Cell clustering method, apparatus and electronic device

Abstract

The application relates to the technical field of artificial intelligence, in particular to a cell clustering method, device and electronic equipment. The method comprises the following steps: acquiring MDT data coverage boundary maps of a plurality of cells within a first preset time and network index change data of the plurality of cells within a second preset time; performing first clustering on the cells with similar coverage form features of the MDT data coverage boundary maps and similar time distribution features of the network index change data, to obtain cells with different cluster labels; acquiring user number change data of the plurality of cells within a third preset time; and performing second clustering on the cells with similar spatiotemporal distribution features of the user number change data in the cells with different cluster labels. The cell clustering method, device and electronic equipment provided in the application can solve the technical problem that the existing cell clustering / classification method is not good in effect.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a cell clustering method, apparatus, and electronic device. Background Technology

[0002] Currently, millions of residential communities use wireless networks. Given the massive scale and complex structure of these networks, wireless network optimization is becoming increasingly sophisticated and personalized. At present, wireless network optimization is achieved through cell clustering / classification. The main cell clustering / classification methods currently available include:

[0003] 1. The coverage scenario field in the community parameters is mainly filled in manually. For example, the community is divided into "residential community", "university", "highway", etc.

[0004] 2. Cluster the communities using latitude and longitude.

[0005] 3. Cluster cells using static cross-sectional data of multiple capacity KPIs (Key Performance Indicators, such as uplink and downlink PRB utilization, traffic, etc.).

[0006] However, the above community clustering / classification methods have the following drawbacks:

[0007] 1. The "Coverage Scene" field in the cell's engineering parameters is often inaccurate, differing from the actual coverage scene of the cell, and may contain missing values. For example, the coverage scene of a cell base station "Guangzhou XX Village" located in an urban village next to a scenic area is listed as "Scenic Area," but it should actually be "Urban Village."

[0008] 2. Clustering communities based on latitude and longitude only groups geographically close communities together, considering only geographical distance. In reality, even geographically close communities may cover different scenarios. For example, a community covering residential areas and a community covering office buildings may be geographically close, but their capacity distribution over time may differ greatly.

[0009] 3. Clustering of cells using static cross-sectional data of multiple capacity KPIs (e.g., uplink / downlink PRB utilization, traffic, etc.): A k-dimensional vector representing the mean of k capacity KPIs over a certain time period is used to describe the capacity characteristics of the cell, and clustering is performed based on this vector. In reality, cell capacity KPIs often exhibit periodic patterns, and static cross-sectional data obviously cannot effectively characterize the periodic characteristics of cell capacity over time.

[0010] This means that existing community clustering / classification methods are not effective. Summary of the Invention

[0011] This application provides a cell clustering method, apparatus, and electronic device to solve the technical problem of poor performance of existing cell clustering / classification methods.

[0012] In a first aspect, embodiments of this application provide a cell clustering method, including:

[0013] Obtain the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period;

[0014] Cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features in the network indicator change data are first clustered to obtain cells with different cluster labels.

[0015] Obtain user number change data for the multiple cells within a third preset time period;

[0016] A second clustering is performed on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels.

[0017] Secondly, embodiments of this application provide a cell clustering device, comprising:

[0018] The first acquisition module is used to acquire the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period.

[0019] The first clustering module is used to perform the first clustering of cells with similar coverage morphology features of the MDT data coverage boundary map and similar temporal distribution features of the network index change data, to obtain cells with different cluster labels.

[0020] The second acquisition module is used to acquire user number change data of the multiple cells within a third preset time period;

[0021] The second clustering module is used to perform a second clustering on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels.

[0022] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the steps of the cell clustering method described in the first aspect.

[0023] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the cell clustering method described in the first aspect.

[0024] The cell clustering method, apparatus, and electronic device provided in this application perform a first clustering of cells with similar coverage morphology features of the MDT data coverage boundary map and similar temporal distribution features of network indicator change data, resulting in cells with different cluster labels. Cells with similar MDT data coverage morphology features and network indicator temporal distribution features are then assigned the same cluster label. During cell clustering, the dynamic spatial features of the MDT data coverage morphology of the cells within a first preset time period and the dynamic temporal distribution features of network indicator change data within a second preset time period are considered simultaneously. A second clustering is then performed on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels. Based on considering the MDT data coverage morphology features and the temporal distribution features of network indicators, the dynamic spatiotemporal distribution features of user number change data within a third preset time period are also considered. This application overcomes the shortcomings of existing methods, such as inaccurate engineering parameter data and the inadequacy of modeling using only spatial or temporal dimension data and static cross-sectional data. This application embodiment considers data in both time and space dimensions, and uses the characteristics of dynamic data of cells within a preset time period for clustering. It abandons static cross-sectional data and adopts a dynamic approach. The cluster labels generated by this application embodiment can fully characterize the MDT data coverage pattern of cells, the time distribution of network indicator change data, and the spatiotemporal distribution of user number change data, thereby improving the effect of cell clustering / classification methods. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is one of the flowcharts illustrating the cell clustering method provided in the embodiments of this application;

[0027] Figure 2 This is a schematic diagram of converting a scatter plot of cell MDT data into a coverage boundary map of cell MDT data according to an embodiment of this application;

[0028] Figure 3 This is the second flowchart illustrating the cell clustering method provided in the embodiments of this application;

[0029] Figure 4 This is the third flowchart illustrating the cell clustering method provided in this application embodiment;

[0030] Figure 5This is a schematic diagram illustrating the convolution operation performed on the user number spatial distribution matrix to obtain multiple third feature vectors, as provided in an embodiment of this application.

[0031] Figure 6 This is the fourth flowchart illustrating the cell clustering method provided in the embodiments of this application;

[0032] Figure 7 This is a schematic diagram of the structure of the cell clustering device provided in the embodiments of this application;

[0033] Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] Figure 1 This is a community clustering method. Please refer to [link / reference]. Figure 1 This application provides a cell clustering method, which may include:

[0036] Step 100: Obtain the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period.

[0037] The electronic device acquires the MDT data coverage boundary map of multiple cells within a first preset time period, as well as the network indicator change data of multiple cells within a second preset time period.

[0038] Here, MDT data refers to Minimization of Drive Tests. MDT data inherently includes the accurate latitude and longitude information of the user's terminal. The MDT data coverage boundary map of the cell within the first preset time period refers to the MDT data coverage boundary map generated based on the user's terminal's geographical location movement over a certain period. Specifically, the MDT data coverage boundary map can be formed from the outer envelope of the MDT data scatter points generated by the cell. Please refer to... Figure 2 , Figure 2 This diagram illustrates the conversion of a scatter plot of MDT data from a cell into a coverage boundary map of the cell's MDT data. It should be noted that the first preset time can be set according to actual conditions; for example, the first preset time could be the coverage boundary map of the cell's MDT data over a 30-day period.

[0039] Since different cells will have different MDT coverage patterns, the MDT coverage pattern of a cell can be obtained by acquiring the coverage boundary map of the cell's MDT data.

[0040] Network metric change data refers to the changes in key network metrics over time within a second preset time period. These key network metrics may include uplink and downlink PRB utilization and traffic. Cell network metric change data within the second preset time period refers to the changes in key network metrics at different time granularities within that cell during that period. It should be noted that the second preset time period can be set according to actual conditions; for example, it could be the network metric change data for each hour within a 30-day period for the cell.

[0041] The temporal distribution of network metric changes varies between different residential communities. For example, in commercial communities, network usage peaks during working hours, while in residential communities it peaks at night. By acquiring network metric change data for a community, its temporal distribution can be determined.

[0042] Step 200: Perform the first clustering on cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features in the network indicator change data to obtain cells with different cluster labels.

[0043] The electronic device performs a first clustering of cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features of network indicator change data, resulting in cells with different cluster labels. This embodiment assigns the same cluster label to cells with similar MDT data coverage morphology features and network indicator temporal distribution features; during cell clustering, it simultaneously considers the dynamic spatial features of the MDT data coverage morphology of the cells within a first preset time period, and the dynamic temporal distribution features of the network indicator change data within a second preset time period.

[0044] In one embodiment, please refer to Figure 3 Step 200: Perform a first clustering on cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features in the network indicator change data, including:

[0045] Step 210: Extract the coverage morphology features of the MDT data coverage boundary map using the first autoencoder to obtain the first feature vector.

[0046] The electronic device extracts the MDT data coverage morphological features from the MDT data coverage boundary map using a first autoencoder to obtain a first feature vector.

[0047] Specifically, step 210 involves extracting the coverage morphology features of the MDT data coverage boundary map using a first autoencoder to obtain a first feature vector, including:

[0048] Step 211: Convert the MDT data coverage boundary map into a grayscale matrix;

[0049] Step 212: Extract the coverage morphological features of the gray value matrix using the first autoencoder to obtain the first feature vector.

[0050] The electronic device converts the MDT data coverage boundary map into a grayscale matrix of size w*h (each element of the matrix is ​​a grayscale value, ranging from 0 to 255), and then uses a first autoencoder to extract the coverage morphology features of the cell MDT data coverage boundary map. After the first autoencoder flattens the grayscale matrix, the input MDT data coverage boundary map becomes a vector image_vector of length w*h. The encoder of the first autoencoder, which contains one or more fully connected layers, compresses this vector into a vector v of length k. The decoder of the first autoencoder decodes this vector and converts it back into a vector image_vector' of length d*d; thus obtaining the first feature vector.

[0051] It should be noted that during the training phase of the first autoencoder, the loss is calculated using the original input vector `image_vector` and the vector `image_vector'` reconstructed by the first autoencoder, ensuring that the image reconstructed by the first autoencoder is as similar as possible to the MDT data coverage boundary map. After the first autoencoder completes training, it is input with the MDT data coverage boundary map of a certain cell, and the first feature vector is calculated as the coverage morphology feature of the cell.

[0052] Step 220: Extract the time distribution features of the network index change data using the second autoencoder to obtain the second feature vector.

[0053] The electronic device extracts the temporal distribution features of the network indicator change data using a second autoencoder to obtain a second feature vector. The basic principle of feature extraction by the second autoencoder is described in step 210. In this embodiment, the input is the entire hourly sequence s of the cell's network indicator change data (e.g., uplink / downlink PRB utilization, traffic, etc.) over the past 30 days: [kpi_1, kpi_2, ..., kpi_n], where n = 30 * 24. kpi_n represents uplink / downlink PRB utilization or traffic, etc. The training process of the second autoencoder involves calculating the gradient of the error with respect to each parameter and adjusting the parameters to make the reconstructed sequence s' as similar as possible to the original input sequence s. After the second autoencoder completes training, the entire hourly sequence s of the cell's network indicator change data over the past 30 days is input, and the calculated second feature vector serves as the temporal distribution feature of the cell's network indicator change data.

[0054] It should be noted that the model structures of the first and second autoencoders can be the same. In other embodiments, the second autoencoder can also be replaced by a variational encoder, that is, the variational encoder is used to extract the time distribution features of the network index change data to obtain the second feature vector.

[0055] Step 230: Concatenate the first feature vector and the second feature vector, and perform the first clustering on the cells where the first feature vector and the second feature vector are similar.

[0056] Based on the obtained first and second feature vectors, the electronic device concatenates the first and second feature vectors, performing the first clustering on cells where both the first and second feature vectors are similar. For example, the MDT data coverage pattern of cell A is similar to that of cell B. Furthermore, the temporal distribution characteristics of network indicator changes in cell A are similar to those in cell B. In this case, cell A and cell B are assigned the same cluster label.

[0057] It should be noted that the embodiments of this application can use the k-means clustering algorithm to cluster similar MDT data coverage morphology features, ultimately obtaining clusters of cells with similar MDT data coverage morphology features. Additionally, the k-means algorithm can be used to cluster similar temporal distribution features of network indicator change data, ultimately obtaining clusters of cells with similar temporal distribution features of network indicator change data.

[0058] In this embodiment, cells with similar MDT data coverage patterns and network indicator temporal distribution characteristics are labeled with the same cluster label; when clustering cells, the dynamic spatial characteristics of the MDT data coverage patterns of the cells in the first preset time period and the dynamic temporal distribution characteristics of the network indicator change data in the second preset time period are considered simultaneously.

[0059] Step 300: Obtain the user number change data of the multiple cells within a third preset time period.

[0060] The electronic device acquires user number change data for the multiple cells within a third preset time period. User number change data refers to the number of terminals using the network within a cell within a certain time period, and the number of users changes dynamically within this time period. User number change data includes user number distribution data at each preset time granularity. For example, it includes hourly user number distribution data over 90 days. The user number distribution data refers to a grid divided into s*s grids within a preset range (e.g., a 1km*1km range) centered on the cell base station. The user number distribution within the grid is denoted as the s*s user number spatial distribution matrix.

[0061] Step 400: Perform a second clustering on the cells with similar spatiotemporal distribution characteristics of user number change data in the cells with different cluster labels.

[0062] The electronic device performs a second clustering on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels.

[0063] For details, please refer to Figure 4 Step 400: Perform a second clustering on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels, including:

[0064] Step 410: Convert the user number distribution data of each preset time granularity in the user number change data of cells with different cluster labels into the corresponding user number spatial distribution matrix.

[0065] Please refer to Figure 4 , Figure 4 The matrix in this example is an s*s grid within a 1km*1km area centered on the cell base station. The distribution of users within the grid is denoted as the s*s user spatial distribution matrix U.

[0066] Step 420: Perform convolution operations on each of the user number spatial distribution matrices to obtain multiple third feature vectors.

[0067] The electronic device performs convolution operations on each of the user spatial distribution matrices to obtain multiple third feature vectors. The electronic device also extracts the geographical distribution features of users within a cell using convolutional layers, resulting in multiple third feature vectors.

[0068] In one embodiment, please refer to Figure 5 Step 420: Perform convolution operations on each of the user number spatial distribution matrices to obtain multiple third feature vectors, including:

[0069] Perform the following operations on each of the aforementioned user number spatial distribution matrices:

[0070] Step 421: Perform convolution operations on the user number spatial distribution matrix using convolution kernels of different preset sizes to obtain multiple feature maps.

[0071] The electronic device performs convolution operations on the user number spatial distribution matrix using convolution kernels of different preset sizes to form different feature maps. The width w of the convolution kernel is consistent with the user number spatial distribution matrix U, both being s. kernel_size = k*s, where k = {2, 3, 4, 5}. Each size of convolution kernel slides across the user number spatial distribution matrix U with a stride of 1, forming multiple m*1 feature maps. In this embodiment, four m*1 feature maps are formed; m varies depending on k, and the specific calculation method is as follows:

[0072] m=1+(s+2*padding-dilation*(k-1)-1) / stride

[0073] Where padding = 0, dilation = 1, stride = 1.

[0074] Step 422: Perform max pooling operation on the multiple feature maps respectively to obtain multiple intermediate vectors.

[0075] Step 423: Concatenate the multiple intermediate vectors sequentially to obtain the third feature vector.

[0076] After performing max-pooling on each feature map, the electronic device generates a 1*1 intermediate vector s_1*1. All intermediate vectors are concatenated to form a third feature vector. The convolution-pooling operation in this embodiment transforms the user spatial distribution matrix U into a third feature vector x.

[0077] In this embodiment of the application, the hour is used as the statistical granularity. Within each hourly time window, the above steps 421 to 423 are executed sequentially on the spatial distribution matrix U of the number of users in the cell to form a series of third feature vectors: x_0, x_1, x_2, ..., x_t.

[0078] Step 430: Input the multiple third feature vectors into a recurrent neural network, and learn the features of the user number change data at different preset time granularities through the recurrent neural network to obtain the spatiotemporal distribution features of the user number change data of the cell.

[0079] The electronic device inputs the multiple third feature vectors into a recurrent neural network, and learns the characteristics of the user number change data at different preset time granularities through the recurrent neural network, thereby obtaining the spatiotemporal distribution characteristics of the user number change data of the cell.

[0080] Specifically, in this embodiment, the multiple third feature vectors are input into a Long Short-Term Memory (LSTM) network for sequence modeling, learning the characteristics of spatial feature changes at different times, and mapping each cell to a vector representing the spatiotemporal distribution characteristics of its user numbers.

[0081] In one embodiment, step 430 involves inputting the plurality of third feature vectors into a recurrent neural network, and learning the characteristics of the user number change data at different preset time granularities through the recurrent neural network to obtain the spatiotemporal distribution characteristics of the user number change data of the cell, including:

[0082] Step 431: Input multiple third feature vectors into the Long Short-Term Memory network;

[0083] Step 432: Control the window of the Long Short-Term Memory Network of a preset length to slide on the multiple third feature vectors, take the vectors of the first n time steps in the window as input, and take the vector of the last time step in the window as output to train the Long Short-Term Memory Network model.

[0084] Step 433: Use the hidden state of the last time step of the trained Long Short-Term Memory Network as the spatiotemporal distribution feature of the user number change data of the cell.

[0085] A window of a preset length controlling the Long Short-Term Memory (LSTM) network slides across multiple third feature vectors. The vectors from the first n time steps within the window are used as input, and the vector from the last time step is used as output. For a single cell, assuming a 90-day time range for modeling, with 24 hours per day, there are 2160 vectors, t = 2160. Assuming a window size of window_size, t - window_size + 1 training samples will be generated. The following example uses window_size = 5:

[0086] Input Intput = [x_n-4, x_n-3, x_n-2, x_n-1];

[0087] Output = x_n;

[0088] n = {window_size - 1, ..., t}.

[0089] During the model learning phase of the Long Short-Term Memory (LSTM) network, the error between the model's output and the true value, as well as the gradients of each model parameter, are calculated and optimized using gradient descent. After model training is complete, the hidden state of the last time step of the trained LSM network is used as the spatiotemporal distribution feature of the user number change data for that cell. This results in each cell being mapped to a vector representing its spatiotemporal distribution characteristics of user numbers, thus obtaining the spatiotemporal distribution feature of the user number change data for that cell.

[0090] It should be noted that in other embodiments, a recurrent neural network with a gated recurrent unit (GRU) can also be used to learn the spatiotemporal distribution characteristics of user number change data.

[0091] Step 440: Perform a second clustering on the communities with similar spatiotemporal distribution characteristics.

[0092] The electronic device performs a second clustering of cells with similar spatiotemporal distribution characteristics. Cells with similar spatiotemporal distribution characteristics of user number change data are assigned to the same cluster.

[0093] Specifically, in one embodiment, step 440, the second clustering of the cells with similar spatiotemporal distribution characteristics, specifically includes:

[0094] The spatiotemporal distribution characteristics of the multiple cells are clustered a second time using the Euclidean distance clustering method.

[0095] This application defines the distance function as Euclidean distance and uses the Euclidean distance clustering method to cluster cells. Each cell is assigned a cluster label, and the cluster label serves as the cell's category. Through the steps of this application embodiment, cells with similar spatiotemporal distribution characteristics of user number change data are assigned to the same cluster.

[0096] Therefore, considering the MDT data coverage morphology characteristics and the temporal distribution characteristics of network indicators, cells with similar MDT data coverage morphology characteristics, temporal distribution characteristics of network indicators, and spatiotemporal distribution characteristics of user number change data will be assigned the same cluster label. Based on clustering, coverage optimization, capacity prediction, and generation of dynamic thresholds for anomaly detection can be performed specifically according to the characteristics of each cluster label. This overcomes the shortcomings of existing methods that classify cells based on static cross-sectional data, geographical location, and cell engineering parameters. Furthermore, the cluster labels for cell classification generated in this application embodiment can be used as input to upper-level machine learning models, demonstrating strong practical value in multiple scenarios.

[0097] In summary, please refer to Figure 6 In the first stage, this application extracts the coverage morphology features of the cell MDT data coverage boundary map and the temporal distribution features of network indicator changes through an autoencoder. In the second stage, it combines convolutional operations and sequence modeling with a long short-term memory network to extract the dynamic spatiotemporal distribution features of user number changes in the cell, followed by clustering. This approach considers not only the coverage morphology features and the temporal distribution features of network indicators in the MDT data, but also the dynamic spatiotemporal distribution characteristics of user number changes in the cell within a third preset time period. This application overcomes the shortcomings of existing methods, such as inaccurate engineering parameter data, modeling using only spatial or temporal dimension data, and static cross-sectional data.

[0098] This application embodiment performs a first clustering of cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features of network indicator change data, resulting in cells with different cluster labels. Cells with similar MDT data coverage morphology features and network indicator temporal distribution features are then assigned the same cluster label. During cell clustering, the dynamic spatial features of the MDT data coverage morphology of the cells within a first preset time period and the dynamic temporal distribution features of network indicator change data within a second preset time period are considered simultaneously. A second clustering is then performed on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels. Based on considering the MDT data coverage morphology features and the temporal distribution features of network indicators, the dynamic spatiotemporal distribution features of user number change data within a third preset time period are also considered. This application embodiment overcomes the shortcomings of existing methods, such as inaccurate engineering parameter data, modeling using only spatial or temporal dimension data, and static cross-sectional data. This application embodiment considers data in both time and space dimensions, and uses the characteristics of dynamic data of cells within a preset time period for clustering. It abandons static cross-sectional data and adopts a dynamic approach. The cluster labels generated by this application embodiment can fully characterize the MDT data coverage pattern of cells, the time distribution of network indicator change data, and the spatiotemporal distribution of user number change data, thereby improving the effect of cell clustering / classification methods.

[0099] The cell clustering device provided in the embodiments of this application is described below. The cell clustering device described below can be referred to in correspondence with the cell clustering method described above.

[0100] Please refer to Figure 7 A community clustering device, comprising:

[0101] The first acquisition module 201 is used to acquire the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period.

[0102] The first clustering module 202 is used to perform the first clustering of cells with similar coverage morphology features of the MDT data coverage boundary map and similar temporal distribution features of the network index change data to obtain cells with different cluster labels.

[0103] The second acquisition module 203 is used to acquire user number change data of the multiple cells within a third preset time period;

[0104] The second clustering module 204 is used to perform a second clustering on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels.

[0105] By performing a first clustering of cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features of network indicator change data, cells with different cluster labels are obtained. Cells with similar MDT data coverage morphology features and network indicator temporal distribution features are then assigned the same cluster label. During cell clustering, the dynamic spatial features of the MDT data coverage morphology of the cells within a first preset time period and the dynamic temporal distribution features of network indicator change data within a second preset time period are considered simultaneously. A second clustering is then performed on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels. Based on considering the MDT data coverage morphology features and the temporal distribution features of network indicators, the dynamic spatiotemporal distribution features of user number change data within the cells are also considered within a third preset time period. This embodiment overcomes the shortcomings of existing methods, such as inaccurate engineering parameter data, modeling using only spatial or temporal dimension data, and static cross-sectional data. This application embodiment considers data in both time and space dimensions, and uses the characteristics of dynamic data of cells within a preset time period for clustering. It abandons static cross-sectional data and adopts a dynamic approach. The cluster labels generated by this application embodiment can fully characterize the MDT data coverage pattern of cells, the time distribution of network indicator change data, and the spatiotemporal distribution of user number change data, thereby improving the effect of cell clustering / classification methods.

[0106] In one embodiment, the first clustering module includes:

[0107] The coverage morphology feature extraction module is used to extract the coverage morphology features of the MDT data coverage boundary map through the first autoencoder to obtain the first feature vector.

[0108] The time distribution feature extraction module is used to extract the time distribution features of the network index change data through the second autoencoder to obtain a second feature vector.

[0109] The splicing and clustering module is used to splice the first feature vector and the second feature vector, and to perform the first clustering on the cells where the first feature vector and the second feature vector are similar.

[0110] In one embodiment, the coverage morphology feature extraction module includes:

[0111] The first conversion module is used to convert the MDT data coverage boundary map into a grayscale matrix.

[0112] The first feature vector calculation module is used to extract the coverage morphological features of the gray value matrix through the first autoencoder to obtain the first feature vector.

[0113] In one embodiment, the second clustering module includes:

[0114] The second conversion module is used to convert the user number distribution data of different time granularities in the cell with different cluster labels into the corresponding user number spatial distribution matrix.

[0115] The third feature vector calculation module is used to perform convolution operations on each of the user number spatial distribution matrices to obtain multiple third feature vectors.

[0116] The spatiotemporal distribution feature extraction module is used to input the multiple third feature vectors into a recurrent neural network, and learn the features of the user number change data at different preset time granularities through the recurrent neural network to obtain the spatiotemporal distribution features of the user number change data of the cell.

[0117] The target clustering module is used to perform a second clustering of the communities with similar spatiotemporal distribution characteristics.

[0118] In one embodiment, the third feature vector calculation module includes:

[0119] The feature map calculation module is used to perform convolution operations on the user number spatial distribution matrix using convolution kernels of different preset sizes to obtain multiple feature maps;

[0120] The intermediate vector calculation module is used to perform max pooling operations on the multiple feature maps respectively to obtain multiple intermediate vectors;

[0121] The third feature vector acquisition module is used to concatenate the multiple intermediate vectors sequentially to obtain the third feature vector.

[0122] In one embodiment, the spatiotemporal distribution feature extraction module includes:

[0123] The third feature vector input module is used to input multiple third feature vectors into the long short-term memory network.

[0124] The training module is used to control a window of a preset length of the long short-term memory network to slide on the multiple third feature vectors, and to train the long short-term memory network by taking the vectors of the first n time steps in the window as input and the vector of the last time step in the window as output.

[0125] The spatiotemporal distribution feature acquisition module is used to take the hidden state of the last time step of the trained Long Short-Term Memory Network as the spatiotemporal distribution feature of the user number change data of the cell.

[0126] In one embodiment, the target clustering module is specifically used to perform a second clustering of the spatiotemporal distribution characteristics of the multiple cells using the Euclidean distance clustering method.

[0127] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call a computer program in the memory 830 to execute the steps of a cell clustering method, such as: acquiring MDT data coverage boundary maps of multiple cells within a first preset time period, and network indicator change data of multiple cells within a second preset time period; performing a first clustering on cells with similar coverage morphology features of the MDT data coverage boundary maps and similar temporal distribution features of the network indicator change data to obtain cells with different cluster labels; acquiring user number change data of the multiple cells within a third preset time period; and performing a second clustering on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels.

[0128] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the cell clustering method provided in the above embodiments, such as: obtaining MDT data coverage boundary maps of multiple cells within a first preset time period and network indicator change data of multiple cells within a second preset time period; performing a first clustering on cells with similar coverage morphology features of the MDT data coverage boundary maps and similar temporal distribution features of the network indicator change data to obtain cells with different cluster labels; obtaining user number change data of the multiple cells within a third preset time period; and performing a second clustering on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels.

[0130] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program. The computer program is used to cause a processor to execute the steps of the cell clustering method provided in the above embodiments, such as: acquiring MDT data coverage boundary maps of multiple cells within a first preset time period, and network indicator change data of multiple cells within a second preset time period; performing a first clustering on cells with similar coverage morphology features of the MDT data coverage boundary maps and similar temporal distribution features of the network indicator change data to obtain cells with different cluster labels; acquiring user number change data of the multiple cells within a third preset time period; and performing a second clustering on cells with similar spatiotemporal distribution features of user number change data among the cells with different cluster labels.

[0131] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0132] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A cell clustering method, characterized in that, include: Obtain the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period; Cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features in the network indicator change data are first clustered to obtain cells with different cluster labels. Obtain user number change data for the multiple cells within a third preset time period; A second clustering is performed on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels; The second clustering of cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels includes: The user number distribution data of each preset time granularity in the user number change data in cells with different cluster labels are converted into the corresponding user number spatial distribution matrix; Perform convolution operations on each of the user number spatial distribution matrices to obtain multiple third feature vectors; The multiple third feature vectors are input into a recurrent neural network, and the recurrent neural network learns the features of the user number change data at different preset time granularities to obtain the spatiotemporal distribution features of the user number change data of the cell. The communities with similar spatiotemporal distribution characteristics are then clustered a second time.

2. The cell clustering method according to claim 1, characterized in that, The first clustering of cells with similar coverage morphology features in the MDT data coverage boundary map and similar temporal distribution features in the network indicator change data includes: The first feature vector is obtained by extracting the coverage morphology features of the MDT data coverage boundary map through the first autoencoder. The second feature vector is obtained by extracting the time distribution features of the network indicator change data through the second autoencoder. The first feature vector and the second feature vector are concatenated, and the cells whose first feature vector and the second feature vector are similar are subjected to the first clustering.

3. The cell clustering method according to claim 2, characterized in that, The step of extracting the coverage morphology features of the MDT data coverage boundary map through the first autoencoder to obtain the first feature vector includes: The MDT data coverage boundary map is converted into a grayscale matrix; The first feature vector is obtained by extracting the coverage morphological features of the gray value matrix through the first autoencoder.

4. The cell clustering method according to claim 1, characterized in that, The convolution operation is performed on each of the user number spatial distribution matrices to obtain multiple third feature vectors, including: Perform the following operations on each of the aforementioned user number spatial distribution matrices: Multiple feature maps are obtained by performing convolution operations on the user number spatial distribution matrix using convolution kernels of different preset sizes. Max pooling is performed on the multiple feature maps respectively to obtain multiple intermediate vectors; The third feature vector is obtained by concatenating the multiple intermediate vectors sequentially.

5. The cell clustering method according to claim 1, characterized in that, The step of inputting the multiple third feature vectors into a recurrent neural network and learning the features of the user number change data at different preset time granularities through the recurrent neural network to obtain the spatiotemporal distribution features of the user number change data of the cell includes: Input multiple third feature vectors into the long short-term memory network; A window of a preset length controlling the Long Short-Term Memory Network slides over multiple third feature vectors. The vectors of the first n time steps within the window are used as input, and the vector of the last time step within the window is used as output to train the Long Short-Term Memory Network model. The hidden state of the last time step of the trained Long Short-Term Memory network is used as the spatiotemporal distribution feature of the user number change data of the cell.

6. The cell clustering method according to claim 1, characterized in that, The second clustering of the communities with similar spatiotemporal distribution characteristics includes: The spatiotemporal distribution characteristics of the multiple cells are clustered a second time using the Euclidean distance clustering method.

7. A community clustering device, characterized in that, include: The first acquisition module is used to acquire the MDT data coverage boundary map of multiple cells within a first preset time period, and the network indicator change data of multiple cells within a second preset time period. The first clustering module is used to perform the first clustering of cells with similar coverage morphology features of the MDT data coverage boundary map and similar temporal distribution features of the network index change data, to obtain cells with different cluster labels. The second acquisition module is used to acquire user number change data of the multiple cells within a third preset time period; The second clustering module is used to perform a second clustering on cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels; The second clustering of cells with similar spatiotemporal distribution characteristics of user number change data in cells with different cluster labels includes: The user number distribution data of each preset time granularity in the user number change data in cells with different cluster labels are converted into the corresponding user number spatial distribution matrix; Perform convolution operations on each of the user number spatial distribution matrices to obtain multiple third feature vectors; The multiple third feature vectors are input into a recurrent neural network, and the recurrent neural network learns the features of the user number change data at different preset time granularities to obtain the spatiotemporal distribution features of the user number change data of the cell. The communities with similar spatiotemporal distribution characteristics are then clustered a second time.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the cell clustering method according to any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the cell clustering method according to any one of claims 1 to 6.

Images