculation module allows for the rapid construction of a classification model for joint area optimization, such as... Figure 3 As shown, the main steps are as follows:
[0163] Data definition: Cell-level data X, neighbor-level data E, neighbor relationship A, central cell matrix C.
[0164] Training process:
[0165] Step 1: Using 50,000 work orders from a city in the current network as training samples, pull the cell-level data and neighbor cell-level data of the work orders for 7 days in history, and encode them using the encoder module of transformer to obtain the cell-level representation X and the neighbor cell-level data representation E.
[0166] Step 2: Treat the cell-level data X as a node and the neighbor-cell-level data E as an edge, and input them into the cell relationship calculation module to obtain the updated node X.
[0167] Step 3: Connect two fully connected layers as a classification network. The output parameters of the last fully connected layer are equal to the number of fault categories. Classify the nodes into N categories and use the cross-entropy loss function as the loss function to train the model.
[0168] Inference process: Access cell-level data X, neighbor-level data E, neighbor relationships A, central cell matrix C, load the model, and obtain the prediction and classification results.
[0169] After experimental comparison, the results of the previous scheme were as follows: the accuracy of the single-element classification model was 72.5%, the accuracy of the multi-element CLS relationship learning model was 76%, and the accuracy of this embodiment was 82.9%. Therefore, it can be seen that the accuracy has been significantly improved.
[0170] According to another aspect of the embodiments of this disclosure, a modeling apparatus for cell relationships is provided, such as... Figure 4 As shown, the device includes:
[0171] The acquisition module 401 is used to acquire cell-level data and neighbor-cell-level data within a region based on indicators representing cell status collected by the wireless network monitoring center. The cell-level data is used to represent the status of each cell included in the region, and the neighbor-cell-level data is used to represent the influence relationship between each cell and its neighboring cells within the region.
[0172] The calculation module 402 is used to obtain a relationship representation model that characterizes the influence relationship between cells within the area based on the cell-level data and the neighbor cell-level data and using a graph neural network algorithm.
[0173] In one or more embodiments, the computing module 402 is used for:
[0174] The cell-level data and neighbor cell-level data are preprocessed, including data segmentation and dimensionality compression.
[0175] The preprocessed cell-level data is used as nodes, and the preprocessed neighbor-cell-level data is used as edges, which are then input into a graph neural network algorithm.
[0176] Based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, calculate the nodes and edges of the Lth layer of the graph neural network algorithm, where L is a positive integer greater than or equal to 2; the nodes of the first layer of the graph neural network algorithm are preprocessed cell-level data, and the edges of the first layer are preprocessed neighbor-cell-level data;
[0177] When the Lth layer is the last layer of the graph neural network algorithm, a relational representation model for representing the influence relationship between cells within the area is obtained based on the nodes and edges of the Lth layer.
[0178] If the Lth layer is not the last layer of the graph neural network algorithm, update L = L + 1, and return to the steps of executing the nodes and edges of the (L-1)th layer based on the graph neural network algorithm, and calculating the nodes and edges of the Lth layer of the graph neural network algorithm, until the updated Lth layer is the last layer of the graph neural network algorithm.
[0179] In one or more embodiments, the computing module 402 is further configured to:
[0180] Based on each node of the (L-1)th layer and the linear matrix of the Lth layer of the graph neural network algorithm, the attention score representing the association relationship between any two adjacent nodes in the Lth layer is calculated.
[0181] Based on the attention score and the edge between the two adjacent nodes in the (L-1)th layer, calculate the edge between the target initial adjacent nodes corresponding to the two adjacent nodes in the Lth layer;
[0182] Using the edge between the target initial adjacent nodes in the Lth layer and the two adjacent nodes in the (L-1)th layer, calculate the target initial adjacent nodes in the Lth layer;
[0183] Obtain the cell type represented by the initial neighboring nodes of the target;
[0184] Based on the cell type represented by the target initial neighbor node, the final neighbor node of the Lth layer is determined from the target initial neighbor node and the two neighbor nodes; and the edge between the final neighbor node of the Lth layer is determined from the edge between the two neighbor nodes and the edge between the target initial neighbor node.
[0185] In one or more embodiments, the computing module 402 is further configured to:
[0186] If the cell type represented by the target initial neighbor node is a central cell, the target initial neighbor node is taken as the final neighbor node of the Lth layer.
[0187] If the cell type represented by the initial neighboring node of the target is an edge cell, the two neighboring nodes are taken as the final neighboring nodes of the Lth layer;
[0188] Determining the edges between the final adjacent nodes in the Lth layer from the edges between the two adjacent nodes and the edges between the target initial adjacent nodes includes:
[0189] When the cell type represented by the target initial neighbor node is a central cell, the edge between the target initial neighbor node is taken as the edge between the final neighbor node in the Lth layer;
[0190] When the cell type represented by the initial neighboring node of the target is an edge cell, the edge between the two neighboring nodes is taken as the edge between the final neighboring nodes of the Lth layer.
[0191] In one or more embodiments, the computing module 402 is further configured to:
[0192] Based on the number of signal handovers and coverage overlap between cells, obtain the n cells with the highest relationship to the cells represented by the initial adjacent nodes of the target.
[0193] The n cells with the highest relationship density are taken as the neighboring cells of the cell represented by the target initial neighboring node;
[0194] When all neighboring cells are within the coverage area of the area, the cell type represented by the target initial neighboring node is a central cell; the coverage area of the area is determined based on the signal coverage equipment installed within the area.
[0195] If not all neighboring cells are within the coverage area of the area, the target initial neighboring node represents the cell type as an edge cell.
[0196] In one or more embodiments, the computing module 402 is further configured to:
[0197] Based on the acquisition frequency and acquisition time range of the wireless network monitoring center, the cell-level data and the neighbor cell-level data are segmented according to a preset time length to obtain cell-level data and neighbor cell-level data in time block form;
[0198] Dimensional compression is performed on the neighboring cell-level data in the form of time blocks.
[0199] The cell relationship modeling apparatus and the cell relationship modeling method provided in this disclosure are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.
[0200] This disclosure also provides a computer device for performing the above-described cell relationship modeling method. Please refer to... Figure 5 It illustrates a schematic diagram of a computer device provided by some embodiments of this disclosure. For example... Figure 5As shown, the computer device 50 includes: a processor 500, a memory 501, a bus 502, and a communication interface 503. The processor 500, the communication interface 503, and the memory 501 are connected via the bus 502. The memory 501 stores a computer program that can run on the processor 500. When the processor 500 runs the computer program, it executes the cell relationship modeling method provided in any of the foregoing embodiments of this disclosure.
[0201] The memory 501 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this device network element and at least one other network element is achieved through at least one communication interface 503 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.
[0202] Bus 502 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. Memory 501 is used to store programs. After receiving an execution instruction, processor 500 executes the program. The cell relationship modeling method disclosed in any of the foregoing embodiments of this disclosure can be applied to processor 500, or implemented by processor 500.
[0203] The processor 500 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 500 or by instructions in software form. The processor 500 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPTA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 501. The processor 500 reads the information in memory 501 and, in conjunction with its hardware, completes the steps of the above method.
[0204] The computer device provided in this disclosure and the cell relationship modeling method provided in this disclosure are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.
[0205] This disclosure also provides a computer-readable storage medium corresponding to the cell relationship modeling method provided in the foregoing embodiments. The computer-readable storage medium is an optical disc, on which a computer program (i.e., a computer program product) is stored. When the computer program is run by a processor, it executes the cell relationship modeling method provided in any of the foregoing embodiments.
[0206] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.
[0207] The computer-readable storage medium provided in the above embodiments of this disclosure and the cell relationship modeling method provided in the embodiments of this disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.
[0208] This disclosure also provides a computer program product; please refer to [link / reference]. Figure 6 The computer program product 600 carries program code, namely computer program 601. The instructions included in the computer program 601 can be used to execute the steps of the cell relationship modeling method described in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.
[0209] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.
[0210] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.
[0211] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.
[0212] Additionally, as used herein, the “or” used in a list of items beginning with “at least one” indicates a separate list, such that a list of, for example, “at least one of A, B, or C” means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word “exemplary” does not imply that the described example is preferred or better than other examples.
[0213] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.
[0214] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.
[0215] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.
[0216] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.
Claims
1. A method for modeling community relationships, characterized in that, include: Based on the indicators representing cell status collected by the wireless network monitoring center, cell-level data and neighbor-cell-level data within the area are obtained. The cell-level data is used to represent the status of each cell in the area, and the neighbor-cell-level data is used to represent the influence relationship between each cell and its neighboring cells in the area. Based on the cell-level data and the neighbor cell-level data, a graph neural network algorithm is used to obtain a relationship representation model that characterizes the influence relationship between cells within the area.
2. The cell relationship modeling method as described in claim 1, characterized in that, Based on the cell-level data and the neighboring cell-level data, a graph neural network algorithm is used to obtain a relationship representation model for characterizing the influence relationships between cells within the area, including: The cell-level data and neighbor cell-level data are preprocessed, including data segmentation and dimensionality compression. The preprocessed cell-level data is used as nodes, and the preprocessed neighbor-cell-level data is used as edges, which are then input into a graph neural network algorithm. Based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, the nodes and edges of the Lth layer of the graph neural network algorithm are calculated, where L is a positive integer greater than or equal to 2; the nodes of the first layer of the graph neural network algorithm are preprocessed cell-level data, and the edges of the first layer are preprocessed neighbor-cell-level data; When the Lth layer is the last layer of the graph neural network algorithm, a relational representation model for representing the influence relationship between cells within the area is obtained based on the nodes and edges of the Lth layer. If the Lth layer is not the last layer of the graph neural network algorithm, update L = L + 1, and return to the steps of executing the nodes and edges of the (L-1)th layer based on the graph neural network algorithm, and calculating the nodes and edges of the Lth layer of the graph neural network algorithm, until the updated Lth layer is the last layer of the graph neural network algorithm.
3. The cell relationship modeling method as described in claim 2, characterized in that, Based on the nodes and edges of the (L-1)th layer of the graph neural network algorithm, calculate the nodes and edges of the Lth layer of the graph neural network algorithm, including: Based on each node of the (L-1)th layer and the linear matrix of the Lth layer of the graph neural network algorithm, the attention score representing the association relationship between any two adjacent nodes in the Lth layer is calculated. Based on the attention score and the edge between the two adjacent nodes in the (L-1)th layer, calculate the edge between the target initial adjacent nodes corresponding to the two adjacent nodes in the Lth layer; Using the edge between the target initial adjacent nodes in the Lth layer and the two adjacent nodes in the (L-1)th layer, calculate the target initial adjacent nodes in the Lth layer; Obtain the cell type represented by the initial neighboring nodes of the target; Based on the cell type represented by the target initial neighbor node, the final neighbor node of the Lth layer is determined from the target initial neighbor node and the two neighbor nodes; And, from the edges between the two adjacent nodes and the edges between the target initial adjacent nodes, the edges between the final adjacent nodes of the Lth layer are determined.
4. The cell relationship modeling method as described in claim 3, characterized in that, The cell types include central cells and edge cells. A central cell is a cell whose neighboring cells are all located in the area, and an edge cell is a cell whose neighboring cells are not all located in the area. Based on the cell type represented by the target initial neighbor node, the final neighbor node of the Lth layer is determined from the target initial neighbor node and the two neighbor nodes, including: If the cell type represented by the target initial neighbor node is a central cell, the target initial neighbor node is taken as the final neighbor node of the Lth layer. If the cell type represented by the initial neighboring node of the target is an edge cell, the two neighboring nodes are taken as the final neighboring nodes of the Lth layer; Determining the edges between the final adjacent nodes in the Lth layer from the edges between the two adjacent nodes and the edges between the target initial adjacent nodes includes: When the cell type represented by the target initial neighbor node is a central cell, the edge between the target initial neighbor node is taken as the edge between the final neighbor node in the Lth layer; When the cell type represented by the initial neighboring node of the target is an edge cell, the edge between the two neighboring nodes is taken as the edge between the final neighboring nodes of the Lth layer.
5. The cell relationship modeling method as described in claim 3, characterized in that, Obtaining the cell type represented by the initial neighboring nodes of the target includes: Based on the number of signal handovers and coverage overlap between cells, obtain the n cells with the highest relationship to the cells represented by the initial adjacent nodes of the target. The n cells with the highest relationship density are taken as the neighboring cells of the cell represented by the target initial neighboring node; When all neighboring cells are within the coverage area of the area, the cell type represented by the target initial neighboring node is a central cell; the coverage area of the area is determined based on the signal coverage equipment installed within the area. If not all neighboring cells are within the coverage area of the area, the target initial neighboring node represents the cell type as an edge cell.
6. The cell relationship modeling method as described in claim 2, characterized in that, The cell-level data and neighbor-cell-level data are preprocessed, including: Based on the acquisition frequency and acquisition time range of the wireless network monitoring center, the cell-level data and the neighbor cell-level data are segmented according to a preset time length to obtain cell-level data and neighbor cell-level data in time block form; Dimensional compression is performed on the neighboring cell-level data in the form of time blocks.
7. A modeling device for community relationships, characterized in that, include: The acquisition module is used to acquire cell-level data and neighbor-cell-level data within a region based on indicators representing cell status collected by the wireless network monitoring center. The cell-level data is used to represent the status of each cell included in the region, and the neighbor-cell-level data is used to represent the influence relationship between each cell and its neighboring cells within the region. The calculation module is used to obtain a relationship representation model that characterizes the influence relationship between cells within the area, based on the cell-level data and the neighbor cell-level data, using a graph neural network algorithm.
8. A computer embedded device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.
