Urban mobile network traffic generation method and device, electronic equipment and storage medium

Abstract

The application provides a kind of urban mobile network traffic generation method, device, electronic equipment and storage medium, the method comprises: extracting multiple entities based on city data information, and constructing city knowledge graph based on the relationship between entities;Knowledge graph embedding vector corresponding to base station is obtained based on city knowledge graph;Knowledge graph embedding vector and noise are input to the generator network of the generated adversarial network model trained to obtain generated traffic data;Different time scale urban mobile network traffic pattern generation adversarial network and continuous time urban mobile network traffic generation adversarial network in the generated adversarial network model are all input to the generator network based on the knowledge graph embedding vector of multiple base stations transmitting known network traffic to obtain generated traffic sample data, then the real traffic data and generated traffic sample data are discriminated based on discriminator network, and are trained in opposition.By the application, reliable urban mobile network traffic can be generated.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, apparatus, electronic device, and storage medium for generating urban mobile network traffic. Background Technology

[0002] Urban mobile network traffic generation refers to the generation of mobile network traffic within a city. In cities, the number of mobile devices is increasing daily, including but not limited to smartphones and various IoT devices. These devices access the mobile network through nearby base stations and generate diverse mobile network traffic when performing their functions. This mobile network traffic reflects the activity of network domains in multiple areas and time periods within the city, and is of great significance for the planning, construction, and optimization of urban mobile networks.

[0003] However, current urban mobile network traffic generation methods primarily generate daily traffic fluctuations without fully considering the impact of the urban environment. They also fail to generate traffic data with longer time scales and finer spatial granularity, resulting in insufficient practicality and accuracy. Therefore, generating large-scale, accurate, and reliable urban mobile network traffic data has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for generating urban mobile network traffic, which addresses the shortcomings of existing technologies in terms of the practicality and accuracy of urban mobile network traffic generation, and enables the generation of reliable urban mobile network traffic.

[0005] This invention provides a method for generating urban mobile network traffic. The method includes: extracting multiple entities based on urban data information and constructing an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; obtaining knowledge graph embedding vectors corresponding to the base stations based on the urban knowledge graph; inputting the knowledge graph embedding vectors and noise into a generator network of a trained generative adversarial network (GAN) model to obtain generated traffic data output by the generator network; the generated traffic data is urban mobile network traffic in the context of the urban data information; wherein the GAN model includes urban mobile network traffic pattern GANs at different time scales and continuous-time urban mobile network traffic GANs, both of which are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic being input into the generator network to obtain generated traffic sample data, and then performing adversarial training on real traffic data and the generated traffic sample data based on a discriminator network, wherein the real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

[0006] According to a method for generating urban mobile network traffic provided by the present invention, the step of obtaining the knowledge graph embedding vector corresponding to the base station based on the urban knowledge graph specifically includes: obtaining the knowledge graph embedding vector corresponding to the base station through a knowledge graph embedding model based on the urban knowledge graph.

[0007] According to a method for generating urban mobile network traffic based on the present invention, the step of adversarial training on real traffic data and generated traffic sample data based on a discriminator network specifically includes: projecting the real traffic data and the generated traffic sample data to obtain real traffic data projection vectors and generated traffic sample data projection vectors, respectively; processing the real traffic data projection vectors and generated traffic sample data projection vectors through a multilayer perceptron, layer normalization, and activation function to obtain the discrimination result output by the discriminator network; and performing adversarial training on the urban mobile network traffic pattern generative adversarial network at different time scales based on the discrimination result and loss function.

[0008] According to a method for generating urban mobile network traffic provided by the present invention, the loss function is determined using the following formula:

[0009]

[0010] Where L represents the loss function; D(S) represents the discrimination result of the real traffic data; This indicates the discrimination result of the generated traffic sample data; This indicates the distribution of the actual traffic data; This indicates the distribution of the generated traffic sample data; This represents uniform sampling between real and generated sample pairs; Indicates the discrimination result of the sample; λ represents the distribution of the sample; λ represents the weight of the penalty term; Indicates gradient operation; represents the expectation operation; ||.||2 represents the L2 norm operation.

[0011] According to a method for generating urban mobile network traffic provided by the present invention, the urban mobile network traffic pattern generation adversarial network at different time scales includes an urban mobile network traffic pattern generation adversarial network at a daily scale, wherein the urban mobile network traffic pattern at a daily scale is a traffic pattern averaged by base stations over multiple days on a daily time scale; the generator network includes a first generator network; the discriminator network includes a first discriminator network; the generated traffic sample data includes urban mobile network traffic pattern sample data at a daily scale; the discrimination result includes a first discrimination result; the knowledge graph embedding vectors of multiple base stations transmitting known network traffic are input to the generator network to obtain generated traffic sample data, specifically including: presetting a first preset traffic pattern matrix, wherein the first preset traffic pattern matrix is ​​an initialized traffic pattern at a daily scale; inputting the knowledge graph embedding vectors of the base stations transmitting known network traffic and noise to the first generator network, and passing them through a multilayer perceptron, layer normalization, and excitation... The active function converts the knowledge graph embedding vector of the base station transmitting known network traffic and the noise into a first projection vector; the first projection vector is multiplied by the first preset traffic pattern matrix, and then processed by an activation function to obtain the city mobile network traffic pattern sample data on a daily scale; the adversarial training of the real traffic data and the generated traffic sample data based on the discriminator network specifically includes: projecting the real traffic data and the city mobile network traffic pattern sample data on a daily scale to obtain the real traffic data projection vector and the city mobile network traffic pattern sample data projection vector on a daily scale, respectively; processing the real traffic data projection vector and the city mobile network traffic pattern sample data projection vector on a daily scale through a multilayer perceptron, layer normalization, and activation function to obtain the first discrimination result output by the first discriminator network; and adversarial training of the city mobile network traffic pattern generative adversarial network on a daily scale based on the first discrimination result and the loss function.

[0012] According to a method for generating urban mobile network traffic provided by the present invention, the first preset traffic pattern matrix is ​​determined in the following manner: the real traffic data is clustered based on a clustering algorithm to obtain multiple first cluster centers; the multiple first cluster centers are used as the initial values ​​of each basis vector in the first preset traffic pattern matrix, and the first preset traffic pattern matrix is ​​obtained based on the initial values.

[0013] According to a method for generating urban mobile network traffic provided by the present invention, the urban mobile network traffic pattern generation adversarial network at different time scales includes a weekly urban mobile network traffic pattern generation adversarial network, wherein the weekly urban mobile network traffic pattern is a traffic pattern in which the traffic transmitted by the base station over multiple weeks is averaged on a weekly time scale; the generator network includes a second generator network; the discriminator network includes a second discriminator network; the generated traffic sample data includes weekly urban mobile network traffic pattern sample data; the discrimination result includes a second discrimination result; the knowledge graph embedding vectors of multiple base stations transmitting known network traffic are input to the generator network to obtain generated traffic sample data, specifically including: presetting a second preset traffic pattern matrix, wherein the second preset traffic pattern matrix is ​​an initialized weekly traffic pattern that does not include the daily traffic pattern; inputting the knowledge graph embedding vectors of the base stations transmitting known network traffic and noise to the second generator network, and converting the knowledge graph embedding vectors of the base stations transmitting known network traffic and the noise into a second projection vector through a multilayer perceptron, layer normalization, and activation function; The second projection vector and the second preset traffic pattern matrix are multiplied and then processed by an activation function to obtain weekly traffic pattern sample data that does not include the daily traffic pattern. Based on the weekly traffic pattern sample data that does not include the daily traffic pattern and the duplicate sample data of the daily urban mobile network traffic pattern sample data, the weekly urban mobile network traffic pattern sample data is obtained. The adversarial training of the discriminator network on the real traffic data and the generated traffic sample data specifically includes: projecting the real traffic data and the weekly urban mobile network traffic pattern sample data to obtain real traffic data projection vectors and weekly urban mobile network traffic pattern sample data projection vectors, respectively; processing the real traffic data projection vectors and the weekly urban mobile network traffic pattern sample data projection vectors through a multilayer perceptron, layer normalization, and activation function to obtain the second discrimination result output by the second discriminator network; and adversarial training of the weekly urban mobile network traffic pattern generative adversarial network based on the second discrimination result and the loss function.

[0014] According to a method for generating urban mobile network traffic provided by the present invention, the second preset traffic pattern matrix is ​​determined in the following manner: the real traffic data is clustered based on a clustering algorithm to obtain multiple second cluster centers; the multiple second cluster centers are used as the initial values ​​of each basis vector in the second preset traffic pattern matrix, and the second preset traffic pattern matrix is ​​obtained based on the initial values.

[0015] According to the present invention, a method for generating urban mobile network traffic includes a generator network comprising a third generator network; a discriminator network comprising a third discriminator network; generated traffic sample data comprising continuous-time urban mobile network traffic sample data; and a discrimination result comprising a third discrimination result. The method for generating traffic sample data by inputting knowledge graph embedding vectors of multiple base stations transmitting known network traffic into the generator network specifically includes: inputting the knowledge graph embedding vectors of the base stations transmitting known network traffic and noise into the third generator network, and obtaining a transformed sequence through a multilayer perceptron and layer normalization, wherein the sequence length of the transformed sequence is the same as the sequence length of the continuous-time urban mobile network traffic sample data; and inputting the transformed sequence as initial values ​​into three temporal convolutional neural networks with different kernel sizes. The process involves processing the data using a multilayer perceptron and a first activation function to obtain continuous-time urban mobile network traffic fluctuation sample data. Based on this continuous-time urban mobile network traffic fluctuation sample data, and repeated sample data from the weekly-scale urban mobile network traffic pattern sample data, continuous-time urban mobile network traffic sample data is obtained. The adversarial training of the real traffic data and the generated traffic sample data using a discriminator network specifically includes: inputting the real traffic data and the continuous-time urban mobile network traffic sample data into three temporal convolutional neural networks with different kernel sizes, and processing them using a multilayer perceptron and a second activation function to obtain a third discrimination result output by the third discriminator network; and performing adversarial training on the continuous-time urban mobile network traffic generation adversarial network based on the third discrimination result and a loss function.

[0016] This invention also provides an urban mobile network traffic generation device, the device comprising: a construction module for extracting multiple entities based on urban data information and constructing an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; a processing module for obtaining knowledge graph embedding vectors corresponding to the base stations based on the urban knowledge graph; and a training module for inputting the knowledge graph embedding vectors and noise into a generator network of a trained generative adversarial network model to obtain generated traffic data output by the generator network; wherein the generated traffic data is urban traffic data generated in the context of the urban data information. The city's mobile network traffic; wherein, the generative adversarial network model includes a city mobile network traffic pattern generative adversarial network at different time scales and a continuous-time city mobile network traffic generative adversarial network. Both the city mobile network traffic pattern generative adversarial network at different time scales and the continuous-time city mobile network traffic generative adversarial network are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic being input into the generator network to obtain generated traffic sample data. Then, adversarial training is performed on the real traffic data and the generated traffic sample data based on the discriminator network. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the urban mobile network traffic generation method as described above.

[0018] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the urban mobile network traffic generation method as described above.

[0019] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the urban mobile network traffic generation method as described above.

[0020] This invention provides a method, apparatus, electronic device, and storage medium for generating urban mobile network traffic. It constructs an urban knowledge graph using urban data information and designs a generative model using a generative adversarial network (GAN) to gradually capture the daily, weekly, and long-term changing patterns of urban mobile network traffic, thereby obtaining a trained GAN model. The knowledge graph embedding vectors corresponding to base stations, obtained based on the urban knowledge graph, are then input into the trained GAN model to obtain urban mobile network traffic data within the context of urban data information. This invention achieves the generation of urban mobile network traffic data with high accuracy and reliability. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention 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 invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is one of the flowcharts illustrating the urban mobile network traffic generation method provided by the present invention;

[0023] Figure 2 This is a schematic diagram of the process of adversarial training based on a discriminator network on real traffic data and generated traffic sample data provided by the present invention;

[0024] Figure 3 This is a schematic diagram of the training process of the Generative Adversarial Network for Urban Mobile Network Traffic Patterns on a Daily Scale, provided by the present invention.

[0025] Figure 4 This is a schematic diagram of a training scenario for a generative adversarial network for urban mobile network traffic patterns based on a daily scale, provided by the present invention.

[0026] Figure 5 This is a schematic diagram of the training process of the urban mobile network traffic pattern generative adversarial network provided by the present invention.

[0027] Figure 6 This is a schematic diagram of a training scenario for a generative adversarial network for urban mobile network traffic patterns based on a weekly scale, provided by the present invention.

[0028] Figure 7 This is a schematic diagram of a training scenario for the continuous-time urban mobile network traffic generation adversarial network provided by the present invention.

[0029] Figure 8 This is a schematic diagram of a training scenario for the continuous-time urban mobile network traffic generation adversarial network provided by the present invention.

[0030] Figure 9 This is the second flowchart illustrating the urban mobile network traffic generation method provided by the present invention;

[0031] Figure 10 This is a schematic diagram of the urban mobile network traffic generation device provided by the present invention;

[0032] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0034] The objective of this invention is to generate reliable large-scale urban mobile network traffic based on real urban mobile network traffic samples. Since different base stations are distributed across various areas of a city, serving surrounding business districts, POIs (Points of Interest), etc., the traffic they transmit changes in real time as mobile users and devices access the mobile network and generate data demands. In application, an urban knowledge graph can first be constructed using urban environmental information. Based on this, a generative model is designed using granular learning and generative adversarial networks to gradually capture the daily, weekly, and long-term (corresponding to continuous time) variation patterns of urban mobile network traffic. Urban knowledge is then introduced into the generative model through knowledge graph embedding technology and conditional mechanisms. Finally, the generative model is trained, and reliable large-scale urban mobile network traffic is obtained based on the final trained generative model.

[0035] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 1 Please provide an explanation.

[0036] Figure 1 This is one of the flowcharts illustrating the urban mobile network traffic generation method provided by the present invention.

[0037] In an exemplary embodiment of the present invention, combined with Figure 1 As shown, the method for generating urban mobile network traffic may include steps 110 to 130, which will be described in detail below.

[0038] In step 110, multiple entities are extracted based on urban data information, and an urban knowledge graph is constructed based on the relationships between entities. The entities include at least base stations, regions, business districts, POIs, POI categories, and brands.

[0039] In one example, urban data information may include existing environmental information and planning information for the city. In one embodiment, six entities are primarily considered in the city: base stations, regions, business districts, Points of Interest (POIs), POI categories, and brands. The relationships between these entities may include belonging, being located at, being adjacent, or being similar. In application, triples can be constructed from the various entities and their relationships, and then a city knowledge graph can be built based on these triples.

[0040] In step 120, the knowledge graph embedding vector corresponding to the base station is obtained based on the city knowledge graph.

[0041] In one embodiment, the knowledge graph embedding vector corresponding to the base station can be obtained based on the city knowledge graph in the following way: based on the city knowledge graph, the knowledge graph embedding vector corresponding to the base station is obtained through the knowledge graph embedding model.

[0042] In one example, the knowledge graph embedding model TuckER can be used to obtain the i-th base station BS. i The corresponding knowledge graph embedding vector K i .

[0043] In step 130, the knowledge graph embedding vector and noise are input into the generator network of the trained generative adversarial network model to obtain generated traffic data output by the generator network. The generated traffic data is urban mobile network traffic in the context of urban data information. The generative adversarial network model includes urban mobile network traffic pattern generative adversarial networks at different time scales and continuous-time urban mobile network traffic generative adversarial networks. Both the urban mobile network traffic pattern generative adversarial networks at different time scales and the continuous-time urban mobile network traffic generative adversarial networks are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic being input into the generator network to obtain generated traffic sample data. Then, adversarial training is performed on the real traffic data and the generated traffic sample data based on the discriminator network. The real traffic data is the traffic data transmitted by base stations transmitting known network traffic.

[0044] It should be noted that urban mobile network traffic in the context of urban data information can be understood as the performance of urban mobile network traffic based on existing urban data information and under existing mobile network planning.

[0045] The urban mobile network traffic patterns at different time scales can include daily-scale and weekly-scale urban mobile network traffic pattern generative adversarial networks (GANs). Daily-scale urban mobile network traffic patterns represent traffic transmitted by base stations over multiple days, averaged on a daily time scale. Weekly-scale urban mobile network traffic patterns represent traffic transmitted by base stations over multiple weeks, averaged on a weekly time scale. Continuous-time urban mobile network traffic GANs refer to urban mobile network traffic GANs operating over a continuous long period. This continuous long period is longer than a week; in one example, a continuous long period could refer to a month.

[0046] In application, the output results obtained from the weekly urban mobile network traffic pattern generative adversarial network can be applied to the training of the continuous-time urban mobile network traffic generative adversarial network, thereby improving the model training accuracy and fast convergence of the continuous-time urban mobile network traffic generative adversarial network.

[0047] This invention provides a method, apparatus, electronic device, and storage medium for generating urban mobile network traffic. It constructs an urban knowledge graph using urban data information and designs a generative model using a generative adversarial network (GAN) to gradually capture the daily, weekly, and long-term changing patterns of urban mobile network traffic, thereby obtaining a trained GAN model. The knowledge graph embedding vectors corresponding to base stations, obtained based on the urban knowledge graph, are then input into the trained GAN model to obtain urban mobile network traffic data within the context of urban data information. This invention achieves the generation of urban mobile network traffic data with high accuracy and reliability.

[0048] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 2 The process of adversarial training based on a discriminator network on real traffic data and generated traffic sample data is explained.

[0049] In an exemplary embodiment of the present invention, combined with Figure 2 As can be seen, adversarial training based on the discriminator network on real traffic data and generated traffic sample data can include steps 210 and 230, which will be described in detail below.

[0050] In step 210, the real traffic data and the generated traffic sample data are projected to obtain the real traffic data projection vector and the generated traffic sample data projection vector, respectively.

[0051] In application, the discriminator network contains a traffic pattern basis vector matrix, which corresponds to the traffic pattern basis vector matrix in the generator network. In one example, the real traffic data and generated traffic sample data can be multiplied by the traffic pattern basis vector matrix in the discriminator network to obtain the projection vectors of the real traffic data and the generated traffic sample data.

[0052] In step 220, the projection vectors of the real traffic data and the projection vectors of the generated traffic sample data are processed by a multilayer perceptron, layer normalization, and activation function to obtain the discrimination result output by the discriminator network.

[0053] In one embodiment, the projection vectors of real traffic data and generated traffic sample data can be processed through a multilayer perceptron, layer normalization, and activation functions to obtain the discrimination result output by the discriminator network. The activation function can be a sigmoid activation function.

[0054] In step 230, adversarial training is performed on the generative adversarial network for urban mobile network traffic patterns at different time scales, based on the discrimination results and loss function.

[0055] In one embodiment, Wasserstein distance with added gradient penalty can be used as the loss function. Based on the discrimination results and the loss function, adversarial networks are generated and adversarial training is performed on urban mobile network traffic patterns at different time scales. The loss function can be determined using the following formula:

[0056]

[0057] Where L represents the loss function; D(S) represents the discrimination result of the actual traffic data; This indicates the discrimination result of generating traffic sample data; This represents the distribution of actual traffic data; This indicates the distribution of the generated traffic sample data; This represents uniform sampling between real and generated sample pairs; Indicates the discrimination result of the sample; λ represents the distribution of the sample; λ represents the weight of the penalty term; Indicates gradient operation; represents the expectation operation; ||.||2 represents the L2 norm operation.

[0058] During training, the generator network generates samples that are as close to the real samples as possible by minimizing the loss function, while the discriminator network determines the authenticity of the samples as much as possible by maximizing the loss function.

[0059] In the scenario of a city mobile network traffic pattern generative adversarial network with a daily scale, the generator network may include a first generator network; the discriminator network may include a first discriminator network; the generated traffic sample data may include city mobile network traffic pattern sample data with a daily scale; and the discrimination result may include a first discrimination result.

[0060] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 3 The training process of a generative adversarial network for urban mobile network traffic patterns on a daily scale is explained.

[0061] In an exemplary embodiment of the present invention, combined with Figure 3 As can be seen, the training process of the urban mobile network traffic pattern generative adversarial network on a daily scale can include steps 310 to 360, which will be described in detail below.

[0062] In step 310, a first preset flow pattern matrix is ​​preset, wherein the first preset flow pattern matrix is ​​an initialized flow pattern on a daily scale.

[0063] Combination Figure 4 To clarify, a day-scale urban mobile network traffic pattern generative adversarial network can include a first generator network G. d and the first discriminator network D d Two parts.

[0064] In one embodiment, a first preset traffic pattern matrix can be preset. Among them, the first preset traffic pattern matrix This corresponds to the traffic pattern basis vector matrix in a generator network described earlier. In the first generator network G... d In the first preset traffic pattern matrix This represents possible daily-scale flow patterns, where each vector represents a basis vector for a flow pattern, and combining multiple basis vectors can yield diverse flow patterns.

[0065] In one embodiment, the first preset traffic pattern matrix can be determined as follows: real traffic data is clustered using a clustering algorithm to obtain multiple first cluster centers; these multiple first cluster centers are used as the initial values ​​of each basis vector in the first preset traffic pattern matrix, and the first preset traffic pattern matrix is ​​obtained based on these initial values. The clustering algorithm can be the K-means algorithm. In this embodiment, using multiple first cluster centers as the initial values ​​of each basis vector in the first preset traffic pattern matrix and obtaining the first preset traffic pattern matrix based on these initial values ​​can improve training efficiency and accelerate convergence.

[0066] In step 320, the knowledge graph embedding vector and noise of the base station transmitting known network traffic are input into the first generator network, and the knowledge graph embedding vector and noise of the base station transmitting known network traffic are converted into the first projection vector through a multilayer perceptron, layer normalization and activation function.

[0067] In one embodiment, the knowledge graph of a base station transmitting known network traffic can be embedded into vector K. i (correspond Figure 4 K) and noise (corresponding to Figure 4 The input Z) is the first generator network G. dThe method employs a multilayer perceptron, layer normalization, and activation functions to transform the knowledge graph embedding vectors and noise of base stations transmitting known network traffic into a first projection vector. The activation function can be a SoftMax activation function. The first projection vector represents the traffic pattern basis vector matrix (corresponding to a first preset traffic pattern matrix). A way to combine basis vectors.

[0068] In step 330, the first projection vector and the first preset traffic pattern matrix are multiplied together, and then processed by an activation function to obtain urban mobile network traffic pattern sample data on a daily scale.

[0069] In one embodiment, the first projection vector and the first preset traffic pattern matrix can be multiplied, and then processed by an activation function to obtain daily-scale urban mobile network traffic pattern sample data. The daily-scale urban mobile network traffic pattern sample data can be represented as follows: The activation function can be the GELU activation function.

[0070] In step 340, the actual traffic data and the daily urban mobile network traffic pattern sample data are projected to obtain the actual traffic data projection vector and the daily urban mobile network traffic pattern sample data projection vector, respectively.

[0071] In one embodiment, in the first discriminator network D d There is also a flow pattern basis vector matrix in it. Real traffic data and daily-scale urban mobile network traffic pattern sample data can be input into the first discriminator network D. d Then, with the flow pattern basis vector matrix Multiplying these vectors yields the projection of the sample onto multiple traffic pattern basis vectors (corresponding to the projection vector of the actual traffic data and the projection vector of the urban mobile network traffic pattern sample data on a daily scale).

[0072] In step 350, the projection vector of the real traffic data and the projection vector of the urban mobile network traffic pattern sample data on a daily scale are processed by a multilayer perceptron, layer normalization and activation function to obtain the first discrimination result output by the first discriminator network.

[0073] In one embodiment, the first discriminator network D is obtained by passing the projection vectors of real traffic data and the projection vectors of urban mobile network traffic pattern sample data on a daily scale through a multilayer perceptron, layer normalization, and a sigmoid activation function. d The first discriminant results D(R) and D(F) are output.

[0074] In step 360, adversarial training is performed on the urban mobile network traffic pattern generative adversarial network based on the first discrimination result and the loss function.

[0075] In one embodiment, adversarial training can be performed on a daily-scale urban mobile network traffic pattern generative adversarial network based on the first discrimination results D(R) and D(F) and the loss function.

[0076] In the scenario where the urban mobile network traffic pattern generative adversarial network is based on a weekly scale, the generator network may include a second generator network; the discriminator network may include a second discriminator network; the generated traffic sample data may include urban mobile network traffic pattern sample data based on a weekly scale; and the discrimination result may include a second discrimination result.

[0077] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 5 The training process of a generative adversarial network for urban mobile network traffic patterns on a weekly scale is explained.

[0078] In an exemplary embodiment of the present invention, combined with Figure 5 As can be seen, the training process of the urban mobile network traffic pattern generative adversarial network on a weekly scale can include steps 510 to 570, which will be described in detail below.

[0079] In step 510, a second preset traffic pattern matrix is ​​preset, wherein the second preset traffic pattern matrix is ​​an initialized traffic pattern based on a week, excluding traffic patterns based on days.

[0080] Combination Figure 6 To clarify, a weekly-scale urban mobile network traffic pattern generative adversarial network can include a second generator network G. w Second discriminator network D w Two parts.

[0081] In one embodiment, a second preset traffic pattern matrix can be preset. Among them, the second preset traffic pattern matrix This can correspond to the traffic pattern basis vector matrix in a generator network as described above.

[0082] To eliminate the influence between traffic patterns at different scales, a second generator network G is used in a weekly-scale urban mobile network traffic pattern generative adversarial network. wFirst, sample data of traffic patterns on a weekly scale can be generated without daily traffic patterns. Then, the daily traffic pattern sample data generated in the previous step (corresponding to the repeatedly generated daily weekday traffic pattern and the repeatedly generated daily weekend traffic pattern in the figure) is added to the daily traffic in the weekly traffic pattern to obtain complete sample data of urban mobile network traffic patterns on a weekly scale.

[0083] Second generator network G w In the second preset traffic pattern matrix This represents possible flow patterns on a weekly scale, where each vector represents a basis vector for a flow pattern, and combining multiple basis vectors can yield diverse flow patterns.

[0084] In one embodiment, the second preset traffic pattern matrix can be determined as follows: real traffic data is clustered using a clustering algorithm to obtain multiple second cluster centers; these multiple second cluster centers are used as the initial values ​​of each basis vector in the second preset traffic pattern matrix, and the second preset traffic pattern matrix is ​​obtained based on these initial values. The clustering algorithm can be the K-means algorithm. In this embodiment, using multiple second cluster centers as the initial values ​​of each basis vector in the second preset traffic pattern matrix and obtaining the second preset traffic pattern matrix based on these initial values ​​can improve training efficiency and accelerate convergence.

[0085] In step 520, the knowledge graph embedding vector and noise of the base station transmitting known network traffic are input into the second generator network, and the knowledge graph embedding vector and noise of the base station transmitting known network traffic are converted into a second projection vector through a multilayer perceptron, layer normalization and activation function.

[0086] In one embodiment, the knowledge graph of a base station transmitting known network traffic can be embedded into a vector (corresponding to...). Figure 6 K) and noise (corresponding to Figure 6 The input Z) is the second generator network G. w The method employs a multilayer perceptron, layer normalization, and activation functions to transform the knowledge graph embedding vectors and noise of base stations transmitting known network traffic into a second projection vector. The activation function can be a SoftMax activation function. The second projection vector represents the traffic pattern basis vector matrix (corresponding to a second preset traffic pattern matrix). A way to combine basis vectors.

[0087] In step 530, the second projection vector and the second preset flow pattern matrix are multiplied together, and then processed by the activation function to obtain flow pattern sample data on a weekly scale that does not include flow patterns on a daily scale.

[0088] In one embodiment, the second projection vector and the second preset traffic pattern matrix are multiplied together, and then processed by an activation function to obtain weekly traffic pattern sample data that does not include daily traffic patterns. The activation function can be the GELU activation function.

[0089] In step 540, based on the weekly traffic pattern sample data (which does not include daily traffic patterns) and the duplicate sample data of the daily urban mobile network traffic pattern sample data, the weekly urban mobile network traffic pattern sample data is obtained.

[0090] In one embodiment, the data can be based on weekly traffic pattern sample data that does not include daily traffic patterns, and duplicate sample data of daily urban mobile network traffic pattern sample data (corresponding to...). Figure 6 By analyzing the repeatedly generated daily traffic patterns on weekdays and weekends, we obtain sample data of urban mobile network traffic patterns on a weekly scale. This sample data is represented as follows: in, This represents the sample data of urban mobile network traffic patterns generated in the previous step, on a daily scale. The repetition.

[0091] It should be noted that, in this embodiment, the duplicate sample data is the daily-scale urban mobile network traffic pattern sample data that has been repeatedly processed so that the length of the duplicate sample data is the same as the length of the weekly-scale traffic pattern sample data containing the daily-scale traffic pattern.

[0092] In step 550, the actual traffic data and the weekly urban mobile network traffic pattern sample data are projected to obtain the actual traffic data projection vector and the weekly urban mobile network traffic pattern sample data projection vector, respectively.

[0093] In one embodiment, in the second discriminator network D w There is also a flow pattern basis vector matrix in it. Real traffic data and weekly-scale urban mobile network traffic pattern sample data can be input into the second discriminator network D. w Then, with the flow pattern basis vector matrix Multiplying these vectors yields the projection of the sample onto multiple traffic pattern basis vectors (corresponding to the projection vector of the actual traffic data and the projection vector of the urban mobile network traffic pattern sample data on a weekly scale).

[0094] In step 560, the projection vector of the real traffic data and the projection vector of the urban mobile network traffic pattern sample data on a weekly scale are processed by a multilayer perceptron, layer normalization and activation function to obtain the second discrimination result output by the second discriminator network.

[0095] In one embodiment, the second discriminator network D is obtained by passing the projection vectors of real traffic data and the projection vectors of weekly urban mobile network traffic pattern sample data through a multilayer perceptron, layer normalization, and a sigmoid activation function. w The output consists of the second discrimination results D(R) and D(F).

[0096] In step 570, adversarial training is performed on the urban mobile network traffic pattern generative adversarial network based on the second discrimination result and the loss function.

[0097] In one embodiment, adversarial training can be performed on a weekly-scale generative adversarial network for urban mobile network traffic patterns based on the second discrimination results D(R) and D(F) and the loss function.

[0098] In the scenario of a continuous-time urban mobile network traffic generation adversarial network, the generator network may include a third generator network; the discriminator network may include a third discriminator network; the generated traffic sample data may include continuous-time urban mobile network traffic sample data; and the discrimination result may include a third discrimination result.

[0099] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 7 The training process of a continuous-time urban mobile network traffic generative adversarial network is explained.

[0100] In an exemplary embodiment of the present invention, combined with Figure 7 As can be seen, the training process of the continuous-time urban mobile network traffic generative adversarial network may include steps 710 to 750, and each step will be described below.

[0101] In step 710, the knowledge graph embedding vector of the base station transmitting known network traffic and noise are input into the third generator network, and the transformed sequence is obtained through multilayer perceptron and layer normalization. The sequence length of the transformed sequence is the same as the sequence length of the continuous-time urban mobile network traffic sample data.

[0102] In one embodiment, the third generator and the third discriminator of the continuous-time urban mobile network traffic generative adversarial network are both designed based on a temporal convolutional neural network (TCN). The TCN includes an input layer, multiple hidden layers, and an output layer. Causal convolutions are performed within each layer, and the results are output to the next layer. Temporal causal relationships are thus propagated between layers, with the stride of each propagation determined by the shape of the convolutional kernel.

[0103] Combination Figure 8 To clarify, a continuous-time urban mobile network traffic generation adversarial network can include a third generator network (corresponding to the long-time residual traffic generator in the diagram) G. r With the third discriminator network (corresponding to the long-term residual flow discriminator in the diagram) D r .

[0104] In one embodiment, the knowledge graph of a base station transmitting known network traffic can be embedded into a vector (corresponding to...). Figure 8 K) and noise (corresponding to Figure 8 The Z) input is fed into the third generator network G. r The sequence is then transformed into a sequence of the same length as the sequence to be generated using a multilayer perceptron and layer normalization. This sequence is then used as the initial value input into three temporal convolutional neural networks (TCNs) with different kernel sizes (corresponding to TCNs with kernel sizes of months, weeks, and days, respectively, as shown in the diagram). The kernel sizes of the three temporal convolutional neural networks are days, weeks, and months, respectively, indicating that each value in each layer is influenced by the sequence values ​​from the previous day, week, and month in the previous layer.

[0105] In step 720, the transformed sequence is used as the initial value and input into three temporal convolutional neural networks with different kernel sizes. The sequence is then processed by a multilayer perceptron and a first activation function to obtain continuous-time urban mobile network traffic fluctuation sample data.

[0106] In one embodiment, the outputs of three temporal convolutional neural networks can be passed through a multilayer perceptron and the GELU activation function to obtain continuous-time urban mobile network traffic fluctuation sample data.

[0107] In step 730, continuous-time urban mobile network traffic sample data is obtained based on continuous-time urban mobile network traffic fluctuation sample data and repeated sample data of urban mobile network traffic pattern sample data on a weekly scale.

[0108] In one embodiment, the data can be based on continuous-time urban mobile network traffic fluctuation sample data and repeated sample data of urban mobile network traffic patterns on a weekly scale (corresponding to...). Figure 8 The weekly traffic pattern generated repeatedly in the data yields continuous-time urban mobile network traffic sample data. in, This represents the sample data of urban mobile network traffic patterns generated in the previous step, on a weekly scale. The repetition.

[0109] It should be noted that, in this embodiment, the repeated sample data refers to the repeated processing of the urban mobile network traffic pattern sample data on a weekly scale, so that the length of the repeated sample data is the same as the length of the continuous time urban mobile network traffic sample data.

[0110] In step 740, real traffic data and continuous-time urban mobile network traffic sample data are input into three temporal convolutional neural networks with different kernel sizes, and processed by a multilayer perceptron and a second activation function to obtain the third discrimination result output by the third discriminator network.

[0111] In one embodiment, in the third discriminator network D r In this process, continuous-time urban mobile network traffic sample data can be obtained. Inputting real traffic data with the third generator network G r In three similar temporal convolutional neural networks with daily, weekly, and monthly convolutional kernel sizes, after passing through a multilayer perceptron and a sigmoid activation function, the third discriminant results D(R) and D(F) output by the third discriminator network can be obtained.

[0112] In step 750, adversarial training is performed on the continuous-time urban mobile network traffic generative adversarial network based on the third discrimination result and the loss function.

[0113] The loss function is the same as that used for training urban mobile network traffic pattern generative adversarial networks on a daily or weekly scale.

[0114] In one embodiment, adversarial training can be performed on a generative adversarial network for continuous-time urban mobile network traffic patterns based on third discrimination results D(R) and D(F) and a loss function.

[0115] It should be noted that the real-world traffic data used in training the daily, weekly, and extended-time urban mobile network traffic pattern generative adversarial networks (GANs) are different. In actual application, the real-world traffic data corresponds to the respective trained models.

[0116] To further introduce the urban mobile network traffic generation method provided by this invention, the following will be combined with... Figure 9 Please provide an explanation.

[0117] Figure 9 This is the second flowchart illustrating the urban mobile network traffic generation method provided by the present invention.

[0118] In an exemplary embodiment of the present invention, such as Figure 9 As shown, a city knowledge graph can be constructed based on entities such as base stations, regions, business districts, POIs, POI categories, and brands in the city, as well as the relationships between these entities. Furthermore, generative adversarial networks can be constructed to generate daily traffic patterns, weekly traffic patterns, and long-term (corresponding to continuous time) urban mobile network traffic.

[0119] Each generative adversarial network (GAN) consists of a generator and a discriminator. The generator generates urban mobile network traffic for corresponding durations (corresponding to Stage 1: daily traffic pattern generation; Stage 2: weekly traffic pattern generation; and Stage 3: long-term traffic generation, respectively, as shown in the diagram). The discriminator distinguishes between real data (corresponding to real daily traffic patterns, real weekly traffic patterns, and real traffic, respectively, as shown in the diagram) and generated data (corresponding to generated daily traffic patterns, generated weekly traffic patterns, and generated long-term traffic, respectively, as shown in the diagram). The generator and discriminator are trained adversarially to gradually obtain a reliable model.

[0120] Because base stations continuously transmit urban mobile network traffic across various areas of a city, they can provide high spatial resolution. In applications, the traffic transmitted by each base station can be used as a sample when generating urban mobile network traffic data. The traffic transmitted by each base station over a period of time can be represented as... Among them, V t Let V represent the traffic transmitted by the base station in the t-th time segment, and T represent the total time length. Since the primary concern is the distribution of traffic in urban mobile networks, it can be normalized using the L2 norm of the traffic sequence V, thus obtaining... Where |V|2 represents the L2 norm of the flow sequence V.

[0121] Furthermore, the problem of generating urban mobile network traffic can be expressed as follows: Given a real dataset of IoT device network traffic... Among them, S i Let O represent the normalized network traffic transmitted by the i-th base station in the dataset. Given a real IoT traffic dataset O, this invention, based on the idea of ​​granular learning, designs and trains a generative adversarial network to generate a reliable dataset.

[0122] To better generate traffic patterns at different time scales in urban mobile network traffic, we can define daily-scale traffic patterns (corresponding to the daily-scale urban mobile network traffic patterns mentioned earlier) and weekly-scale traffic patterns (corresponding to the weekly-scale urban mobile network traffic patterns mentioned earlier). The daily-scale traffic pattern refers to the average of base station traffic V over multiple days on a daily time scale, which can be expressed as... in, This represents the flow rate within the t-th hour of each day; This refers to the total number of days. Weekly traffic patterns refer to the average base station traffic V over multiple weeks, calculated on a weekly time scale, and can be expressed as... in, This represents the traffic flow in the t-th hour of each week. This refers to the total number of weeks.

[0123] In application, generative adversarial networks can be designed and trained separately, first generating a daily-scale traffic pattern S. d And in S d Based on this, a weekly-scale flow pattern S is generated. w Finally in S w Based on this, long-term urban mobile network traffic (corresponding to the traffic data generated above) S is generated.

[0124] To further illustrate the urban mobile network traffic generation method of the present invention, the application process of the urban mobile network traffic generation method is described below.

[0125] In one embodiment, a city plans to build a new urban area on the city's edge and wants to know how the city's mobile network traffic will perform under the existing mobile network plan before construction, so as to use this as a reference for planning and construction.

[0126] Based on the existing mobile network traffic and urban environmental information of the built-up areas of the city, the urban mobile network traffic generation method provided by this invention can generate urban mobile network traffic that may be generated within one month after the completion of the new urban area, according to this existing information and the environmental information and network planning in the urban plan.

[0127] First, by utilizing the city's existing environmental and planning information, multiple entities such as base stations, regions, business districts, POIs, POI categories, and brands can be extracted. Based on the relationships between entities, a city knowledge graph can be constructed, and the knowledge graph embedding vector can be learned.

[0128] Furthermore, the existing mobile network traffic training model is utilized in the established areas. Data from the past month is used as the training set to calculate the hourly traffic transmitted by each base station, yielding a traffic sequence sample V. Based on these traffic samples, traffic patterns with daily and weekly periods are calculated. K-means clustering is then performed, and the cluster centers are used to initialize the traffic pattern basis vector matrix M in the model. d Considering the balance between computational complexity and performance, and controlling the clustering to 32 classes, the traffic pattern basis vector matrix M is... d The dimension is T×32, where T = 24×31, which is the number of hours in a month.

[0129] After pre-training, M=32 is selected as the number of traffic sequence samples used for training in each batch. In each iteration, the discriminator is trained and updated N=5 times, and then the generator is updated once. When training the discriminator, M traffic sequences are first randomly selected as real samples. Then, the knowledge graph embedding vectors of the base stations corresponding to these M traffic sequences and the randomly generated noise in the (0,1) interval are used to generate M simulated samples through the generator. Subsequently, the simulated samples and real samples are simultaneously input into the discriminator. The loss function is calculated based on the discriminator's discrimination result, and the adaptive moment estimation optimizer is used to maximize the loss function and update the discriminator parameters.

[0130] After training the discriminator N times, M knowledge graph embedding vectors of randomly selected traffic sequences and randomly generated noise in the (0,1) interval are used to generate M simulation samples. The discriminator with updated parameters is then used to distinguish these simulation samples. The loss function is calculated based on the discrimination results, and an adaptive moment estimation optimizer is used to minimize the loss function and update the generator parameters. After approximately 250 iterations, the model reaches the convergence threshold, and training is complete.

[0131] Finally, by simultaneously inputting the knowledge graph embedding vector corresponding to each planned base station in the new urban area and the randomly generated noise vector in the (0,1) interval into the generator, the urban mobile network traffic that may be generated within one month after the completion of the new urban area can be obtained.

[0132] As described above, the urban mobile network traffic generation method, apparatus, electronic device, and storage medium provided by this invention construct an urban knowledge graph using urban data information and design a generative model using a generative adversarial network (GAN) to gradually capture the daily, weekly, and long-term changing patterns of urban mobile network traffic, thereby obtaining a trained GAN model. The knowledge graph embedding vectors corresponding to base stations, obtained based on the urban knowledge graph, are then input into the trained GAN model to obtain urban mobile network traffic within the context of urban data information. This invention achieves the generation of urban mobile network traffic with high accuracy and reliability.

[0133] Based on the same concept, the present invention also provides an urban mobile network traffic generation device.

[0134] The urban mobile network traffic generation device provided by the present invention is described below. The urban mobile network traffic generation device described below can be referred to in correspondence with the urban mobile network traffic generation method described above.

[0135] Figure 10 This is a schematic diagram of the urban mobile network traffic generation device provided by the present invention.

[0136] In an exemplary embodiment of the present invention, combined with Figure 10 As can be seen, the urban mobile network traffic generation device may include a construction module 1010, a processing module 1020, and a training module 1030. Each module will be described in detail below.

[0137] The construction module 1010 can be configured to extract multiple entities based on urban data information and construct an urban knowledge graph based on the relationships between entities. Entities can include at least base stations, regions, business districts, points of interest (POIs), POI categories, and brands.

[0138] The processing module 1020 can be configured to obtain the knowledge graph embedding vector corresponding to the base station based on the city knowledge graph.

[0139] The training module 1030 can be configured to input knowledge graph embedding vectors and noise into the generator network of the trained generative adversarial network model to obtain generated traffic data output by the generator network. The generated traffic data refers to urban mobile network traffic within the context of urban data information.

[0140] The generative adversarial network (GAN) model can include urban mobile network traffic pattern GANs at different time scales and continuous-time urban mobile network traffic GANs. Both the urban mobile network traffic pattern GANs at different time scales and the continuous-time urban mobile network traffic GANs are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic being input into the generator network to obtain generated traffic sample data. Then, the discriminator network is used to perform adversarial training on the real traffic data and the generated traffic sample data. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

[0141] In an exemplary embodiment of the present invention, the processing module 1020 may obtain the knowledge graph embedding vector corresponding to the base station based on the city knowledge graph in the following manner: based on the city knowledge graph, the knowledge graph embedding vector corresponding to the base station is obtained through the knowledge graph embedding model.

[0142] In an exemplary embodiment of the present invention, the training module 1030 may perform adversarial training on real traffic data and generated traffic sample data based on the discriminator network in the following manner: the real traffic data and generated traffic sample data are projected to obtain real traffic data projection vectors and generated traffic sample data projection vectors, respectively; the real traffic data projection vectors and generated traffic sample data projection vectors are processed through a multilayer perceptron, layer normalization, and activation function to obtain the discrimination result output by the discriminator network; based on the discrimination result and loss function, adversarial training is performed on the urban mobile network traffic pattern generative adversarial network at different time scales.

[0143] In an exemplary embodiment of the present invention, the training module 1030 may determine the loss function using the following formula:

[0144]

[0145] Where L represents the loss function; D(S) represents the discrimination result of the actual traffic data; This indicates the discrimination result of generating traffic sample data; This represents the distribution of actual traffic data; This indicates the distribution of the generated traffic sample data; This represents uniform sampling between real and generated sample pairs; Indicates the discrimination result of the sample; λ represents the distribution of the sample; λ represents the weight of the penalty term; This represents gradient operation; represents the expectation operation; ||.||2 represents the L2 norm operation.

[0146] In an exemplary embodiment of the present invention, the urban mobile network traffic pattern generation adversarial network at different time scales may include an urban mobile network traffic pattern generation adversarial network at a daily scale, wherein the urban mobile network traffic pattern at a daily scale is a traffic pattern in which the traffic transmitted by the base station over multiple days is averaged on a daily time scale; the generator network may include a first generator network; the discriminator network may include a first discriminator network; the generated traffic sample data may include urban mobile network traffic pattern sample data at a daily scale; and the discrimination result may include a first discrimination result.

[0147] The training module 1030 can input the knowledge graph embedding vectors of multiple base stations transmitting known network traffic into the generator network in the following manner to obtain generated traffic sample data: a first preset traffic pattern matrix is ​​preset, wherein the first preset traffic pattern matrix is ​​an initialized traffic pattern on a daily scale; the knowledge graph embedding vectors of base stations transmitting known network traffic and noise are input into the first generator network, and the knowledge graph embedding vectors of base stations transmitting known network traffic and noise are converted into a first projection vector through a multilayer perceptron, layer normalization and activation function; the first projection vector and the first preset traffic pattern matrix are multiplied and then processed by the activation function to obtain urban mobile network traffic pattern sample data on a daily scale.

[0148] The training module 1030 can perform adversarial training on real traffic data and generated traffic sample data based on the discriminator network in the following way: Project the real traffic data and the daily-scale urban mobile network traffic pattern sample data to obtain the real traffic data projection vector and the daily-scale urban mobile network traffic pattern sample data projection vector, respectively; process the real traffic data projection vector and the daily-scale urban mobile network traffic pattern sample data projection vector through a multilayer perceptron, layer normalization, and activation function to obtain the first discrimination result output by the first discriminator network; and perform adversarial training on the daily-scale urban mobile network traffic pattern generative adversarial network based on the first discrimination result and the loss function.

[0149] In an exemplary embodiment of the present invention, the training module 1030 may determine the first preset traffic pattern matrix in the following manner: clustering real traffic data based on a clustering algorithm to obtain multiple first cluster centers; using the multiple first cluster centers as the initial values ​​of each basis vector in the first preset traffic pattern matrix, and obtaining the first preset traffic pattern matrix based on the initial values.

[0150] In an exemplary embodiment of the present invention, the urban mobile network traffic pattern generation adversarial network at different time scales may include a weekly urban mobile network traffic pattern generation adversarial network, wherein the weekly urban mobile network traffic pattern is a traffic pattern of base station traffic averaged over multiple weeks on a weekly time scale; the generator network may include a second generator network; the discriminator network may include a second discriminator network; the generated traffic sample data may include weekly urban mobile network traffic pattern sample data; and the discrimination result may include a second discrimination result.

[0151] The training module 1030 can input the knowledge graph embedding vectors of multiple base stations transmitting known network traffic into the generator network in the following manner to obtain generated traffic sample data: A second preset traffic pattern matrix is ​​preset, wherein the second preset traffic pattern matrix is ​​an initialized weekly traffic pattern matrix that does not include daily traffic patterns; the knowledge graph embedding vectors and noise of the base stations transmitting known network traffic are input into the second generator network, and the knowledge graph embedding vectors and noise of the base stations transmitting known network traffic are converted into a second projection vector through a multilayer perceptron, layer normalization, and activation function; the second projection vector is multiplied by the second preset traffic pattern matrix, and then processed by the activation function to obtain weekly traffic pattern sample data that does not include daily traffic patterns; based on the weekly traffic pattern sample data that does not include daily traffic patterns and the repeated sample data of the daily urban mobile network traffic pattern sample data, weekly urban mobile network traffic pattern sample data is obtained.

[0152] The training module 1030 can perform adversarial training on real traffic data and generated traffic sample data based on the discriminator network in the following way: Project the real traffic data and the weekly-scale urban mobile network traffic pattern sample data to obtain the real traffic data projection vector and the weekly-scale urban mobile network traffic pattern sample data projection vector, respectively; process the real traffic data projection vector and the weekly-scale urban mobile network traffic pattern sample data projection vector through a multilayer perceptron, layer normalization, and activation function to obtain the second discrimination result output by the second discriminator network; and perform adversarial training on the weekly-scale urban mobile network traffic pattern generative adversarial network based on the second discrimination result and the loss function.

[0153] In an exemplary embodiment of the present invention, the training module 1030 may determine the second preset traffic pattern matrix in the following manner: clustering real traffic data based on a clustering algorithm to obtain multiple second cluster centers; using the multiple second cluster centers as the initial values ​​of each basis vector in the second preset traffic pattern matrix, and obtaining the second preset traffic pattern matrix based on the initial values.

[0154] In an exemplary embodiment of the present invention, the generator network may include a third generator network; the discriminator network may include a third discriminator network; the generated traffic sample data may include continuous-time urban mobile network traffic sample data; and the discrimination result may include a third discrimination result.

[0155] The training module 1030 can obtain generated traffic sample data by inputting the knowledge graph embedding vectors of multiple base stations transmitting known network traffic into the generator network in the following manner: The knowledge graph embedding vectors of base stations transmitting known network traffic and noise are input into the third generator network, and a transformed sequence is obtained through a multilayer perceptron and layer normalization, wherein the sequence length of the transformed sequence is the same as the sequence length of the continuous-time urban mobile network traffic sample data; the transformed sequence is used as the initial value and input into three temporal convolutional neural networks with different kernel sizes, and processed through a multilayer perceptron and a first activation function to obtain continuous-time urban mobile network traffic fluctuation sample data; based on the continuous-time urban mobile network traffic fluctuation sample data and the repeated sample data of the weekly urban mobile network traffic pattern sample data, continuous-time urban mobile network traffic sample data is obtained.

[0156] The training module 1030 can perform adversarial training on real traffic data and generated traffic sample data based on the discriminator network in the following way: input the real traffic data and continuous-time urban mobile network traffic sample data into three temporal convolutional neural networks with different kernel sizes, and process them through a multilayer perceptron and a second activation function to obtain the third discrimination result output by the third discriminator network; based on the third discrimination result and the loss function, perform adversarial training on the continuous-time urban mobile network traffic generation adversarial network.

[0157] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communications bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communications bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a method for generating urban mobile network traffic. This method includes: extracting multiple entities based on urban data information and constructing an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; obtaining knowledge graph embedding vectors corresponding to base stations based on the urban knowledge graph; inputting the knowledge graph embedding vectors and noise into the generator network of a trained generative adversarial network (GAN) model to obtain generated traffic data output by the generator network; the generated traffic data is urban mobile network traffic in the context of urban data information; wherein the GAN model includes urban mobile network traffic pattern GANs at different time scales and continuous-time urban mobile network traffic GANs, both of which are based on inputting knowledge graph embedding vectors of multiple base stations transmitting known network traffic into the generator network to obtain generated traffic sample data, and then performing adversarial training on the real traffic data and generated traffic sample data based on a discriminator network, where the real traffic data is the traffic data transmitted by base stations transmitting known network traffic.

[0158] Furthermore, the logical instructions in the aforementioned memory 1130 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 the present invention, 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 the present invention. 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.

[0159] On the other hand, the present invention 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 execute the urban mobile network traffic generation method provided by the above methods. This method includes: extracting multiple entities based on urban data information and constructing an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; obtaining the knowledge graph embedding vector corresponding to the base station based on the urban knowledge graph; and inputting the knowledge graph embedding vector and noise into the generator network of a trained generative adversarial network model to obtain the generator network. The generated traffic data output by the network is urban mobile network traffic in the context of urban data information. Among them, the generative adversarial network model includes urban mobile network traffic pattern generative adversarial networks at different time scales and continuous-time urban mobile network traffic generative adversarial networks. Both the urban mobile network traffic pattern generative adversarial networks at different time scales and the continuous-time urban mobile network traffic generative adversarial networks are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic. The vectors are input into the generator network to obtain generated traffic sample data. Then, the discriminator network is used to perform adversarial training on the real traffic data and the generated traffic sample data. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

[0160] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the urban mobile network traffic generation method provided by the above methods. This method includes: extracting multiple entities based on urban data information and constructing an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; obtaining knowledge graph embedding vectors corresponding to base stations based on the urban knowledge graph; inputting the knowledge graph embedding vectors and noise into a generator network of a trained generative adversarial network model to obtain generated traffic data output by the generator network; generating traffic... The data refers to urban mobile network traffic within the context of urban data information. The generative adversarial network (GAN) model includes urban mobile network traffic pattern GANs at different time scales and continuous-time urban mobile network traffic GANs. Both the urban mobile network traffic pattern GANs at different time scales and the continuous-time urban mobile network traffic GANs are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic. These vectors are input into the generator network to obtain generated traffic sample data. Then, the discriminator network is used to perform adversarial training on the real traffic data and the generated traffic sample data. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

[0161] 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.

[0162] 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.

[0163] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of the present invention, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.

Claims

1. A method for generating urban mobile network traffic, characterized in that, The method includes: Multiple entities are extracted based on urban data information, and an urban knowledge graph is constructed based on the relationships between entities. The entities include at least base stations, regions, business districts, POIs, POI categories, and brands. Based on the city knowledge graph, the knowledge graph embedding vector corresponding to the base station is obtained; The knowledge graph embedding vector and noise are input into the generator network of the trained generative adversarial network model to obtain the generated traffic data output by the generator network; the generated traffic data is the urban mobile network traffic in the context of the urban data information. The generative adversarial network (GAN) model includes GANs for urban mobile network traffic patterns at different time scales and GANs for urban mobile network traffic over continuous time. Both GANs are based on knowledge graph embedding vectors from multiple base stations transmitting known network traffic, which are input into a generator network to generate traffic sample data. Then, a discriminator network performs adversarial training on real traffic data and the generated traffic sample data. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic. The GANs for urban mobile network traffic patterns at different time scales include those at the daily scale and those at the weekly scale. The GAN for urban mobile network traffic patterns over a continuous long period, where the duration is longer than a week, is generated first at the daily scale, then at the weekly scale, and finally, at the weekly scale, the continuous long-term urban mobile network traffic data is generated. Specifically, the adversarial training based on the discriminator network on real traffic data and generated traffic sample data includes: The real traffic data and the generated traffic sample data are projected to obtain the real traffic data projection vector and the generated traffic sample data projection vector, respectively. The projection vectors of the real traffic data and the projection vectors of the generated traffic sample data are processed by a multilayer perceptron, layer normalization, and activation function to obtain the discrimination result output by the discriminator network. Based on the discrimination results and loss function, adversarial training is performed on the urban mobile network traffic pattern generative adversarial network at different time scales, wherein the Wasserstein distance with added gradient penalty is used as the loss function.

2. The urban mobile network traffic generation method according to claim 1, characterized in that, The step of obtaining the knowledge graph embedding vector corresponding to the base station based on the city knowledge graph specifically includes: Based on the city knowledge graph, the knowledge graph embedding vector corresponding to the base station is obtained through a knowledge graph embedding model.

3. The urban mobile network traffic generation method according to claim 1, characterized in that, The loss function is determined using the following formula: ； in, Represents the loss function; This indicates the discrimination result of the actual traffic data; This indicates the discrimination result of the generated traffic sample data; This indicates the distribution of the actual traffic data; This indicates the distribution of the generated traffic sample data; This represents uniform sampling between real and generated sample pairs; Indicates the discrimination result of the sample; Indicates the distribution of the sample; Indicates the weight of the penalty item; Indicates gradient operation; This represents the expectation operation; This represents the L2 norm operation.

4. The urban mobile network traffic generation method according to claim 1, characterized in that, The city mobile network traffic pattern generation adversarial network at different time scales includes a city mobile network traffic pattern generation adversarial network on a daily scale, wherein the city mobile network traffic pattern on a daily scale is a traffic pattern averaged by base station traffic over multiple days on a daily time scale; the generator network includes a first generator network; the discriminator network includes a first discriminator network; the generated traffic sample data includes city mobile network traffic pattern sample data on a daily scale; and the discrimination result includes a first discrimination result. The knowledge graph embedding vectors based on multiple base stations transmitting known network traffic are input into the generator network to obtain generated traffic sample data, specifically including: A first preset traffic pattern matrix is ​​set, wherein the first preset traffic pattern matrix is ​​an initialized traffic pattern on a daily scale; The knowledge graph embedding vector of the base station transmitting known network traffic and the noise are input into the first generator network, and the knowledge graph embedding vector of the base station transmitting known network traffic and the noise are converted into a first projection vector through a multilayer perceptron, layer normalization and activation function; The first projection vector and the first preset traffic pattern matrix are multiplied together, and then processed by an activation function to obtain the urban mobile network traffic pattern sample data on a daily scale. The adversarial training based on the discriminator network on real traffic data and generated traffic sample data specifically includes: The actual traffic data and the daily urban mobile network traffic pattern sample data are projected to obtain the actual traffic data projection vector and the daily urban mobile network traffic pattern sample data projection vector, respectively. The projection vectors of the real traffic data and the projection vectors of the urban mobile network traffic pattern sample data on a daily scale are processed by a multilayer perceptron, layer normalization and activation function to obtain the first discrimination result output by the first discriminator network. Based on the first discrimination result and the loss function, adversarial training is performed on the city mobile network traffic pattern generative adversarial network measured on a daily scale.

5. The urban mobile network traffic generation method according to claim 4, characterized in that, The first preset traffic pattern matrix is ​​determined in the following way: The real traffic data is clustered based on a clustering algorithm to obtain multiple first cluster centers; The first cluster centers are used as the initial values ​​for each basis vector in the first preset traffic pattern matrix, and the first preset traffic pattern matrix is ​​obtained based on the initial values.

6. The urban mobile network traffic generation method according to claim 4, characterized in that, The city mobile network traffic pattern generation adversarial network at different time scales includes a week-scale city mobile network traffic pattern generation adversarial network, wherein the week-scale city mobile network traffic pattern is a traffic pattern of base station transmission over multiple weeks averaged on a weekly time scale; the generator network includes a second generator network; the discriminator network includes a second discriminator network; the generated traffic sample data includes week-scale city mobile network traffic pattern sample data; and the discrimination result includes a second discrimination result. The knowledge graph embedding vectors based on multiple base stations transmitting known network traffic are input into the generator network to obtain generated traffic sample data, specifically including: A second preset traffic pattern matrix is ​​set, wherein the second preset traffic pattern matrix is ​​an initialized weekly traffic pattern that does not include the daily traffic pattern. The knowledge graph embedding vector of the base station transmitting known network traffic and the noise are input into the second generator network, and the knowledge graph embedding vector of the base station transmitting known network traffic and the noise are converted into a second projection vector through a multilayer perceptron, layer normalization and activation function; The second projection vector and the second preset traffic pattern matrix are multiplied together, and then processed by an activation function to obtain weekly traffic pattern sample data that does not include the daily traffic pattern. Based on the weekly traffic pattern sample data that does not include the daily traffic pattern, and the duplicate sample data of the daily urban mobile network traffic pattern sample data, the weekly urban mobile network traffic pattern sample data is obtained. The adversarial training based on the discriminator network on real traffic data and generated traffic sample data specifically includes: The actual traffic data and the weekly urban mobile network traffic pattern sample data are projected to obtain the actual traffic data projection vector and the weekly urban mobile network traffic pattern sample data projection vector, respectively. The projection vectors of the real traffic data and the projection vectors of the sample data of urban mobile network traffic patterns on a weekly scale are processed by a multilayer perceptron, layer normalization and activation function to obtain the second discrimination result output by the second discriminator network. Based on the second discrimination result and the loss function, adversarial training is performed on the urban mobile network traffic pattern generative adversarial network measured on a weekly scale.

7. The urban mobile network traffic generation method according to claim 6, characterized in that, The second preset traffic pattern matrix is ​​determined in the following way: The real traffic data is clustered based on a clustering algorithm to obtain multiple second cluster centers; The second cluster centers are used as the initial values ​​for each basis vector in the second preset traffic pattern matrix, and the second preset traffic pattern matrix is ​​obtained based on the initial values.

8. The urban mobile network traffic generation method according to claim 6, characterized in that, The generator network includes a third generator network; the discriminator network includes a third discriminator network; the generated traffic sample data includes continuous-time urban mobile network traffic sample data; the discrimination result includes a third discrimination result; The knowledge graph embedding vectors based on multiple base stations transmitting known network traffic are input into the generator network to obtain generated traffic sample data, specifically including: The knowledge graph embedding vector of the base station transmitting known network traffic and noise are input into the third generator network, and the transformed sequence is obtained through a multilayer perceptron and layer normalization, wherein the sequence length of the transformed sequence is the same as the sequence length of the continuous time urban mobile network traffic sample data. The transformed sequence is used as the initial value and input into three temporal convolutional neural networks with different kernel sizes. The data is then processed by a multilayer perceptron and a first activation function to obtain continuous-time urban mobile network traffic fluctuation sample data. The continuous-time urban mobile network traffic fluctuation sample data is obtained based on the continuous-time urban mobile network traffic fluctuation sample data and the repeated sample data of the weekly urban mobile network traffic pattern sample data. The adversarial training based on the discriminator network on real traffic data and generated traffic sample data specifically includes: The real traffic data and the continuous-time urban mobile network traffic sample data are input into three temporal convolutional neural networks with different kernel sizes, and processed by a multilayer perceptron and a second activation function to obtain the third discrimination result output by the third discriminator network. Based on the third discrimination result and the loss function, adversarial training is performed on the continuous-time urban mobile network traffic generation adversarial network.

9. A device for generating urban mobile network traffic, characterized in that, The apparatus is used to implement the urban mobile network traffic generation method according to any one of claims 1 to 8, and the apparatus comprises: Construction module: used to extract multiple entities based on urban data information and construct an urban knowledge graph based on the relationships between entities, wherein the entities include at least base stations, regions, business districts, POIs, POI categories, and brands; The processing module is used to obtain the knowledge graph embedding vector corresponding to the base station based on the city knowledge graph; The training module is used to input the knowledge graph embedding vector and noise into the generator network of the trained generative adversarial network model to obtain the generated traffic data output by the generator network; the generated traffic data is the urban mobile network traffic in the context of the urban data information. The generative adversarial network (GAN) model includes a city mobile network traffic pattern GAN at different time scales and a continuous-time city mobile network traffic GAN. Both the city mobile network traffic pattern GAN at different time scales and the continuous-time city mobile network traffic GAN are based on the knowledge graph embedding vectors of multiple base stations transmitting known network traffic being input into the generator network to obtain generated traffic sample data. Then, adversarial training is performed on the real traffic data and the generated traffic sample data based on the discriminator network. The real traffic data is the traffic data transmitted by the base stations transmitting known network traffic.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the urban mobile network traffic generation method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the urban mobile network traffic generation method as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the urban mobile network traffic generation method as described in any one of claims 1 to 8.

Images