Traffic prediction method and device, equipment, storage medium and program product

Abstract

The invention discloses a traffic prediction method and device, equipment, a storage medium and a program product. According to the specific technical scheme, the method comprises the following steps: for each base station, acquiring a crowd transfer trend and historical flow data of the base station; according to the crowd transfer trend, a crowd transfer adjacency matrix is constructed, elements of the crowd transfer adjacency matrix represent the trend that the crowd of the row base station is transferred to the column base station, the row base station is the base station corresponding to the row where the elements are located, and the column base station is the base station corresponding to the column where the elements are located; constructing a traffic transfer matrix according to the historical traffic data; calculating a sum value of the flow transfer matrix and the group transfer adjacency matrix to obtain a base station adjacency weight matrix; constructing a data matrix according to the historical traffic data; and performing traffic prediction according to the base station adjacency weight matrix and the data matrix to obtain a traffic prediction result. Thus, the traffic transfer trend between the base stations can be accurately judged by integrating the two dimensions of the traffic transfer trend and the crowd transfer trend, and the accuracy of traffic prediction is improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a method, apparatus, device, storage medium and program product for traffic prediction. Background Technology

[0002] In mobile communication networks, by predicting the traffic of base stations and then dynamically allocating network resources according to the prediction results, network congestion can be avoided and the quality of base station services can be improved.

[0003] Currently, traffic prediction is performed using algorithms such as Spatio-Temporal Variational Graph Auto-Encoder (ST-VGAE), Graph Convolutional Network (GCN), Graph Neural Network (GNN), and Spatio-Temporal Graph Convolutional Network (ST-GCN) in conjunction with a pre-defined adjacency matrix. The adjacency matrix represents the correlation between adjacent cellular network regions.

[0004] However, in practice, the flow of people changes in real time, and the flow of people causes the traffic load of different base stations to change dynamically. Therefore, a fixed adjacency matrix cannot accurately reflect the correlation between cellular network areas, thus reducing the accuracy of traffic prediction. Summary of the Invention

[0005] This application provides a method, apparatus, device, storage medium, and program product for traffic prediction, which can comprehensively determine the traffic transfer trend between base stations by considering both traffic transfer trends and population transfer trends. Finally, traffic prediction is performed based on historical traffic data and the base station adjacency weight matrix, thus improving the accuracy of traffic prediction.

[0006] In a first aspect, embodiments of this application provide a method for traffic prediction, applied to an electronic device, the method comprising: For each base station, the population transfer trend and historical traffic data of the base station are obtained. The population transfer trend is predicted based on the movement trajectory of the population. Based on the population transfer trend, a population transfer adjacency matrix is ​​constructed. The elements of the population transfer adjacency matrix represent the trend of population transfer from row base stations to column base stations. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located. A traffic transfer matrix is ​​constructed based on the historical traffic data, and the elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations; The sum of the traffic transfer matrix and the group transfer adjacency matrix is ​​calculated to obtain the base station adjacency weight matrix; A data matrix is ​​constructed based on the historical traffic data. The elements in the data matrix are used to represent the traffic data of the column base station at the row historical time. The row historical time is the historical time corresponding to the row where the element is located. Traffic prediction is performed based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result.

[0007] In one possible implementation, obtaining the population shift trend of the base station includes: Obtain historical user distribution data for each sub-period of the historical time period for the base station; For each historical user distribution data point, the historical user distribution data is clustered to obtain a cluster. Based on the cluster center of the corresponding s cluster in each sub-time period of the base station, determine the movement trajectory of each cluster center; The population transfer trend at the base station is determined based on the movement trajectory.

[0008] In one possible implementation, constructing a population transfer adjacency matrix based on the population transfer trend includes: For each base station, the weight coefficient of the base station is determined based on the user distribution data in each cluster of the base station; The weight coefficients are used as the values ​​of the elements in the group transition adjacency matrix to obtain the group transition adjacency matrix.

[0009] In one possible implementation, determining the weight coefficients of the base station based on user distribution data in each cluster of the base station includes: For each cluster, calculate the distance between the distribution data of each user in the cluster and the cluster center of the cluster; For each cluster, sum the distances corresponding to the distribution data of each user to obtain the first sum value; The average value is calculated based on the first sum value corresponding to each cluster to obtain the weight coefficient.

[0010] In one possible implementation, the step of performing traffic prediction based on the base station adjacency weight matrix and the data matrix to obtain a traffic prediction result includes: The data matrix is ​​input into the feature extraction network in the traffic prediction model. The feature extraction network extracts features from the data matrix based on the base station adjacency weight matrix to obtain encoded features. The encoded features are input into the multilayer feedforward neural network in the traffic prediction model. The mean and variance of the encoded features are calculated by the multilayer feedforward neural network, and latent variables are calculated based on the mean and variance. The latent variables are input into the reconstruction network of the traffic prediction model, and reconstruction features are constructed based on the reconstruction network. The reconstructed features are input into the hidden layer of the traffic prediction model. The hidden layer performs a weighted summation of the features of different dimensions in the reconstructed features to obtain the traffic prediction result.

[0011] In one possible implementation, before inputting the data matrix into the feature extraction network of the traffic prediction model, the method further includes: Obtain training data; The training data is input into the neural network model to obtain the sample reconstruction features output by the reconstruction network in the neural network model, the sample flow prediction results output by the neural network model, and the sample encoding features output by the feature extraction network in the neural network model. Calculate the reconstruction loss based on the sample reconstruction features and the sample encoding features; Based on the sample flow prediction results and the label data of the training data, the mean squared error loss and relative entropy loss are calculated. The total loss function is obtained by weighted summation of the reconstruction loss, the mean squared error loss, and the relative entropy loss. The neural network model is adjusted using the total loss function to obtain the traffic prediction model.

[0012] In one possible implementation, after performing traffic prediction based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result, the method further includes: For each base station, actual traffic data is obtained, which includes the number of users using preset services within the signal coverage area of ​​the base station, total user data, and actual total traffic. Calculate the correction coefficient based on the number of users, the total number of users, the actual total traffic, and the traffic prediction result; Obtain the future group transition adjacency matrix at the next time step adjacent to the traffic prediction result; The product of the correction coefficient and the future group transfer adjacency matrix is ​​calculated to obtain the corrected group transfer adjacency matrix, which is used for traffic prediction at the next time step.

[0013] Secondly, embodiments of this application provide a traffic prediction apparatus applied to an electronic device, the apparatus comprising: The acquisition module is used to acquire the population transfer trend and historical traffic data of each base station, wherein the population transfer trend is predicted based on the movement trajectory of the population. A construction module is used to construct a population transfer adjacency matrix based on the population transfer trend. The elements of the population transfer adjacency matrix represent the trend of population transfer from row base stations to column base stations. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located. The construction module is used to construct a traffic transfer matrix based on the historical traffic data, wherein the elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations; The calculation module is used to calculate the sum of the traffic transfer matrix and the group transfer adjacency matrix to obtain the base station adjacency weight matrix; The construction module is used to construct a data matrix based on the historical traffic data. The elements in the data matrix are used to represent the traffic data of the column base station at the row historical time, and the row historical time is the historical time corresponding to the row where the element is located. The prediction module is used to perform traffic prediction based on the base station adjacency weight matrix and the data matrix, and obtain the traffic prediction result.

[0014] Thirdly, embodiments of this application provide an electronic device, the device including: a processor and a memory storing computer program instructions; The method of flow prediction, as described in the first aspect, is implemented when the processor executes computer program instructions.

[0015] Fourthly, embodiments of this application provide a computer storage medium on which computer program instructions are stored, which, when executed by a processor, implement the traffic prediction method as described in the first aspect.

[0016] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the traffic prediction method as described in the first aspect.

[0017] This application discloses a method, apparatus, device, storage medium, and program product for traffic prediction. For each base station, an electronic device acquires population transfer trends and historical traffic data within the signal coverage area of ​​that base station. A traffic transfer matrix is ​​constructed based on the historical traffic data, where each element represents the trend of traffic transfer from a row base station to a column base station. The row base station is the base station corresponding to the row of the element, and the column base station is the base station corresponding to the column of the element. The traffic transfer matrix allows for the determination of traffic changes between base stations in historical timeframes. Then, a group transfer adjacency matrix is ​​constructed based on the population transfer trends, where each element represents the trend of population transfer from a row base station to a column base station. The sum of the traffic transfer matrix and the group transfer adjacency matrix is ​​calculated to obtain the base station adjacency weight matrix. Thus, the base station adjacency weight matrix is ​​constructed based on real-time collected population transfer trends and traffic transfer values. By combining these two dimensions, the traffic transfer trend and population transfer trend can be accurately determined, allowing for accurate determination of traffic transfer trends between base stations. Finally, traffic prediction is performed based on historical traffic data and the base station adjacency weight matrix, improving the accuracy of traffic prediction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a traffic prediction method provided in an embodiment of this application; Figure 2 This is an exemplary schematic diagram of a population transfer provided in an embodiment of this application; Figure 3 This is an exemplary schematic diagram illustrating a population migration trend provided in an embodiment of this application; Figure 4 This is a flowchart illustrating a method for determining population migration trends provided in an embodiment of this application; Figure 5 This is an exemplary schematic diagram of a user distribution data image provided in an embodiment of this application; Figure 6 This is an exemplary schematic diagram of head distribution data provided in an embodiment of this application; Figure 7 This is an exemplary schematic diagram of a clustering algorithm processing procedure provided in an embodiment of this application; Figure 8 This is an exemplary schematic diagram of abnormal data provided in an embodiment of this application; Figure 9This is an exemplary schematic diagram of a normal data clustering center provided in an embodiment of this application; Figure 10 This is an exemplary schematic diagram of a cluster center provided in an embodiment of this application; Figure 11 This is a flowchart illustrating a method for determining traffic prediction results provided in an embodiment of this application; Figure 12 This is a flowchart illustrating a traffic prediction model training method provided in an embodiment of this application; Figure 13 This is an exemplary schematic diagram of a traffic prediction method provided in an embodiment of this application; Figure 14 This is an exemplary schematic diagram of another traffic prediction method provided in the embodiments of this application; Figure 15 This is a schematic diagram of the structure of a flow prediction device provided in an embodiment of this application; Figure 16 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0021] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0022] To address the problems existing in the prior art, embodiments of this application provide a method, apparatus, device, storage medium, and program product for traffic prediction. The following first describes the traffic prediction method provided by embodiments of this application. This method is applied to electronic devices, such as… Figure 1 As shown, the method includes: S101. For each base station, obtain the population migration trend and historical traffic data of the base station.

[0023] The population migration trend is predicted based on the population's movement trajectory. Historical traffic data includes the base station load traffic at each historical moment.

[0024] S102. Construct a population migration adjacency matrix based on population migration trends.

[0025] In this matrix, the elements of the group transfer adjacency matrix represent the trend of the population moving from the row base station to the column base station. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located.

[0026] In one example, such as Figure 2 As shown, Figure 2 This is a schematic diagram of crowd movement provided in an embodiment of this application. At time t, the crowd is within the signal coverage area of ​​sensing base station A. At time t+1, the crowd moves to the signal coverage area of ​​sensing base station B. The group transition adjacency matrix is ​​then given by... In this matrix, the row base stations include inductive base station A and inductive base station B, and the column base stations include inductive base station A and inductive base station B. An element with a value of 1 indicates that the population at inductive base station A is migrating to inductive base station B.

[0027] In the group transition adjacency matrix is In this case, it means that the users within the signal coverage area of ​​the sensing base station B include users transferred from other sensing base stations besides sensing base station A. Therefore, the number of users within the signal coverage area of ​​sensing base station B is greater than the number of users of sensing base station A.

[0028] S103. Construct a traffic transfer matrix based on historical traffic data.

[0029] The elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations.

[0030] In one example, the flow transfer matrix can be: The row base stations include the first base station, the second base station, the third base station, and the fourth base station, and the column base stations also include the first base station, the second base station, the third base station, and the fourth base station. Taking the element in the first row and fourth column as an example, if the value of this element is 1, the corresponding row base station is the first base station, and the corresponding column base station is the fourth base station. This indicates that the traffic of the first base station is transferred to the fourth base station.

[0031] S104. Calculate the sum of the traffic transfer matrix and the group transfer adjacency matrix to obtain the base station adjacency weight matrix.

[0032] Continuing with the example above, the group transition adjacency matrix is: The base station adjacency weight matrix can be: .

[0033] like Figure 3 As shown, the base station adjacency weight matrix corresponds to the primary sensing base station, sensing base station A, sensing base station B, and sensing base station C. The arrows between the base stations indicate traffic transfer trends. For example, if sensing base station A points towards the primary sensing base station, it means that the traffic load on sensing base station A is decreasing, while the traffic load on the primary sensing base station is increasing. The numerical values ​​corresponding to the arrows represent the adjacency weights between the base stations. For example, if the value on the arrow pointing from sensing base station A to the primary sensing base station is 3, it means that traffic from sensing base station A is transferred to the primary sensing base station, resulting in a decrease in traffic from sensing base station A and an increase in traffic from the primary sensing base station. Furthermore, the larger the value, the greater the transferred traffic.

[0034] S105. Construct a data matrix based on historical traffic data.

[0035] In this data matrix, the elements represent the traffic data of the column base station at the row historical time, and the row historical time is the historical time corresponding to the row in which the element is located.

[0036] Specifically, the data matrix includes traffic data corresponding to multiple historical moments, and the traffic data includes the number of users.

[0037] In one example, the data matrix can be represented as: For example, the element in the first row and first column is... 50 represents the traffic of the column base station at time t-4, and 100 represents the user data of the column base station at time t-4.

[0038] S106. Based on the base station adjacency weight matrix and data matrix, perform traffic prediction to obtain the traffic prediction result.

[0039] Using the method provided in this application, for each base station, the electronic device acquires the population transfer trend and historical traffic data within the signal coverage area of ​​that base station. A traffic transfer matrix is ​​constructed based on the historical traffic data, where each element of the traffic transfer matrix represents the trend of traffic transfer from the row base station to the column base station. The row base station is the base station corresponding to the row of the element, and the column base station is the base station corresponding to the column of the element. The traffic transfer matrix can be used to determine the traffic changes between base stations in historical time. Then, a group transfer adjacency matrix is ​​constructed based on the population transfer trend, where each element represents the trend of population transfer from the row base station to the column base station. The sum of the traffic transfer matrix and the group transfer adjacency matrix is ​​then calculated to obtain the base station adjacency weight matrix. Thus, the base station adjacency weight matrix is ​​constructed based on the real-time collected population transfer trend and traffic transfer values. By combining the two dimensions of traffic transfer trend and population transfer trend, the traffic transfer trend between base stations can be accurately determined. Finally, traffic prediction is performed based on historical traffic data and the base station adjacency weight matrix, improving the accuracy of traffic prediction.

[0040] Regarding S102 above, based on the population migration trend, a population migration adjacency matrix is ​​constructed, which can be specifically implemented as S1021-S1024, such as... Figure 4 As shown: S1021. Obtain historical user distribution data for each sub-period of the historical time period for the base station.

[0041] The historical time period includes multiple sub-time periods. For example, the sub-time periods include time period t-1, time period t-2, and time period t-3.

[0042] Historical user distribution data includes two types of data collected in real time by the sensing base station: image data and sensing data on the distance, speed, and altitude of user equipment.

[0043] like Figure 5 As shown, Figure 5 An example is shown of image data collected by a sensing base station. Using this image data, a head recognition algorithm is employed to extract the distribution of heads within the image data, thereby determining the distribution of electronic devices. For example... Figure 6 As shown, Figure 6 An exemplary diagram illustrates the distribution of heads. Each person is represented by a data point, thus obtaining the head distribution data.

[0044] S1022. For each historical user distribution data, perform clustering processing on the historical user distribution data to obtain clusters.

[0045] Historical user distribution data is represented by head distribution data. Based on the extracted head distribution data, the head distribution data is clustered to obtain multiple clusters.

[0046] Specifically, such as Figure 7 As shown, Figure 7 An example is provided illustrating the implementation process of a clustering algorithm. Among them, Figure 7 The diagram includes Figures (a) to (f). Figure (a) shows the extracted head distribution data, and the two crosses in Figure (b) represent the initial cluster centers. The process begins by obtaining the first data point from the head distribution data, calculating the distance between this data point and each initial cluster center, and adding this data point to the cluster of the nearest cluster center. Then, for each cluster, the mean of the existing data points in that cluster is calculated, and this mean is used as the updated cluster center. The process is repeated for the second data point, iterating through the cluster centers until all head distribution data has been traversed. This process yields the two clusters shown in Figure (f).

[0047] The method for calculating the cluster center location is as follows: First, calculate the distance between the head distribution data in the cluster and the original cluster center. Then, calculate the average distance for each head distribution data to obtain the updated cluster center location.

[0048] The distance between the head distribution data and the original cluster centers is calculated using the following formula: .

[0049] in, This indicates the number of heads distributed in the cluster. These represent the longitude and latitude positions of the head distribution data, respectively.

[0050] Then, the updated cluster center positions are calculated using the following formula: .

[0051] in, This represents the cluster center of the k-th cluster. This represents the head distribution data of the i-th person in the k-th cluster. This indicates the number of clusters.

[0052] S1023. Determine the movement trajectory of each cluster center based on the cluster center of the corresponding cluster in each sub-time period of the base station.

[0053] Specifically, for each base station, the electronic device can determine the clusters of each sub-time period within the historical time period, and then construct the movement trajectory of the cluster centers by comparing the positions of the cluster centers of adjacent sub-time periods.

[0054] Furthermore, in the aforementioned clustering process, after calculating the distance between each head distribution data point and each cluster center, outlier data points are filtered based on the distances corresponding to each head distribution data point. For each head distribution data point, if the distance between the head distribution data point and each cluster center is greater than or equal to a preset distance threshold, that head distribution data point is considered an outlier. Specifically, as follows... Figure 8 As shown, Figure 8 An exemplary diagram illustrates a normal data point set and an abnormal data point set. The normal data point set includes a first head distribution data, which is data whose distance to at least one cluster center is less than a preset distance threshold. The abnormal data point set includes a second head distribution data, which is data whose distance to all cluster centers is greater than or equal to a preset distance threshold.

[0055] exist Figure 8 Based on the illustrated embodiment, the centroid position of the normal data point set is calculated. For example... Figure 9 As shown, Figure 9 An example is shown of the centroid of a normal data point aggregation.

[0056] In practical implementation, the number of clusters within the signal coverage area of ​​a sensing base station can be multiple, such as... Figure 10 As shown, Figure 10 Five clusters of the sensing base station are illustrated as an example.

[0057] S1024. Determine the population transfer trend of the base station based on the movement trajectory.

[0058] Understandably, after determining the movement trajectory of the cluster centers, the movement direction and speed of the cluster centers in the latest sub-time period are determined according to the movement trajectory. Specifically, based on obtaining the movement trajectory, a tangent line to the movement trajectory is established at the position of the cluster centers in the latest sub-time period, and the direction of the tangent line is taken as the movement direction of the cluster centers.

[0059] The method provided in this application's embodiments involves clustering historical user distribution data for each sub-time period to obtain multiple clusters. By calculating the movement trajectories of the cluster centers, the population migration trends within the base station signal coverage area can be determined. Thus, clustering allows for the quantification of user movement trajectories, reducing computational load and improving computational efficiency.

[0060] Based on the above-mentioned determination of population migration trends through clustering algorithms, step S102, which constructs a population migration adjacency matrix according to the population migration trends, can be implemented as follows: For each base station, the weight coefficient of the base station is determined based on the user distribution data in each cluster of the base station; the weight coefficient is used as the value of the element in the group transition adjacency matrix to obtain the group transition adjacency matrix.

[0061] To determine the weighting coefficients of base stations based on user distribution data in each cluster, the specific implementation can be as follows: For each cluster, calculate the distance between each user's data distribution within the cluster and the cluster center. For each cluster, sum the distances corresponding to each user's data distribution to obtain a first sum. Based on the first sum for each cluster, calculate the average to obtain the weight coefficient.

[0062] Using the method provided in this application, the number of users within the current base station signal coverage area can be determined based on the user distribution data of clusters. The more users there are, the higher the corresponding base station load, and thus the higher the base station's weight coefficient. The weight coefficient of the base station is determined based on the user distribution data. This weight coefficient serves as an element value in the group transition adjacency matrix, thereby constructing the group transition adjacency matrix and improving its accuracy. Furthermore, the distance between the user distribution data and the cluster centers is used, and the distances corresponding to each user distribution data point are summed to obtain a first sum. The average value of the first sum for each cluster is then calculated. In this way, the traffic load of each base station is quantified through clustering, improving the accuracy of the calculation results.

[0063] In some embodiments of this application, traffic prediction is performed based on the base station adjacency weight matrix and data matrix in step S106 to obtain the traffic prediction result. Specifically, this can be implemented as steps S1061-S1064, such as... Figure 11 As shown: S1061. Input the data matrix into the feature extraction network in the traffic prediction model. The feature extraction network extracts features from the data matrix based on the base station adjacency weight matrix to obtain the encoded features.

[0064] Before using the base station adjacency weight matrix for feature extraction, the electronic device processes the base station adjacency weight matrix to generate an adjacency matrix, and then uses the adjacency matrix for feature extraction.

[0065] Specifically, the electronic device calculates the degree matrix of the base station adjacency weight matrix, and then normalizes the base station adjacency weight matrix using the degree matrix to obtain the adjacency matrix.

[0066] In one example, the base station adjacency weight matrix is: The calculated degree matrix is ​​as follows: The method for calculating the adjacency matrix is ​​as follows: .

[0067] in, This represents the adjacency matrix.

[0068] Based on the calculated adjacency matrix, the feature extraction network uses the adjacency matrix to perform spatial convolution on the data matrix to obtain spatial convolution features. Then, it uses a convolution kernel of a preset size to extract temporal features from the spatial convolution features, thereby obtaining encoded features.

[0069] In one example, the data matrix is: The adjacency matrix is: Taking the data from base station 0 at time t as an example, the spatial convolution result is: .

[0070] Then, the feature extraction network uses a convolutional kernel of size 3 to extract temporal features from the spatial features. The temporal features are: That is, the time feature is represented as a 32-dimensional matrix with three rows and four columns.

[0071] S1062. Input the encoded features into the multilayer feedforward neural network in the traffic prediction model, calculate the mean and variance of the encoded features through the multilayer feedforward neural network, and calculate the latent variables based on the mean and variance.

[0072] In this process, the multilayer feedforward neural network first flattens the encoded features to obtain row vectors, and then calculates the mean and variance of the row vectors.

[0073] Specifically, the multilayer feedforward neural network calculates the product of the preset parameters and the variance to obtain the first product, and then calculates the product of the first product and the mean to obtain the latent variables.

[0074] In one example, a multilayer feedforward neural network pair Flattened into row vectors Then the mean was calculated to be... The variance is The latent variables obtained are .in, They follow a standard normal distribution with a mean of 0 and a covariance matrix equal to the identity matrix. , It is an identity matrix.

[0075] S1063. Input the latent variables into the reconstruction network in the traffic prediction model, and construct reconstruction features based on the reconstruction network.

[0076] In this process, after receiving latent variables, the reconstruction network determines whether the latent variables meet a preset size. If the latent variables do not meet the preset size, they are expanded to the preset size. If the latent variables are smaller than the preset size, they are copied to obtain expanded variables.

[0077] Then, the reconstructed network uses an adjacency matrix to aggregate the neighbor information of the extended variables, thereby obtaining the reconstructed features.

[0078] In one example, taking the coding features of base station 1 as an example, the expanded variable is: Then, the neighbor information is aggregated using the adjacency matrix. .

[0079] in, This represents the activation function. This represents the adjacency matrix.

[0080] S1064. Input the reconstructed features into the hidden layer of the traffic prediction model. The hidden layer performs a weighted summation of the features of different dimensions in the reconstructed features to obtain the traffic prediction result.

[0081] In this approach, different weights are pre-set for each dimension of the data matrix. For example, if the data matrix includes two dimensions: user count and traffic, the weight for user count is 0.7, and the weight for traffic is 0.3. The reconstructed features are... The hidden layer then calculates the traffic prediction result as follows: Where 0.1 represents the bias term of the hidden layer. For each feature, the traffic prediction result is calculated using the above method, and the traffic prediction result is... The weight of user numbers is greater than that of traffic volume, thus strengthening the dominant influence of population flow dynamics on traffic prediction.

[0082] It should be noted that after obtaining the traffic prediction result through the above calculation, the traffic prediction result is inversely normalized. That is, the product of the traffic prediction result and the preset scaling factor is calculated to obtain the actual traffic prediction result.

[0083] In one example With a preset scaling factor of 136, the calculated actual traffic prediction result is as follows: .

[0084] Based on the above actual traffic prediction results, the main sensing base station is determined to be in The flow rate at time t shows an upward trend compared to the flow rate at time t, and this trend continues... and The continuous intensification of this phenomenon confirms a significant correlation between the dynamic evolution of the adjacency matrix caused by user scale expansion and traffic prediction results. Specifically, The predicted values ​​at any given time are highly valuable because they directly reflect the current user growth trend; however, , The flow prediction results at any given time are due to the use of The adjacency matrix prediction at any given time leads to a systematic overestimation of traffic growth, thus weakening its reference value. Therefore, in The adjacency matrix is ​​reconstructed based on user migration trends and the number of users within the signal range of each base station, and more timely traffic prediction results are obtained through a new round of iterative calculations.

[0085] The method provided in this application involves inputting a data matrix into a traffic prediction model. The feature extraction network within the model extracts features from the data matrix based on the base station adjacency weight matrix to obtain encoded features. This feature extraction, using the base station adjacency weight matrix, integrates traffic transfer trends between base stations and population transfer trends, improving the accuracy of the encoded features. Then, a multi-layer feedforward neural network is used to calculate the mean and variance of the encoded features, and the variance and mean are used to calculate latent variables. A reconstruction network is then used to reconstruct the latent network, obtaining reconstructed features. Based on these reconstructed features, a traffic prediction result can be obtained. Because the feature extraction process integrates traffic transfer trends between base stations and population transfer trends, the accuracy of the encoded features is improved, further enhancing the accuracy of the traffic prediction result.

[0086] Furthermore, after the aforementioned electronic devices calculate the traffic prediction results, they connect the traffic prediction results to the 5G-A network digital twin platform to simulate the congestion risks under different scheduling strategies. This allows for the pre-allocation of sufficient channel resources to each base station to handle the traffic growth caused by user migration, thus avoiding network congestion through proactive capacity expansion.

[0087] The following combination Figure 12 The training process of the above traffic prediction model is introduced, such as... Figure 12 As shown, the method includes: S1201. Obtain training data.

[0088] S1202. Input the training data into the neural network model to obtain the sample reconstruction features output by the reconstruction network in the neural network model, the sample flow prediction results output by the neural network model, and the sample encoding features output by the feature extraction network in the neural network model.

[0089] S1203. Calculate the reconstruction loss based on the sample reconstruction features and sample coding features.

[0090] S1204. Based on the sample flow prediction results and the label data of the training data, the mean squared error loss and relative entropy loss are calculated.

[0091] S1205. The reconstruction loss, mean squared error loss, and relative entropy loss are weighted and summed to obtain the total loss function.

[0092] Specifically, the total loss function is calculated using the following formula: .

[0093] in, Represents the total loss function. Indicates the reconstruction loss. This represents the relative entropy loss. This represents the mean squared error loss. This is a preset hyperparameter; for example, the preset hyperparameter can be 0.1. This indicates the sample flow prediction results. This represents the label data.

[0094] S1206. Adjust the neural network model using the total loss function to obtain the traffic prediction model.

[0095] Using the method provided in this application, training data is input into a neural network model for training, resulting in sample reconstruction features output by the reconstruction network, sample encoding features output by the feature extraction network, and sample traffic prediction results output by the neural network model. Based on these three parameters and the label data of the training data, reconstruction loss, relative entropy loss, and mean squared error loss are calculated respectively. A weighted sum of these losses yields the total loss function. By decomposing the total loss function into reconstruction loss, relative entropy loss, and mean squared error loss, precise training of each network layer in the neural network model can be achieved, thereby improving the accuracy of model training and enhancing model performance.

[0096] In some embodiments of this application, after performing traffic prediction based on the base station adjacency weight matrix and the data matrix in S106 above to obtain the traffic prediction result, the method further includes: Step 1: For each base station, obtain actual traffic data, which includes the number of users using the preset service within the base station's signal coverage area, total user data, and actual total traffic.

[0097] Step 2: Calculate the correction factor based on the number of users, total number of users, actual total traffic, and traffic prediction results.

[0098] The specific calculation method for the correction factor is as follows: The electronic device first calculates the ratio of the number of users to the total number of users to obtain the user ratio; then it calculates the difference between the actual total traffic and the traffic prediction result to obtain the first difference; it calculates the ratio of the first difference to the traffic prediction result to obtain the traffic service behavior ratio; finally, it performs a weighted sum of the user ratio and the traffic service behavior ratio to obtain the correction coefficient.

[0099] Specifically, the formula for calculating the correction factor is as follows: .

[0100] in, This represents the correction factor. The intercept term is used to represent the value at... and When it is 0, The value of . Indicates the proportion of users. Indicates the proportion of traffic-related business activities. and These are regression coefficients, used to represent the effect of a 1-unit increase in the user ratio and traffic / business behavior ratio on [the outcome of the regression]. The marginal effect The random error term follows a normal distribution. N(0, )。

[0101] Step 3: Obtain the future group transition adjacency matrix for the next time step adjacent to the flow prediction result.

[0102] Step 4: Calculate the product of the correction coefficient and the future group transfer adjacency matrix to obtain the corrected group transfer adjacency matrix. The corrected group transfer adjacency matrix is ​​used for flow prediction at the next time step.

[0103] The method provided in this application, after obtaining the traffic prediction result, acquires the actual traffic data corresponding to each base station. Then, based on the number of users using the preset service, the total number of users, and the actual total traffic in the actual traffic data, a correction coefficient is calculated. This correction coefficient is then used to correct the group transfer adjacency matrix for the next time step, resulting in a more accurate future group transfer adjacency matrix. In this way, the group transfer adjacency matrix is ​​adjusted in real time, thereby improving the accuracy of the traffic prediction result for the next time step.

[0104] Figure 13 An example is shown of a method for traffic prediction, such as Figure 13 As shown, the method consists of three parts: dual-source data input, dynamic traffic prediction, and prediction result output.

[0105] The dual-source data input includes inputting historical traffic data to determine long-term time-series patterns, and using a 5G-A integrated sensing base station to perform real-time analysis of the pedestrian flow heatmap based on edge computing power. The pedestrian flow heatmap is used to represent user distribution data.

[0106] After obtaining the temporal patterns of historical traffic data and the pedestrian flow heatmap, data processing and feature extraction are performed on these two types of data. First, data cleaning is conducted, followed by group transfer quantization. Specifically, quantization is achieved using three algorithms: generating movement vectors, calculating cluster centroids, and a head detection algorithm.

[0107] Data cleaning refers to removing abnormal sample data such as negative traffic and data exceeding the theoretical bandwidth limit, and verifying outliers in the original data one by one, and deleting data that does not conform to business logic or physical constraints.

[0108] Furthermore, depending on the specific business scenario, traffic anomalies caused by special events are further filtered out. For example, special events might include base station alarms or cell alarms. Specifically, if abnormally low traffic values ​​are indeed present during the alarm period, the alarm period data is corrected using interpolation or replacement methods based on the historical average traffic during the same period without alarms. Alternatively, statistical methods can be used to identify and remove outlier data points.

[0109] Group transition quantization refers to the process of constructing a group transition adjacency matrix.

[0110] Historical traffic weights are obtained from historical traffic data. Then, population migration trends are determined from the actual number of users collected. Adjacency weights are fused from the historical traffic weights and population migration trends to obtain the adjacency weights. The historical traffic weights are represented by the traffic transfer matrix in the above embodiments, and the population migration trends are represented by the group transfer adjacency matrix in the above embodiments. The adjacency weights are represented by the base station adjacency weight matrix in the above embodiments.

[0111] Electronic devices input adjacency weights and group transfer adjacency matrices into the ST-VGAE traffic prediction model to obtain traffic prediction results, i.e., the predicted value at time t+1. The predicted value at time t+1 is then input into the digital twin platform. The digital twin platform simulates the congestion risk under different scheduling strategies and expands the channel resources of each base station in advance to accommodate the traffic growth caused by the migration of user groups, thus avoiding network congestion through proactive expansion.

[0112] In addition, after obtaining the traffic prediction results, the electronic device obtains the actual traffic data at time t+1, and then performs user group image separation and business behavior analysis based on the actual traffic data. By comparing with the actual number of users, the user ratio and traffic business behavior ratio are determined, and then the correction coefficient is calculated. The adjacency weight at time t+1 is corrected using the correction coefficient, thereby improving the accuracy of the adjacency weight at time t+1.

[0113] In the above Figure 13 Based on the illustrated embodiments, combined with Figure 14This application introduces a traffic prediction method provided in its embodiments, such as... Figure 14 As shown: The traffic prediction method includes obtaining historical base station traffic, inputting the historical base station traffic into the ST-VGAE prediction model, the prediction model outputting the predicted base station traffic, adjusting the base station channel resources through load balancing, and finally obtaining the real-time base station traffic to correct the base station adjacency weight matrix.

[0114] In this embodiment, the electronic device calculates the weights of sensing base station A and sensing base station B based on the crowd transfer trend between sensing base station A and sensing base station B. These weights are then added to the ST-VGAE prediction model. By combining historical traffic and crowd migration trends, the base station adjacency weight matrix in the traffic prediction process can be dynamically adjusted according to changes in pedestrian flow, thereby improving the accuracy of the traffic prediction results.

[0115] Based on the same concept, embodiments of this application provide a traffic flow prediction apparatus, such as... Figure 15 As shown, the device includes: The acquisition module 1501 is used to acquire the population transfer trend and historical traffic data of each base station, wherein the population transfer trend is predicted based on the movement trajectory of the population. The construction module 1502 is used to construct a population transfer adjacency matrix based on the population transfer trend. The elements of the population transfer adjacency matrix represent the trend of population transfer from row base stations to column base stations. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located. The construction module 1502 is used to construct a traffic transfer matrix based on the historical traffic data, wherein the elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations; The calculation module 1503 is used to calculate the sum of the traffic transfer matrix and the group transfer adjacency matrix to obtain the base station adjacency weight matrix; The construction module 1502 is used to construct a data matrix based on the historical traffic data. The elements in the data matrix are used to represent the traffic data of the column base station at the row historical time. The row historical time is the historical time corresponding to the row where the element is located. The prediction module 1504 is used to perform traffic prediction based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result.

[0116] In one possible implementation, module 1501 is specifically used for: Obtain historical user distribution data for each sub-period of the historical time period for the base station; For each historical user distribution data point, the historical user distribution data is clustered to obtain a cluster. Based on the cluster centers of the corresponding clusters of the base stations in each sub-time period, the movement trajectory of each cluster center is determined; The population transfer trend at the base station is determined based on the movement trajectory.

[0117] In one possible implementation, module 1502 is specifically used for: For each base station, the weight coefficient of the base station is determined based on the user distribution data in each cluster of the base station; The weight coefficients are used as the values ​​of the elements in the group transition adjacency matrix to obtain the group transition adjacency matrix.

[0118] In one possible implementation, module 1502 is specifically used for: For each cluster, calculate the distance between the distribution data of each user in the cluster and the cluster center of the cluster; For each cluster, sum the distances corresponding to the distribution data of each user to obtain the first sum value; The average value is calculated based on the first sum value corresponding to each cluster to obtain the weight coefficient.

[0119] In one possible implementation, the prediction module 1504 is specifically used for: The data matrix is ​​input into the feature extraction network in the traffic prediction model. The feature extraction network extracts features from the data matrix based on the base station adjacency weight matrix to obtain encoded features. The encoded features are input into the multilayer feedforward neural network in the traffic prediction model. The mean and variance of the encoded features are calculated by the multilayer feedforward neural network, and latent variables are calculated based on the mean and variance. The latent variables are input into the reconstruction network of the traffic prediction model, and reconstruction features are constructed based on the reconstruction network. The reconstructed features are input into the hidden layer of the traffic prediction model. The hidden layer performs a weighted summation of the features of different dimensions in the reconstructed features to obtain the traffic prediction result.

[0120] In one possible implementation, before inputting the data matrix into the feature extraction network of the traffic prediction model, the apparatus further includes: Module 1501 is also used to acquire training data; The training module is used to input the training data into the neural network model and obtain the sample reconstruction features output by the reconstruction network in the neural network model, the sample traffic prediction results output by the neural network model, and the sample encoding features output by the feature extraction network in the neural network model. The calculation module 1503 is further configured to calculate the reconstruction loss based on the sample reconstruction features and the sample encoding features; The calculation module 1503 is also used to calculate the mean squared error loss and the relative entropy loss based on the sample flow prediction result and the label data of the training data. The calculation module 1503 is also used to perform a weighted summation of the reconstruction loss, the mean square error loss and the relative entropy loss to obtain the total loss function; The adjustment module is also used to adjust the neural network model using the total loss function to obtain the traffic prediction model.

[0121] In one possible implementation, the acquisition module 1501 is further configured to acquire actual traffic data for each base station, the actual traffic data including the number of users using preset services within the signal coverage area of ​​the base station, total user data, and actual total traffic. The calculation module 1503 is further configured to calculate a correction coefficient based on the number of users, the total number of users, the actual total traffic, and the traffic prediction result; The acquisition module 1501 is also used to acquire the future group transfer adjacency matrix of the next time step adjacent to the traffic prediction result; The calculation module 1503 is also used to calculate the product of the correction coefficient and the future group transfer adjacency matrix to obtain the corrected group transfer adjacency matrix, which is used for traffic prediction at the next time step.

[0122] It should be noted that the traffic prediction device is the same as the traffic prediction method described above. All implementation methods in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effect.

[0123] Figure 16 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0124] The electronic device may include a processor 1601 and a memory 1602 storing computer program instructions.

[0125] Specifically, the processor 1601 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0126] Memory 1602 may include mass storage for data or instructions. For example, and not limitingly, memory 1602 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1602 may include removable or non-removable (or fixed) media. Where appropriate, memory 1602 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1602 is non-volatile solid-state memory.

[0127] In a particular embodiment, memory 1602 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0128] The processor 1601 implements any of the traffic prediction methods described in the above embodiments by reading and executing computer program instructions stored in the memory 1602.

[0129] In one example, the electronic device may also include a communication interface 1603 and a bus 1604. For example, Figure 16 As shown, the processor 1601, memory 1602, and communication interface 1603 are connected through bus 1604 and complete communication with each other.

[0130] The communication interface 1603 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0131] Bus 1604 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Super Transmission (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 1604 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0132] Furthermore, in conjunction with the traffic prediction methods in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the traffic prediction methods in the above embodiments.

[0133] This application also provides a computer program product, including a computer program that, when executed, implements any of the traffic prediction methods described in the above embodiments.

[0134] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0135] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on machine-readable media or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0136] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0137] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0138] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for traffic flow prediction, characterized in that, Applied to electronic devices, the method includes: For each base station, the population transfer trend and historical traffic data of the base station are obtained. The population transfer trend is predicted based on the movement trajectory of the population. Based on the population transfer trend, a population transfer adjacency matrix is ​​constructed. The elements of the population transfer adjacency matrix represent the trend of population transfer from row base stations to column base stations. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located. A traffic transfer matrix is ​​constructed based on the historical traffic data, and the elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations; The sum of the traffic transfer matrix and the group transfer adjacency matrix is ​​calculated to obtain the base station adjacency weight matrix; A data matrix is ​​constructed based on the historical traffic data. The elements in the data matrix are used to represent the traffic data of the column base station at the row historical time. The row historical time is the historical time corresponding to the row where the element is located. Traffic prediction is performed based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result.

2. The method according to claim 1, characterized in that, The acquisition of the population migration trend of the base station includes: Obtain historical user distribution data for each sub-period of the historical time period for the base station; For each historical user distribution data point, the historical user distribution data is clustered to obtain a cluster. Based on the cluster centers of the corresponding clusters of the base stations in each sub-time period, the movement trajectory of each cluster center is determined; The population transfer trend at the base station is determined based on the movement trajectory.

3. The method according to claim 2, characterized in that, The step of constructing a population migration adjacency matrix based on the population migration trend includes: For each base station, the weight coefficient of the base station is determined based on the user distribution data in each cluster of the base station; The weight coefficients are used as the values ​​of the elements in the group transition adjacency matrix to obtain the group transition adjacency matrix.

4. The method according to claim 3, characterized in that, The step of determining the weight coefficient of the base station based on the user distribution data in each cluster of the base station includes: For each cluster, calculate the distance between the distribution data of each user in the cluster and the cluster center of the cluster; For each cluster, sum the distances corresponding to the distribution data of each user to obtain the first sum value; The average value is calculated based on the first sum value corresponding to each cluster to obtain the weight coefficient.

5. The method according to claim 1, characterized in that, The step of performing traffic prediction based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result includes: The data matrix is ​​input into the feature extraction network in the traffic prediction model. The feature extraction network extracts features from the data matrix based on the base station adjacency weight matrix to obtain encoded features. The encoded features are input into the multilayer feedforward neural network in the traffic prediction model. The mean and variance of the encoded features are calculated by the multilayer feedforward neural network, and latent variables are calculated based on the mean and variance. The latent variables are input into the reconstruction network of the traffic prediction model, and reconstruction features are constructed based on the reconstruction network. The reconstructed features are input into the hidden layer of the traffic prediction model. The hidden layer performs a weighted summation of the features of different dimensions in the reconstructed features to obtain the traffic prediction result.

6. The method according to claim 5, characterized in that, Before inputting the data matrix into the feature extraction network of the traffic prediction model, the method further includes: Obtain training data; The training data is input into the neural network model to obtain the sample reconstruction features output by the reconstruction network in the neural network model, the sample flow prediction results output by the neural network model, and the sample encoding features output by the feature extraction network in the neural network model. Calculate the reconstruction loss based on the sample reconstruction features and the sample encoding features; Based on the sample flow prediction results and the label data of the training data, the mean squared error loss and relative entropy loss are calculated. The total loss function is obtained by weighted summation of the reconstruction loss, the mean squared error loss, and the relative entropy loss. The neural network model is adjusted using the total loss function to obtain the traffic prediction model.

7. The method according to claim 1, characterized in that, After performing traffic prediction based on the base station adjacency weight matrix and the data matrix to obtain the traffic prediction result, the method further includes: For each base station, actual traffic data is obtained, which includes the number of users using preset services within the signal coverage area of ​​the base station, the total number of users, and the actual total traffic. Calculate the correction coefficient based on the number of users, the total number of users, the actual total traffic, and the traffic prediction result; Obtain the future group transition adjacency matrix at the next time step adjacent to the traffic prediction result; The product of the correction coefficient and the future group transfer adjacency matrix is ​​calculated to obtain the corrected group transfer adjacency matrix, which is used for traffic prediction at the next time step.

8. A device for flow prediction, characterized in that, Applied to electronic devices, the device includes: The acquisition module is used to acquire the population transfer trend and historical traffic data of each base station, wherein the population transfer trend is predicted based on the movement trajectory of the population. A construction module is used to construct a population transfer adjacency matrix based on the population transfer trend. The elements of the population transfer adjacency matrix represent the trend of population transfer from row base stations to column base stations. The row base station is the base station corresponding to the row where the element is located, and the column base station is the base station corresponding to the column where the element is located. The construction module is used to construct a traffic transfer matrix based on the historical traffic data, wherein the elements of the traffic transfer matrix represent the trend of traffic transfer from row base stations to column base stations; The calculation module is used to calculate the sum of the traffic transfer matrix and the group transfer adjacency matrix to obtain the base station adjacency weight matrix; The construction module is used to construct a data matrix based on the historical traffic data. The elements in the data matrix are used to represent the traffic data of the column base station at the row historical time, and the row historical time is the historical time corresponding to the row where the element is located. The prediction module is used to perform traffic prediction based on the base station adjacency weight matrix and the data matrix, and obtain the traffic prediction result.

9. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing computer program instructions; The processor executes computer program instructions to implement the flow prediction method as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the traffic prediction method as described in any one of claims 1-7.

11. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the traffic prediction method as described in any one of claims 1-7.

Images