User mobile cellular network data synthesis method and system based on deep generative adversarial network

Through the method of building multi-module joint training in deep generation adversarial networks, the problem of combining spatiotemporal sequences and user attributes in mobile cellular network data synthesis is solved, the reliability and quality of the data are improved, and it is suitable for the modeling requirements of multi-dimensional features.

CN119094402AActive Publication Date: 2024-12-06CENT SOUTH UNIV
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411229668.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-12-06
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

The existing mobile cellular network data synthesis method cannot effectively combine spatio-temporal sequences and user attributes, resulting in poor reliability and quality of synthetic data and cannot meet the modeling needs of multi-dimensional features.

Method used

Using a method based on deep generation adversarial network, a user multi-dimensional attribute synthesis module, a user's traffic sequence synthesis module and a user space base station connection sequence synthesis module are constructed. By jointly training these modules, the user attributes, traffic time series and spatial base station connection sequence are learned. Features, generate high-quality synthetic data.

Benefits of technology

The reliability and quality of synthetic data are improved, so that it has correlation with real data, can effectively capture user behavior patterns and spatial base station correlation, and is suitable for mobile network computing communities that lack real data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094402A_ABST
    Figure CN119094402A_ABST
Patent Text Reader

Abstract

The invention discloses a user mobile cellular network data synthesis method and system based on a deep generative adversarial network, and the method comprises the steps: constructing a user multi-dimensional attribute synthesis module, a user traffic sequence synthesis module and a user space base station connection sequence synthesis module; and carrying out joint training on the user multi-dimensional attribute synthesis module, the user use flow sequence synthesis module and the user space base station connection sequence synthesis module, and synthesizing data by using the trained user multi-dimensional attribute synthesis module, the user use flow sequence synthesis module and the user space base station connection sequence synthesis module. According to the method, the data synthesis models are constructed for the three types of mobile network data with the associated attributes respectively, and then the data synthesis models are subjected to joint training, so that the data synthesized by the trained models not only have the attributes of the data, but also have the relevance with other data, and the data synthesis efficiency is improved. And the reliability and the quality of the synthesized data are greatly improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data synthesis, and in particular to a method and system for synthesizing user mobile cellular network data based on a deep generative adversarial network. Background Art

[0002] Mobile cellular networks are widely deployed in smart cities, which is essential for realizing wireless communication and data transmission of mobile users. At the same time, a large amount of accumulated mobile cellular data plays a vital role in various real-world applications such as cellular traffic prediction, network optimization, urban planning, and human mobility modeling.

[0003] Despite the huge data potential, the inaccessibility of large-scale cellular network data has also brought great challenges to data-driven research in this field. This is because network operators are usually reluctant to share data due to concerns about leaking business secrets or infringing user privacy. Therefore, limited data access has become a huge obstacle for the network system community to advance data-driven research and development.

[0004] In order to fully realize the research value of this data, a promising approach is to generate and share data synthesized from real data. In recent years, researchers have extensively explored data synthesis techniques, including generating single cellular traffic data, radio data, traffic of IoT devices, and network header traffic. In addition, these techniques have been extended to synthesizing various sensor data, such as millimeter wave radar, Doppler radar, and inertial measurement unit data.

[0005] However, due to limitations in data availability and quality, most existing works focus on generating single network traffic and fail to model multi-dimensional feature combinations such as entity attributes and spatial mobility. For example, in a mobile cellular network environment, different user groups show different temporal patterns in application (App) traffic usage, and each user also shows different spatial base station association characteristics.

[0006] Therefore, although most existing methods can meet the needs of synthesizing network and sensor data in different scenarios, they cannot be directly applied to the task of generating mobile cellular network data. This is because mobile cellular network data synthesis requires combining spatiotemporal sequences and user attributes to achieve different downstream applications. Summary of the invention

[0007] The present invention provides a method and system for synthesizing user mobile cellular network data based on a deep generative adversarial network, which is used to solve the technical problems of the existing mobile network data synthesis method in the prior art and the poor reliability and quality of the synthesized data.

[0008] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0009] A method for synthesizing user mobile cellular network data based on a deep generative adversarial network comprises the following steps:

[0010] Constructing a user multi-dimensional attribute synthesis module, wherein the user multi-dimensional attribute synthesis module is used to learn attribute characteristics of real users with different attributes and synthesize user attributes according to the attribute characteristics of the real users;

[0011] Constructing a user traffic sequence synthesis module, which is used to learn the traffic time series characteristics of real traffic sequences of real users with different attributes, and synthesize the traffic time series of users with different attributes according to the traffic time series characteristics; the input of the user traffic sequence synthesis module includes the output of the user multi-dimensional attribute synthesis module;

[0012] Constructing a user space base station connection sequence synthesis module, which is used to learn the spatial connection features of the real space base station connection sequences of real users with different attributes, and synthesize the space base station connection sequences of users with different attributes according to the spatial connection features; the input of the user space base station connection sequence synthesis module includes the output of the user multi-dimensional attribute synthesis module;

[0013] The user multi-dimensional attribute synthesis module, the user traffic sequence synthesis module, and the user space base station connection sequence synthesis module are jointly trained, and the trained user multi-dimensional attribute synthesis module, the user traffic sequence synthesis module, and the user space base station connection sequence synthesis module are used to synthesize data.

[0014] Preferably, the user multi-dimensional attribute synthesis module is a generative adversarial network, including an attribute generator and an attribute discriminator;

[0015] The attribute generator includes an attribute encoder and an attribute decoder. The attribute encoder generates a coding feature h according to the input attribute noise vector. e , the attribute decoder is used to decode the encoded feature h e Decode and synthesize user attribute vector A generated ;

[0016] The attribute discriminator is used to transform the synthetic attribute vector A generated With the real attribute vector A real Compare and determine the synthetic attribute vector A generated and adjust the network parameters of the attribute generator according to the judgment result until the attribute identifier determines the synthetic attribute vector A generated is true.

[0017] Preferably, the user multi-dimensional attribute synthesis module uses one-hot encoding to encode the input, real user attributes to obtain a real encoding vector, and constructs a real attribute vector A based on the real encoding vectors of multiple different users. real ;

[0018] The attribute encoder includes a first Linear layer, a Leaky ReLU layer, and a BatchNorm layer connected in sequence, and the attribute decoder includes a second Linear layer and a Softmax layer connected in sequence, and the output end of the BatchNorm layer is connected to the input end of the second Linear layer;

[0019] The attribute discriminator includes a third Linear layer and a second Leaky ReLU layer connected in sequence; the input of the third Linear layer is a synthetic attribute vector A generated With the real attribute vector A real The stitching vector.

[0020] Preferably, the user uses a traffic sequence synthesis module to generate a conditional timing adversarial network, including: a timing generator and a timing discriminator, wherein the timing generator is used to synthesize the input synthetic attribute vector A generated The time series vector T corresponding to the user's usage traffic generated The timing discriminator is used to transform the time series vector T generated Compared with the actual usage traffic sequence T real Compare and judge the time series vector T according to the comparison result generated and adjust the network parameters of the timing generator according to the judgment result until the timing discriminator judges the synthesized time series vector T generated is true.

[0021] Preferably, the input of the timing generator includes a timing noise vector and a synthetic attribute vector A generated , the timing generator comprises:

[0022] The splicing layer is used to combine the time series noise vector and the synthetic attribute vector A generated To splice;

[0023] TCN block, for capturing the synthetic attribute vector A based on the timing noise vector generated Use the short-term time series characteristics of traffic;

[0024] Transformer encoding block, synthesizing the time series vector T based on the short-term time series features generated ;

[0025] The timing discriminator includes a fourth Linear layer and a third Leaky ReLU layer connected in sequence; the input of the fourth Linear layer is a time series vector T generated Compared with the actual usage traffic sequence T real The stitching vector.

[0026] Preferably, the TCN block satisfies:

[0027] h TCN =Leaky ReLU(Dropout(TCN([T nosie , A generated ])))

[0028] Among them, h TCN is a short-term time series feature; TCN is a time series convolutional network, Dropout is an overfitting function, and LeakyReLU is a rectified linear activation function;

[0029] The Transformer encoding block satisfies:

[0030] T generated =Norm(TransEncoder(h TCN ))

[0031] Among them, TransEncoder is the autoencoder, and Norm is the normalization function;

[0032] and / or

[0033] The actual usage flow sequence T real Obtained through the following methods:

[0034] Obtain usage traffic sequences of multiple real users, standardize the usage traffic sequences, and convert the multiple standardized usage traffic sequences into a three-dimensional matrix, where the dimensions of the three-dimensional matrix include user attributes, traffic time steps, and traffic types.

[0035] Preferably, the user space base station connection sequence synthesis module is a conditional space generative adversarial network, including: a space generator and a space discriminator, the space generator is used to synthesize the user space base station connection sequence according to the input spatial noise vector and the user attribute vector; the space discriminator is used to identify the authenticity of the synthesized user space base station connection sequence, and adjust the network parameters of the space generator according to the identification result until the space discriminator identifies the synthesized user space base station connection sequence as true.

[0036] Preferably, the space generator includes multiple sub-generators, and the space identifier includes multiple sub-identifiers. The multiple sub-generators, multiple sub-identifiers, and multiple users correspond to each other one by one. Each sub-generator is used to capture the correlation characteristics of the space base station connection sequence of its corresponding user, and synthesize the user space base station connection sequence of the corresponding user according to the correlation characteristics of the space base station connection sequence. Each sub-generator is used to identify whether the user space base station connection sequence synthesized by its corresponding sub-generator is true, and adjust the network parameters of the corresponding sub-generator according to the identification result until the corresponding sub-generator identifies that the synthesized user space base station connection sequence is true.

[0037] Preferably, each sub-generator includes an LSTM network, a Linear network, and a Softmax network connected in sequence;

[0038] The LSTM network is used to capture the correlation characteristics of the spatial base station connection sequence of its corresponding user;

[0039] The linear network is used to calculate the probability matrix of the user space base station connection sequence according to the correlation characteristics of the space base station connection sequence;

[0040] The probability matrix satisfies:

[0041]

[0042] Among them, P i represents the i-th user u i The probability matrix of and For user u i The weight matrix and bias vector of the Linear layer; Represents user u i The eigenvector of the correlation of the spatial base station connection sequence; Represents user u i The spatial noise vector, A generated Represents user u i The synthetic user attributes of

[0043] The sub-discriminators all include a fifth Linear layer and a fourth Leaky ReLU layer connected in sequence; the input of the fifth Linear layer is a concatenated vector of a synthetic user space base station connection sequence and a real user space base station connection sequence.

[0044] Preferably, the loss function of the joint training is as follows:

[0045]

[0046] Among them, G ais the attribute generator, D a is the attribute discriminator, L(G a ,D a ) is the loss between the attribute generator and the attribute discriminator, G t is the timing generator, D t is the timing discriminator, L(G t ,D t ) is the loss between the timing generator and the timing discriminator, G s is the space generator, D s is the spatial discriminator, L(G s ,D s ) is the loss between the spatial generator and the spatial discriminator, γ1, γ2, γ3 are hyperparameters that balance the importance of the two losses, is the generated sample, Represents G a , G t , G s Generator distribution, E represents the expected function, represents the gradient descent function.

[0047] A computer system comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0048] The present invention has the following beneficial effects:

[0049] 1. In the present invention, three types of mobile network data with associated attributes are respectively constructed into data synthesis models (i.e., a new framework CellSyn is designed, which consists of three modules based on generative adversarial networks: user App usage traffic sequence synthesis module, user space base station connection sequence synthesis module, user multi-dimensional attribute synthesis module, which respectively learn dynamic time patterns, heterogeneous spatial patterns and complex multi-dimensional relationships with user attributes), and then the data synthesis model is jointly trained, so that the data synthesized by the trained model not only has the attributes of the data itself, but also has the correlation with other data, which greatly improves the reliability and quality of the synthesized data. In addition, the present invention learns and synthesizes large-scale multi-dimensional mobile cellular network data by perceiving the behavior patterns of users associated with mobile networks. The data synthesized by the present invention is of great value to the mobile network computing community that lacks real data.

[0050] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0052] Figure 1 This is the framework diagram of CellSyn, a data synthesis framework for user-related mobile network behavior.

[0053] Figure 2(a) is the framework diagram of the user multi-dimensional attribute synthesis module

[0054] Figure 2(b) is the framework diagram of the user App usage traffic sequence synthesis module

[0055] Figure 2(c) is a diagram of the user space base station connection sequence synthesis module framework

[0056] Figure 3(a) is the CDF graph of the EMD value of the upstream traffic

[0057] Figure 3(b) is the CDF graph of the EMD value of the downlink traffic

[0058] Figure 3(c) is the CDF graph of the EMD value of the base station ID sequence

[0059] Figure 3(d) is the CDF graph of the JSD values ​​of gender

[0060] Figure 3(e) is the CDF graph of the JSD value of age

[0061] Figure 4(a) is the EMD average value of App usage traffic

[0062] Figure 4(b) is the EMD average value of base station ID

[0063] Figure 4(c) is the EMD average value of user attributes

[0064] Figure 5 It is the Acc@3 graph of the next connected base station prediction. DETAILED DESCRIPTION

[0065] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0066] The technical solution of the present application is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Glossary:

[0068] One-Hot Encoding: It is a commonly used data encoding technique used to represent discrete features in the form of binary vectors.

[0069] GAN: It is the abbreviation of Generative Adversarial Network. It is a deep learning model composed of a generator and a discriminator. Its goal is to generate realistic new samples similar to real data by training the generator and the discriminator against each other.

[0070] Linear layer: used to perform linear transformation on the input, thereby introducing nonlinear relationships and learning appropriate weights and bias parameters to achieve more complex function mapping.

[0071] Leaky ReLU layer: is a variant of the rectified linear unit used for nonlinear transformation of the activation function.

[0072] BatchNorm layer: is a commonly used neural network layer, a technique used to standardize input data in deep learning models.

[0073] Softmax layer: It is an activation function commonly used in neural networks, used to convert the output of the network into a probability distribution.

[0074] Temporal Convolutional Network (TCN): is a model based on the convolutional neural network (CNN) structure, which is specifically used for modeling and forecasting time series data.

[0075] Transformer Encoder Block: It is the basic building block in the Transformer model and is used to handle the encoding and feature extraction of input sequence data.

[0076] LSTM network: LSTM (Long Short-Term Memory) network is a variant of recurrent neural network (RNN) used to process and model sequence data, especially sequence data with long-term dependencies.

[0077] In order to solve the challenges of multi-scale time variations, complex multi-dimensional relationships, and heterogeneous user base station association patterns in the process of synthesizing cellular traffic data, this paper proposes a method for synthesizing user mobile cellular network data based on deep generative adversarial networks, such as Figure 1 As shown, it is divided into the following steps:

[0078] S1 In order to synthesize user attributes, App usage flow sequence and space base station connection sequence, the present invention designs three modules to achieve it. The following is a description of each module:

[0079] 1) User multi-dimensional attribute synthesis module: This module uses the user attribute generative adversarial network (AT-GAN) to capture and synthesize different user attributes. As shown in Figure 2(a), the AT-GAN consists of an attribute generator and an attribute discriminator: the attribute generator includes an encoder and a decoder. The encoder includes a linear layer, a leaky ReLU layer, and a batch-norm layer. The decoder includes a linear layer and a softmax layer; the attribute discriminator includes a linear layer and a softmax layer. The attribute generator adopts an encoder-decoder structure. The encoder maps the noise vector to the latent space through a series of layers (including linear layers, leaky ReLU layers, and batch-norm layers) and passes it to the decoder. The decoder decodes the vector of the latent space into user attributes, including anonymous user identifiers (uid), age, gender, and App preference categories. The attribute discriminator is used to judge the authenticity of the generated user attributes.

[0080] 2) User App Usage Traffic Sequence Synthesis Module: This module uses the Conditional Temporal Generative Adversarial Network (CT-GAN) to synthesize App usage traffic sequences based on user attributes. As shown in Figure 2(b), the CT-GAN consists of two parts: a temporal generator and a temporal discriminator. The temporal generator consists of a TCN (Teporal Convolutional Network) and a Transformer encoding block. The TCN is used to capture short-term (e.g., daily and weekly) temporal patterns, and the Transformer encoding block is used to capture long-term (e.g., monthly granularity) dependencies in sequential data. The temporal discriminator consists of a Linear layer and a Leaky ReLU layer. The temporal generator uses user attributes as conditions to synthesize App usage traffic sequences, and the temporal discriminator is used to distinguish between synthesized traffic sequences and real traffic sequences.

[0081] 3) User space base station connection sequence synthesis module: In this module, a conditional spatial generative adversarial network (CS-GAN) is used to generate a personalized base station connection sequence for each user. As shown in Figure 2(c), the CS-GAN consists of n (n represents the number of users in the real data set) spatial generators and n spatial discriminators. Each spatial generator includes an LSTM network, a Linear layer, and a Softmax layer to learn the association between user attributes and spatial base station connection sequences, and synthesize spatial base station connection sequences for each user. Each spatial discriminator includes a Linear layer and a Leaky ReLU layer to judge the authenticity of the synthesized spatial base station connection sequence.

[0082] S2 uses the user multi-dimensional attribute synthesis module to synthesize user attributes

[0083] Based on the observation that user attributes affect the characteristics of App usage traffic, for example, young users show more traffic consumption than older users. Therefore, it is particularly important to design a modeling mechanism for the joint distribution of user spatiotemporal behavior and attributes. In order to effectively meet this challenge, the present invention uses a user multi-dimensional attribute synthesis module to synthesize user attributes, such as age, gender, and other related factors that affect the association between time-series App usage traffic and spatial base stations. The details are as follows:

[0084] S2.1 First, perform attribute deduplication on all users according to user ID. This step ensures that each unique user ID only has its corresponding attribute information such as gender and age. The purpose is to synthesize diverse and non-repetitive user attributes. By removing duplicate user attributes, it is ensured that each user's attribute only appears once in the process of synthesizing user attributes, providing more changes and diversity for the synthesis of the model, so that the model can learn from various combinations of user attributes and synthesize more diverse and realistic data.

[0085] S2.2 User attributes are discrete categorical values, and these values ​​are preprocessed by converting discrete variables into one-hot encoding. Specifically, given a user attribute variable x, whose value range is [1, n], the formula for one-hot encoding is as follows:

[0086] OneHot(x)=[0,0,…,1,…,0]

[0087] The length of the vector is n, and the element at the xth position is 1, and the elements at other positions are 0. For example, if the user attribute variable x has 3 possible values, the formula for one-hot encoding is as follows:

[0088] OneHot(x=1)=[1,0,0]

[0089] OneHot(x=2)=[0,1,0]

[0090] OneHot(x=3)=[0,0,1]

[0091] Through one-hot encoding, the original user attribute variables are represented as multiple binary features, which are convenient for input into the user multi-dimensional attribute synthesis module for processing.

[0092] S2.3 Stack all the user's attribute one-hot encoding vectors row by row to form a matrix vector of real attributes, and define the matrix vector as A real . The number of rows of the matrix corresponds to the number of users n, and the number of columns corresponds to the vector dimension d=n*c after one-hot encoding, where n represents the dimension of each user attribute variable after one-hot encoding, and c represents the number of user attribute variables. The resulting matrix vector has the dimension (n, d), where n represents the number of users, d represents the attribute one-hot encoding dimension of each user, and each row represents the attribute encoding information of a user. The matrix vector is provided as input data to the user multi-dimensional attribute synthesis module for training and sampling to achieve modeling of user behavior, preferences and other characteristics.

[0093] S2.4 Define an attribute noise vector Where R represents the real number domain, n z is the dimension of the noise vector. This vector contains random attribute information and can be used to synthesize a vector that conforms to the user attribute distribution.

[0094] Through the attribute generator of AT-GAN in the user multi-dimensional attribute synthesis module, the attribute noise vector A noise Sample the distribution characteristics of the user attribute latent space and synthesize the attribute vector A that fits the user attribute distribution generated .

[0095] S2.4.1 Specifically, the attribute generator G a Based on the noise vector A noise As input to synthesize different user attributes, G a This task is implemented using an encoder and a decoder. The encoder consists of several stacked layers and outputs an encoded feature h e , defined as follows:

[0096] h e =BatchNorm(Leaky ReLU(W e A noise +b e ))

[0097] Where We and b e They are the weight matrix and bias vector of the Linear layer in the encoder, respectively. Leaky ReLU and BatchNorm are the activation function and batch normalization operation, which help the model converge and generalize faster.

[0098] S2.4.2 Then, the encoding feature h e is fed into the decoder, and the output of the decoder is the synthetic attribute vector Where R represents the real number domain, n α is the dimension of the attribute vector, given by:

[0099] A generated =Softmax(W d h e +b d )

[0100] Where W d and b d are the weight matrix and bias vector of the Linear layer in the decoder respectively.

[0101] S2.4.3 Then, the synthesized attribute vector A generated and the true attribute matrix vector A real Connect them and provide them to the attribute discriminator D a , D a Responsible for distinguishing the synthetic attribute vector from the real attribute matrix vector. It receives the real attribute matrix vector and the attribute vector synthesized by the attribute generator and tries to distinguish their sources. By continuously optimizing the attribute vector synthesized by the attribute generator, it makes it difficult for the attribute discriminator to distinguish the real attribute matrix vector from the synthesized attribute vector, thereby improving the synthesis ability of the attribute generator.

[0102] Attribute Discriminator D a Using a Linear layer and a Softmax layer gives A generated and A real The output is a binary label y∈{0,1}, indicating whether the input attribute vector is real (y=1) or generated (y=0), which are defined as D a (R) and D a (F).

[0103] S2.4.4 Optimize the attribute generator G through continuous iteration a and attribute discriminator D a , the user multi-dimensional attribute synthesis module can learn the distribution characteristics of the user attribute latent space and synthesize the attribute vector A that conforms to this distribution generated .

[0104] S3 via A generated Use the user app usage traffic sequence synthesis module to synthesize the app usage traffic sequence

[0105] S3.1 The data size of the App usage traffic sequence in the real data is inconsistent, including some very large traffic values ​​(such as 20000MB) and some very small traffic values ​​(such as 0.03MB). This data format is not conducive to the training and learning of the user App usage traffic sequence synthesis module. Therefore, the App usage traffic sequence in the real data is first preprocessed to eliminate the data size difference.

[0106] Normalization is a commonly used data preprocessing technique that maps data of different ranges into a unified range to eliminate scale differences between data. When processing App usage traffic sequences in real data, normalization can improve the training effect and generalization ability of the model.

[0107] Common normalization methods include minimum-maximum scaling and standardization. The present invention adopts minimum-maximum scaling, which is a linear transformation method that scales the data to a specified range, usually [0,1]. The specific normalization formula is as follows:

[0108]

[0109] Among them, X is the App usage flow sequence in real data, X min is the minimum value of the sequence, X max is the maximum value of the sequence. In this way, the minimum value in the App usage sequence is mapped to 0, the maximum value is mapped to 1, and other values ​​are proportionally mapped to this range. Min-max scaling is suitable for situations where the data distribution does not deviate significantly from the normal distribution, and it preserves the relative relationship and distribution shape of the real data.

[0110] Through the minimum-maximum scaling process, the App usage traffic sequence in the real data is converted into a standardized sequence distribution, making the data more in line with the requirements of the user App usage traffic sequence synthesis module.

[0111] S3.2 Convert the normalized sequence distribution into a three-dimensional matrix T of shape (n, t, c) real , where n represents the number of users, t represents the time step, and c represents the number of traffic types.

[0112] Through this matrix representation, each user's App usage traffic sequence is represented as a three-dimensional matrix T real, where each t row represents a user's App usage traffic sequence, and each column represents a traffic type. Specifically, for each user, the traffic data used in one month with hourly granularity is arranged in chronological order, and a user's App usage traffic sequence is constructed. Then, upstream traffic and downstream traffic are regarded as different traffic types, respectively, and a matrix with two traffic types is constructed.

[0113] For example, suppose there are 1,000 users, 31 days of data per month, and two types of traffic: upstream and downstream. Then, the user App usage traffic sequence after the minimum-maximum scaling process is represented as a three-dimensional matrix with a shape of (1000, 31*24, 2).

[0114] In this matrix, the first dimension represents the number of users, which is 1000. The second dimension represents the time step, which is 744, representing each hour of each day. The third dimension represents the number of traffic types, which is 2, representing upstream traffic and downstream traffic respectively.

[0115] This matrix representation helps the user App usage traffic sequence synthesis module to learn the relationship between the time correlation and traffic type in the user App usage traffic sequence.

[0116] S3.3 User attribute vector A synthesized using S2 generated ,CT-GAN using the user App usage traffic sequence synthesis module is used to synthesize user App usage traffic sequences.

[0117] Specifically, the temporal generator G in CT-GAN t The TCN block and the Transformer encoding block are combined to learn multi-scale temporal patterns in time series data. t The input is a time series noise vector and A generated The connection vector of the two, where R represents the real number field, n t Represents the dimensions of the time series noise vector.

[0118] S3.3.1 For short-term temporal pattern learning, G t The TCN block is used to capture short-term (e.g., daily and weekly) temporal patterns. TCN is a one-dimensional convolutional network that uses random convolution and dilated convolution. Specifically, let X t ∈R T represents the connected feature vector at time step t, where T represents the total time steps in the time series. First, the time representation X = {X1, X2, ..., X t} is input into the causal convolution to predict the corresponding output at each time step. Then, the dilated convolution component adopts a deep stack of dilated convolutions to capture long-range temporal patterns. t In the design of G, two extended causal convolutional layers are used, and a normalization layer is included as one of the hidden layers in the TCN block. Subsequently, Leaky ReLU and Dropout are applied to prevent overfitting. t The TCN part in can be expressed as:

[0119] h TCN =LeakyReLU(Dropout(TCN([T nosie ,A generated ])))

[0120] S3.3.2 For long-term temporal pattern learning, G t A Transformer encoding block is used to capture long-term (e.g., monthly granularity) dependencies in user App usage sequences. The Transformer encoding block uses a multi-head self-attention mechanism to dynamically weigh the importance of different time steps in the input sequence, enabling the model to capture long-term temporal patterns. represents the vectors output from the TCN block, where T is the length of the input sequence. These vectors are then processed by the Transformer encoder block to produce a time series vector in G represents the hidden state of the Transformer encoding block at time step t. Finally, a normalization function is applied. t The Transformer part in can be expressed as:

[0121] T generated =Norm(TransEncoder(h TCN ))

[0122] S3.4 Temporal Discriminator D in CT-GAN t Take the time series vector T generated and the three-dimensional matrix T real As input, the output is the identification result, that is, a binary label y∈{0,1}, indicating T generated is real (y=1) or synthetic (fake) (y=0), defined as D t (R) and D t (F). By combining G t and D t ,CT-GAN effectively captures the temporal dependencies inherent in the input user App usage traffic sequences.

[0123] S3.5 optimizes the timing generator G through continuous iteration t and timing discriminator D t , CT-GAN can learn the distribution characteristics of App usage traffic sequence and synthesize the time series vector T that conforms to this distribution generated .

[0124] S4 by A generated Use the space base station connection sequence synthesis module to synthesize the user space base station connection sequence

[0125] S4.1 first preprocesses the user space base station connection sequence in the real data, represents the user space base station connection sequence in the real data as a two-dimensional list, each list in the two-dimensional list represents the space base station connection sequence of each user, and performs an embedding operation on the user space base station connection sequence in the real data according to each list in the two-dimensional list, and represents the discrete base station identifier as a continuous, high-dimensional user space base station connection sequence embedding vector The superscript i represents the i-th user, i=1,2,…,n.

[0126] Specifically, a pre-trained embedding layer is used to convert the base station identifiers of the user space base station connection sequence in the real data into corresponding embedding vectors. This embedding layer is trained based on the user space base station connection sequence dataset in the real data, and learns the similarity and semantic information between base stations.

[0127] For each list in the two-dimensional list, we traverse each base station identifier and convert it into a corresponding embedding vector through the embedding layer. In this way, we get a new two-dimensional list, in which each list consists of the embedding vector of the base station identifier.

[0128] For example, suppose that each list in the two-dimensional list represents a user's spatial base station connection sequence, where the base station identifier is an integer value. If the output dimension of the embedding layer is d, then each base station identifier can be input into the embedding layer to obtain an embedding vector of dimension d. After traversing the entire spatial base station connection sequence, a new two-dimensional list is obtained, where each list consists of an embedding vector of the base station identifier, with a shape of (n, (t, d)), where n represents the number of users, t represents the length of each user's spatial base station connection sequence, and d represents the dimension of the embedding vector.

[0129] S4.2 User attribute vector A synthesized using S2 generated, the user space base station connection sequence synthesis module is used to synthesize the user space base station connection sequence. Considering the uniqueness of each user, it is necessary to capture the unique characteristics of each user's base station association. To achieve this goal, the user space base station connection sequence synthesis module focuses on synthesizing user-specific space base station connection sequences to reflect the inherent uniqueness of user mobility patterns. To this end, the CS-GAN in the space base station connection sequence synthesis module generates a user-specific space base station connection sequence for each user u. i A dedicated space generator and space discriminator are assigned, denoted as and And have the same model structure, the subscript i represents the i-th user, i = 1, 2, …, n.

[0130] S4.2.1 In In the example, the input is a concatenated vector consisting of two key elements: the spatial noise vector and the attribute generator G a Synthesized user attribute vector A generated , where R represents the real number domain, n s represents the dimension of the spatial noise vector. First, Use an LSTM network to capture the correlation in the user-space base station connection sequence and output a feature vector containing the correlation of the user-space base station connection sequence It is expressed as follows:

[0131]

[0132] S4.2.2 Then, Use a personalized Linear layer to customize each user u i In particular, to ensure that the synthesized base station ID comes from each user u i The space base station connection sequence is A user-specific Linear layer is designed in this linear layer. A different output dimension is assigned to the Softmax layer, which corresponds to the dimension of user u i The output of this linear layer is a probability matrix representing the synthetic user space base station connection sequence Where R represents the real number domain, n s represents the dimension of the spatial noise vector, BS represents the base station ID that each user has connected to, and P i It is expressed as follows:

[0133]

[0134] in, and For user u i The weight matrix and bias vector of the Linear layer.

[0135] S4.3 Finally, the probability matrix P i and embedding vector is input to the corresponding spatial discriminator middle, A Linear layer and a Leaky ReLU layer are used, whose output produces a label y∈{0,1} indicating whether the synthesized user space base station connection sequence is true (y=1) or false (y=0), and these operations are defined as and

[0136] S4.4 optimizes the space generator through continuous iteration and spatial discriminator CS-GAN can learn the characteristic distribution of user space base station connection sequences and synthesize the probability matrix P of user space base station connection sequences that conform to this distribution. i , through the probability matrix P i To determine the synthesized user space base station connection sequence.

[0137] S5 training and loss function design

[0138] The entire model training process consists of two stages. First, the attribute generator g is trained a and attribute discriminator D a To synthesize the user attribute vector A generated Then, train the timing generator G t and timing discriminator D t Synthetic time series vector T generated In addition, the training space generator G s and the spatial discriminator D s To synthesize the user space base station connection sequence, for each stage, the Wasserstein distance with gradient penalty is used. This method ensures the robustness and stability of the model training process. At each stage, the training goal of the generator is to minimize the loss, while the training goal of the discriminator is to maximize the loss. The loss function is defined as follows:

[0139]

[0140] Among them, γ1, γ2, and γ3 are hyperparameters that balance the importance of the two losses. is the generated sample, Represents G a , G t , Gs The generator distribution of .

[0141] In addition, in this embodiment, a computer system is disclosed, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

[0142] In order to better illustrate the embodiment of the present invention, the following experiments were conducted in this embodiment:

[0143] The embodiments of the present invention are based on two mobile cellular network datasets of a large network operator, covering 9,933 mobile users and 28,524 cellular base stations. The datasets contain two different time intervals: from 2020.12.01 to 2020.12.31 (referred to as "Dataset-201212"), and from 2021.02.01 to 2021.02.28 (referred to as "Dataset-202102").

[0144] The embodiment of the present invention is run on a machine equipped with 2 NVIDIA Tian V GPUs, each GPU is equipped with 12GB memory and 60GB CPU memory. The experiment uses Python3.8 and PyTorch1.10 framework. In order to ensure the training efficiency, the batch size is set to 25, the maximum number of training epochs is set to 200, and the spatiotemporal sequence length is set to 24. The initial learning rates of AT-GAN, CT-GAN and CS-GAN are set to 3e-6, 3e-6 and 8e-7 respectively. These hyperparameters are determined through multiple experiments to achieve the best performance of their respective models. The specific experimental process is:

[0145] 1. Performance indicators

[0146] The present invention uses different evaluation metrics to assess data fidelity, practicality, and privacy, as shown below.

[0147] In terms of data fidelity evaluation, in order to evaluate the fidelity of synthetic data, the Earth Mover Distance (EMD) (also known as Wasserstein-1 distance) is used to evaluate the distribution of the sequence synthetic data and the real data of uplink traffic, downlink traffic and base station ID, and the Jensen-Shannon Divergence (JSD) is used for the distribution of synthetic data and the real data of user ID (uid), age, gender and App category. Since the EMD of different data fields has different scales, the EMD of each field is normalized to [0.1, 0.9]. Specifically, given two probability distributions P and Q of the set X, the EMD is calculated as:

[0148] EMD(P,Q)=inf∑ x,yγ(x,y)·d(x,y)

[0149] Where γ is a parameter representing the quality of each element x in P to each element y in Q, and d(x,y) is the distance between elements x and y in the metric space. JSD is calculated as:

[0150]

[0151] Where x represents an element in the set X, P(x) and Q(x) are the probabilities of x in the distribution P and Q respectively.

[0152] In terms of data utility evaluation, the utility of data is evaluated through two typical downstream applications, namely App traffic usage prediction and next connection base station prediction. For App traffic usage prediction, two common indicators are used: mean absolute error (MAE) and root mean square error (RMSE), which are calculated as follows:

[0153]

[0154]

[0155] where y i and are the true value and predicted value of the i-th prediction, respectively, and n is the total number of predictions. For the prediction evaluation of the next connected base station, Acc@K is used, which measures the ratio of the number of correctly predicted samples to the total number of predicted samples, calculated as:

[0156]

[0157] Where N is the total number of users, if the connected base stations all appear in the Top-K predicted base station set, Acc@K = 1, otherwise Acc@K = 0. In the embodiment of the present invention, K = 3 is selected to calculate the prediction result.

[0158] 2. Comparative Test

[0159] The present invention uses CTGAN, STAN, DoppelGANger, TabDDPM, NetShare and the technical solution of the embodiment of the present invention to compare and evaluate data fidelity and data utility. The results are as follows:

[0160] (1) Regarding data fidelity, we first compare the data fidelity of the CellSyn method with other baseline models. Each sample is the average value calculated for every 500 users among 9933 users. Figures 3(a), 3(b), and 3(c) depict the CDF graphs of the EMD values ​​of uplink traffic, downlink traffic, and base station ID under different models, respectively. The main observations are as follows: First, CellSyn outperforms all baseline models. For example, at the 80% percentile, the EMD values ​​of CTGAN, STAN, DGAN, TabDDPM, and NetShare on uplink traffic are approximately 0.45, 0.46, 0.82, 0.82, and 0.33, respectively. In contrast, the EMD value of the CellSyn scheme is 0.28, and the EMD value of CellSyn on downlink traffic is only 0.18. Through analysis, the performance advantage of CellSyn is attributed to its comprehensive consideration of the dynamic and multi-scale temporal patterns in the time series, which is not carefully considered by other baseline models. Second, for base station ID, CellSyn achieves an EMD value of only 0.47, which is much better than other baseline models, highlighting the ability of CellSyn in effectively modeling spatial base station association patterns.

[0161] Figure 3(d) and Figure 3(e) are the CDFs of the JSD values ​​obtained by different models in terms of gender and age, respectively. From these figures, we can observe the superior performance of the proposed CellSyn scheme. CellSyn still maintains an overall advantage over other baseline models in its ability to accurately capture the distribution of these classification features.

[0162] Figures 4(a), 4(b) and 4(c) show the average performance of the two indicators on Dataset-202012 as well as the error bars. It can be observed that CellSyn outperforms several baseline models in all indicators. In addition, when observing the results of the error bars, it is found that CellSyn exhibits lower deviations among all baseline models, indicating the robustness of its performance. Although the average EMD of base station IDs obtained by NetShare is slightly lower than that of CellSyn, it is observed that the generated base station ID sequence is not user-specific, which may have a negative impact on the data utility of downstream applications (e.g., prediction of the next connected base station).

[0163] (2) Regarding data utility, the data utility of CellSyn is evaluated through two typical downstream applications: App traffic prediction and prediction of the next connected base station. The synthetic data (denoted as B) generated by the CellSyn method proposed in the present invention based on real data (denoted as A) is used. Both real data and synthetic data are sorted in chronological order based on timestamps. The data is then divided into a 70% training set and a 30% test set. The training set consists of early data, while the test set consists of later data. The prediction model is trained on the training set, and its performance is evaluated on the test set. Finally, the performance indicators of the two scenarios are compared: the real data training set (A) and the real data test set (A′) with the synthetic data training set (B) and the real data test set (A′), so that the generalization ability of the model trained on synthetic data can be evaluated.

[0164] In the experiment of App traffic prediction, an important application of mobile cellular network data is to design App traffic prediction algorithms. In this application, we focus on several specific fields in the dataset, including user ID, uplink traffic, downlink traffic, and timestamp. The goal is to predict the App usage traffic of each user in the next time step, which is a key test of the availability of synthetic App usage traffic. In order to evaluate the performance, three commonly used time series prediction models are used for evaluation: LSTM, ConvLSTM, and Transformer.

[0165] The results show that the real data achieves the lowest MAE and RMSE values, that is, the best prediction effect. It is also observed that CellSyn outperforms all other baseline models in all prediction indicators. For example, when using the LSTM model for uplink traffic prediction on Dataset-202012, the MAE values ​​of synthetic data from CTGAN, DGAN, STAN, TabDDPM, and NetShare are 615.32, 1009.92, 4547.32 and 91.76, 683.6, respectively, while the MAE of synthetic data from CellSyn is only 75.21. Similar results can also be observed in downlink traffic prediction. For example, when using the ConvLSTM model on Dataset-202102, the RMSE values ​​of synthetic data from CTGAN, DGAN, STAN, TabDDPM, and NetShare are 61.5, 272.1, 1737.1, 15.3, and 48.60, respectively, while the RMSE of CellSyn synthetic data is closest to the real data, which is 10.8. The above observations show that CellSyn can effectively learn the temporal patterns of App usage traffic.

[0166] In the experiment of next base station prediction, the prediction of next base station is another key application that relies on mobile cellular network data. The main goal of this task is to predict the next base station that the user may connect to based on past mobility patterns and historical data. To evaluate the performance on synthetic data, four prediction models are used: LSTM, CNN, Attention, and GRU+Attention.

[0167] Figure 5 The Acc@3 of each prediction algorithm on the Dataset-202012 dataset is presented. It should be noted that since the prediction results of the synthetic data generated by CTGAN and DGAN are less than 0.02, their prediction results are not shown here. It can be observed that the real data achieves the best indicator value, however, it is found that the synthetic data generated by CellSyn outperforms all other baseline models in all prediction indicators. For example, when using the Attention network on Dataset-202012, the Acc@3 scores of STAN, TabDDPM, and NetShare are 0.2, 0.21, and 0.3, respectively, while CellSyn reaches 0.62, which is very consistent with the real data. This is due to the customized design of CellSyn on the spatial base station sequence generation module, which is able to learn the correlation of user base station sequences.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for synthesizing user mobile cellular network data based on deep generative adversarial networks, characterized in that: The following steps are involved: Constructing a user multi-dimensional attribute synthesis module, wherein the user multi-dimensional attribute synthesis module is used to learn attribute characteristics of real users with different attributes and synthesize user attributes according to the attribute characteristics of the real users; Constructing a user traffic sequence synthesis module, which is used to learn the traffic time series characteristics of real traffic sequences of real users with different attributes, and synthesize the traffic time series of users with different attributes according to the traffic time series characteristics; the input of the user traffic sequence synthesis module includes the output of the user multi-dimensional attribute synthesis module; Constructing a user space base station connection sequence synthesis module, which is used to learn the spatial connection features of the real space base station connection sequences of real users with different attributes, and synthesize the space base station connection sequences of users with different attributes according to the spatial connection features; the input of the user space base station connection sequence synthesis module includes the output of the user multi-dimensional attribute synthesis module; The user multi-dimensional attribute synthesis module, the user traffic sequence synthesis module, and the user space base station connection sequence synthesis module are jointly trained, and the trained user multi-dimensional attribute synthesis module, the user traffic sequence synthesis module, and the user space base station connection sequence synthesis module are used to synthesize data.

2. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 1, characterized in that: The user multi-dimensional attribute synthesis module is a generative adversarial network, including an attribute generator and an attribute discriminator; The attribute generator includes an attribute encoder and an attribute decoder. The attribute encoder generates a coding feature h according to the input attribute noise vector. e , the attribute decoder is used to decode the encoded feature h e Decode and synthesize user attribute vector A generated ; The attribute discriminator is used to transform the synthetic attribute vector A generated With the real attribute vector A real Compare and determine the synthetic attribute vector A generated and adjust the network parameters of the attribute generator according to the judgment result until the attribute identifier determines the synthetic attribute vector A generated is true.

3. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 2 is characterized in that: The user multi-dimensional attribute synthesis module uses one-hot encoding to encode the input, real user attributes to obtain a real encoding vector. And construct the real attribute vector A based on the real encoding vectors of multiple different users real ; The attribute encoder includes a first Linear layer, a Leaky ReLU layer, and a BatchNorm layer connected in sequence, and the attribute decoder includes a second Linear layer and a Softmax layer connected in sequence, and the output end of the BatchNorm layer is connected to the input end of the second Linear layer; The attribute discriminator includes a third Linear layer and a second Leaky ReLU layer connected in sequence; the input of the third Linear layer is a synthetic attribute vector A generated With the real attribute vector A real The stitching vector.

4. The method for synthesizing user mobile cellular network data based on a deep generative adversarial network according to any one of claims 1 to 3, characterized in that: The user uses the traffic sequence synthesis module to generate an adversarial network for the conditional timing, including: a timing generator and a timing discriminator, wherein the timing generator is used to synthesize the input synthetic attribute vector A generated The time series vector T corresponding to the user's usage traffic generated The timing discriminator is used to transform the time series vector T generated Compared with the actual usage traffic sequence T real Compare and judge the time series vector T according to the comparison result generated and adjust the network parameters of the timing generator according to the judgment result until the timing discriminator judges the synthesized time series vector T generated is true.

5. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 4 is characterized in that: The input of the timing generator includes a timing noise vector and a synthetic attribute vector A generated , the timing generator comprises: The splicing layer is used to combine the time series noise vector and the synthetic attribute vector A generated To splice; TCN block, for capturing the synthetic attribute vector A based on the timing noise vector generated Use the short-term time series characteristics of traffic; Transformer encoding block, synthesizing the time series vector T based on the short-term time series features generated ; The timing discriminator includes a fourth Linear layer and a third Leaky ReLU layer connected in sequence; the input of the fourth Linear layer is a time series vector T generated Compared with the actual usage traffic sequence T real The stitching vector.

6. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 5, characterized in that: The TCN block satisfies: h TCN =Leaky ReLU(Dropout(TCN([T nosie ,A generated ]))) Among them, h TCN is a short-term time series feature; TCN is a time series convolutional network, Dropout is an overfitting function, and Leaky ReLU is a rectified linear activation function; The Transformer encoding block satisfies: T generated =Norm(TransEncoder(h TCN )) Among them, TransEncoder is the autoencoder, and Norm is the normalization function; and / or The actual usage flow sequence T real Obtained through the following methods: Obtain usage traffic sequences of multiple real users, standardize the usage traffic sequences, and convert the multiple standardized usage traffic sequences into a three-dimensional matrix, where the dimensions of the three-dimensional matrix include user attributes, traffic time steps, and traffic types.

7. The method for synthesizing user mobile cellular network data based on a deep generative adversarial network according to any one of claims 1 to 3, characterized in that: The user space base station connection sequence synthesis module is a conditional space generative adversarial network, including: a space generator and a space discriminator, the space generator is used to synthesize the user space base station connection sequence according to the input spatial noise vector and the user attribute vector; the space discriminator is used to identify the authenticity of the synthesized user space base station connection sequence, and adjust the network parameters of the space generator according to the identification result until the space discriminator identifies the synthesized user space base station connection sequence as true.

8. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 7, characterized in that: The space generator includes multiple sub-generators, and the space identifier includes multiple sub-identifiers. The multiple sub-generators, multiple sub-identifiers, and multiple users correspond to each other one by one. Each sub-generator is used to capture the correlation characteristics of the space base station connection sequence of its corresponding user, and synthesize the user space base station connection sequence of the corresponding user according to the correlation characteristics of the space base station connection sequence. Each sub-generator is used to identify whether the user space base station connection sequence synthesized by its corresponding sub-generator is true, and adjust the network parameters of the corresponding sub-generator according to the identification result until the corresponding sub-generator identifies that the synthesized user space base station connection sequence is true.

9. The method for synthesizing user mobile cellular network data based on deep generative adversarial network according to claim 7, characterized in that: Each sub-generator includes an LSTM network, a Linear network, and a Softmax network connected in sequence; The LSTM network is used to capture the correlation characteristics of the spatial base station connection sequence of its corresponding user; The linear network is used to calculate the probability matrix of the user space base station connection sequence according to the correlation characteristics of the space base station connection sequence; The probability matrix satisfies: Among them, P i represents the i-th user u i The probability matrix of and For user u i The weight matrix and bias vector of the Linear layer; Represents user u i The eigenvector of the correlation of the spatial base station connection sequence; Represents user u i The spatial noise vector, A generated Represents user u i The synthetic user attributes of The sub-discriminators all include a fifth Linear layer and a fourth Leaky ReLU layer connected in sequence; the input of the fifth Linear layer is a concatenated vector of a synthetic user space base station connection sequence and a real user space base station connection sequence. The loss function of the joint training is as follows: Among them, G a is the attribute generator, D a is the attribute discriminator, L(G a , D a ) is the loss between the attribute generator and the attribute discriminator, G t is the timing generator, D t is the timing discriminator, L(G t , D t ) is the loss between the timing generator and the timing discriminator, G s is the space generator, D s is the spatial discriminator, L(G s , D s ) is the loss between the spatial generator and the spatial discriminator, γ1, γ2, γ3 are hyperparameters that balance the importance of the two losses, is the generated sample, Represents G a , G t , G s Generator distribution, E represents the expected function, represents the gradient descent function.

10. A computer system 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 computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Citation Information

Patent Citations

  • Sensor data generation model and method based on conditional generative adversarial network

    CN112884076A

  • Multi-party multi-class data generation method and system oriented to longitudinal federated learning

    CN114648099A

  • Mobile user network flow generation method and system based on adversarial network

    CN116208655A

  • Power grid time sequence data generation model system based on CDoubleGAN

    CN117009757A

  • Subsynchronous oscillation data generation method based on multivariable time sequence generative adversarial network

    CN117574760A