A mobile user network traffic generation method and system based on a generative adversarial network

Abstract

The application provides a mobile user network traffic generation method and system based on an adversarial network, comprising: obtaining mobile user behavior feature data, and extracting behavior features from the mobile user behavior feature data; based on the extracted behavior features, simulating multiple network traffic modes and mode switching behaviors of mobile users through multiple pre-stored generators to generate mobile user network traffic sequences; judging the authenticity of the mobile user network traffic sequences through a pre-stored discriminator to generate a judgment result; according to the judgment result, training a preset adversarial network model by using a loss function of the discriminator and a loss function of the generator; and training the adversarial network model to convergence to generate a normalized network traffic sequence dataset. The application solves the problems of unreliable authenticity and low resolution of the generated mobile user network traffic.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network traffic processing, and particularly relates to a mobile user network traffic generation method and system based on an adversarial network. BACKGROUND

[0002] With the rapid expansion of the mobile network user group, collecting and utilizing mobile user network traffic becomes more and more important and popular. Specifically, mobile user traffic can be used in various application scenarios, including user behavior modeling, malware detection, ad fraud detection, sociological intervention, website fingerprinting, etc. In addition, mobile user traffic plays an important role in promoting the development of mobile networks, and can promote the perfection of network planning and optimization. At the same time, there are two difficulties in obtaining mobile user network traffic data: on the one hand, large-scale mobile user network traffic is usually collected and stored by network operators, service providers, etc., and considering the data security and user privacy risks, these data are usually anonymized and added with a large amount of noise before use, resulting in a reduction in data quality and data availability; on the other hand, the mobile user network traffic dataset collected through crowdsourcing and other ways is small in size and limited in the application and research scenarios that can be supported, and expanding the dataset size faces extremely high costs. Therefore, the present application mainly studies a mobile user network traffic generation model to provide a large amount of high-quality mobile user network traffic data at a low cost on the premise of avoiding data security problems and user privacy risks.

[0003] Mobile user network traffic generation belongs to the field of network traffic generation. Currently, there are few machine learning methods for network traffic generation of mobile users, but some methods for other types of network traffic generation can be used for reference. These methods mainly analyze it as a sub-problem of time series generation problem and generate network traffic feature sequence in a period of time. Common traditional methods mainly include autoregressive model (AR model), hidden Markov model (HMM model), recurrent neural network model (RNN model), and generative adversarial network model (GAN model). Among them, the autoregressive model and the hidden Markov model generate the sequence content value at the current time according to the sequence content of the previous T time, without considering the global time sequence features; the recurrent neural network model can learn global time sequence features, but the performance decreases as the traffic duration increases; the original generative adversarial network model is difficult to guarantee the diversity and authenticity of the generated samples when facing complex and variable time sequence. Through observation, there are two challenges in mobile user network traffic: on the micro individual level, a large number of mobile user network traffics show the characteristics of mixed traffic patterns, for example, some user traffics show periodic patterns in units of days, while some user traffics do not show periodicity; for example, some user traffics switch from non-periodic patterns to periodic patterns, while some user traffics always show periodic patterns in units of days and do not switch traffic patterns. On the macro group level, a large number of mobile user network traffics show strong clustering characteristics, that is, the aggregated traffic of users belonging to different categories shows different behaviors. SUMMARY

[0004] The application provides a mobile user network traffic generation method and system based on an adversarial network to solve the problems of unreliable authenticity and low resolution of the generated mobile user network traffic.

[0005] The application provides a mobile user network traffic generation method based on an adversarial network, comprising:

[0006] Obtaining mobile user behavior feature data, extracting behavior features from the mobile user behavior feature data;

[0007] Based on the extracted behavior features, simulating multiple network traffic patterns and mode switching behaviors of mobile users through a plurality of pre-stored generators to generate mobile user network traffic sequences;

[0008] The authenticity of the mobile user network traffic sequence is judged by a pre-stored discriminator, and a judgment result is generated;

[0009] According to the judgment result, the preset adversarial network model is trained by using the loss function of the discriminator and the loss function of the generator;

[0010] The adversarial network model is trained to convergence, and a normalized network traffic sequence dataset is generated.

[0011] According to the mobile user behavior feature data, the behavior features are extracted from the mobile user behavior feature data, and the method specifically comprises:

[0012] The use of mobile application features is obtained from network data packets, and the time mobile behavior features are obtained from the base station addresses accessed by the user when connecting to the mobile network.

[0013] The use of mobile application features is obtained from network data packets, and the time mobile behavior features are obtained from the base station addresses accessed by the user when connecting to the mobile network.

[0014] According to the mobile user behavior feature data, the behavior features are extracted from the mobile user behavior feature data, and the method specifically comprises:

[0015] The generator includes a network traffic pattern generator and a network traffic mode switching behavior generator;

[0016] The network traffic pattern generator simulates the mobile user network traffic pattern based on the extracted behavior features;

[0017] The network traffic mode switching behavior generator learns mode switching based on the simulated mobile user network traffic pattern, and generates a mobile user network traffic sequence.

[0018] According to the mobile user behavior feature data, the behavior features are extracted from the mobile user behavior feature data, and the method specifically comprises:

[0019] The authenticity of the mobile user network traffic sequence is judged by a pre-stored discriminator, and a judgment result is generated;

[0020] The authenticity of the mobile user network traffic sequence is judged by a pre-stored discriminator, and a judgment result is generated;

[0021] According to the mobile user network traffic generation method based on the adversarial network provided by the application, the preset adversarial network model is trained by using the loss function of the discriminator and the loss function of the generator according to the judgment result, and specifically includes the following steps:

[0022] The loss function of the discriminator and the loss function of the generator both include an adversarial loss function, a clustering loss function and a regularization term.

[0023] The adversarial network model is alternately trained by the loss function of the generator and the loss function of the discriminator.

[0024] According to the mobile user network traffic generation method based on the adversarial network provided by the application, the preset adversarial network model is trained to convergence to generate a normalized network traffic sequence data set, and specifically includes the following steps:

[0025] After the adversarial network model is trained to convergence, the real traffic sequence sample is input into the adversarial network model.

[0026] The adversarial network model outputs the normalized network traffic sequence data set.

[0027] The application further provides a mobile user network traffic generation system based on an adversarial network, and the system includes the following modules:

[0028] The behavior feature acquisition module is used for acquiring mobile user behavior feature data and extracting behavior features from the mobile user behavior feature data.

[0029] The mobile user network traffic sequence generation module generates mobile user network traffic sequences by simulating multiple network traffic modes and mode switching behaviors of the mobile user through multiple pre-stored generators based on the extracted behavior features.

[0030] The judgment module is used for judging the authenticity of the mobile user network traffic sequences by using a pre-stored discriminator to generate a judgment result.

[0031] The training module is used for training a preset adversarial network model by using the loss function of the discriminator and the loss function of the generator according to the judgment result.

[0032] The data generation module is used for training the adversarial network model to convergence to generate a normalized network traffic sequence data set.

[0033] The application further provides an electronic device including a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor implements the mobile user network traffic generation method based on the adversarial network as described above when executing the program.

[0034] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the mobile user network traffic generation method based on the adversarial network.

[0035] The application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the mobile user network traffic generation method based on the adversarial network.

[0036] The application provides a mobile user network traffic generation method and system based on an adversarial network, which models multiple network traffic modes of mobile users and mode switching behaviors thereof through multiple generators, and then clusters the mobile user traffic while judging the authenticity of the mobile user traffic by using a discriminator, and finally supervises the model from the micro and macro levels at the same time by using multiple loss functions, so as to generate more real and effective mobile user network traffic, and the spatial coverage range is large enough and the resolution is improved. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0038] Figure 1 is one of the flowcharts of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0039] Figure 2 is another flowchart of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0040] Figure 3 is a third flowchart of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0041] Figure 4 is a fourth flowchart of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0042] Figure 5 is a fifth flowchart of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0043] Figure 6 is a sixth flowchart of the mobile user network traffic generation method based on the adversarial network provided by the application;

[0044] Figure 7 This is a schematic diagram of the module connections of a mobile user network traffic generation system based on an adversarial network provided by the present invention;

[0045] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0046] Figure 9 This invention provides a multi-scale, multi-layered generative adversarial network overall structure diagram for generating mobile user network traffic;

[0047] Figure 10 This is a structural diagram of the mobile user network traffic pattern generator provided by the present invention;

[0048] Figure 11 This is a structural diagram of the mobile user network traffic mode switching behavior generator provided by the present invention;

[0049] Figure 12 This is a structural diagram of the mobile user network traffic discriminator provided by the present invention;

[0050] Figure label:

[0051] 110: Behavioral feature acquisition module; 120: Mobile user network traffic sequence generation module; 130: Judgment module; 140: Training module; 150: Data generation module; 810: Processor; 820: Communication interface; 830: Memory; 840: Communication bus. Detailed Implementation

[0052] 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.

[0053] This invention first defines the concept of user traffic sequence:

[0054] Definition 1 (User Traffic Sequence): Given a mobile user U, his traffic sequence is defined as a sequence of network traffic values ​​generated by him within multiple consecutive equal time intervals. This sequence can be represented as... Where v t This represents the network traffic generated by user U during the t-th time period, where T represents the total number of time periods.

[0055] This chapter focuses primarily on changes in mobile user traffic, and therefore normalizes the user traffic sequences:

[0056] Definition 2 (Normalized User Traffic Sequence): Given a mobile user U, his normalized traffic sequence is defined as X = {x t} T t＝1 = {v t / ||V||2} t＝1 T where x t denotes the normalized network traffic of user U in the t-th time period, and ||V||2is the 2-norm of user traffic sequence V.

[0057] On this basis, this chapter defines the aggregated normalized user traffic sequence:

[0058] Definition 3 (Aggregated Normalized User Traffic Sequence): Given a set of mobile users {U i} N i＝1 and their normalized user traffic sequences {X i} N i＝1 The aggregated normalized user traffic sequence of this set of users is defined as A = {a t} T t＝1 = {∑ N i＝1 x i t} T t＝1 where a t denotes the aggregated normalized user traffic of this set of users in the t-th time period.

[0059] Further, the behavior of mobile users using network traffic can change significantly over time, resulting in significant changes in the normalized user traffic sequence. Therefore, this chapter defines the traffic pattern and traffic pattern switching behavior in the normalized user traffic sequence:

[0060] Definition 4 (Traffic Pattern): A traffic pattern is defined as a segment of normalized user traffic sequence with a consistently constant time series pattern, which can be represented as P = {p {} t} T t＝1 where p t denotes the normalized user traffic in the t-th time period.

[0061] Definition 5 (Traffic Pattern Switching Behavior): Given a mobile user U, his traffic pattern switching behavior is defined as S = {s t} T t＝1 where s t is the one-hot encoding representing the traffic pattern followed by user U in the t-th time period.

[0062] The application is described below Figures 1-6 The application provides a mobile user network traffic generation method based on an adversarial network, which comprises the following steps:

[0063] S100, acquiring mobile user behavior feature data, and extracting behavior features from the mobile user behavior feature data;

[0064] S200, based on the extracted behavior features, simulating multiple network traffic modes and mode switching behaviors of mobile users through multiple pre-stored generators, and generating mobile user network traffic sequences;

[0065] S300, judging the authenticity of the mobile user network traffic sequences through a pre-stored discriminator, and generating a judgment result;

[0066] S400, according to the judgment result, training a preset adversarial network model by using a loss function of the discriminator and a loss function of the generator;

[0067] S500, training the adversarial network model to convergence, and generating a normalized network traffic sequence dataset.

[0068] The application first extracts the space-time behavior features of mobile users, and then designs a multi-scale and multi-level generative adversarial network model for network traffic generation, as shown in the accompanying drawings. Figure 9 The adversarial network model first models multiple network traffic modes and mode switching behaviors of mobile users from a micro perspective by using multiple generators, and then clusters the mobile user traffic while judging the authenticity of the mobile user traffic from a macro perspective by using a discriminator, and finally supervises the model from the micro and macro perspectives by using multiple loss functions to generate more real and effective mobile user network traffic.

[0069] The application first extracts the space-time behavior features of mobile users, and then designs a multi-scale and multi-level generative adversarial network model for network traffic generation, as shown in the accompanying drawings.

[0070] S101, extracting the mobile application usage features and the time mobile behavior features of each user from the mobile user behavior feature data;

[0071] S102, the mobile application usage features are acquired from network data packets, and the time mobile behavior features are acquired from base station addresses accessed by the user when connecting to the mobile network.

[0072] Different mobile users have different behavior characteristics, and these behavior characteristics can have an impact on the network traffic generated by the mobile users. In order to use these behavior characteristics to assist in the generation of user network traffic, the application extracts the mobile application usage features and the space-time mobile behavior features of each user from the mobile user network traffic dataset.

[0073] First, when the user uses the mobile application, the application-related information is contained in the corresponding network data packet. Generally, according to the "User Agent" field in the data packet, the corresponding mobile application type can be obtained, for example, according to the "User Agent" field in the data packet as "QQ / 8.2.17724.201", it can be inferred that the mobile application used by the user at this time is a certain version of QQ. According to the "User Agent" field mainly involved in the data set, the application corresponding to the data packet can be divided into 21 categories: food, hotel, shopping, leisure and entertainment, sports and fitness, school, scenic spot, tourism development, finance, office building, company, business district, factory, industrial zone, science and technology park, economic development zone, high-tech development zone, residential area, life service, town, and village. This section calculates the probability of the network traffic generated by each user belonging to each category as a feature of the user using the mobile application.

[0074] At the same time, the data set contains the base station ID accessed by each user when performing mobile network connection. According to the time of accessing the base station and the geographical location of the base station, the spatio-temporal behavior information of the user can be obtained. This section calculates the total number of base stations accessed by each user, the total number of base stations switched, the daily average number of base stations accessed and the variance, the daily average number of base stations switched and the variance, the turning radius, the total moving distance, the daily average moving distance and the variance, the average time of staying at each base station and the variance, as the spatio-temporal mobile behavior features of the user. Since there are some base stations lacking latitude and longitude information in the data, this part of the base station does not participate in the calculation, and if the base stations involved by the user all lack latitude and longitude information, the relevant statistics are marked as -1.

[0075] Based on the extracted behavior features, a plurality of generators are pre-stored to simulate a plurality of network traffic patterns and mode switching behaviors of the mobile user, and a network traffic sequence of the mobile user is generated, specifically including:

[0076] S201, the generator includes a network traffic pattern generator and a network traffic mode switching behavior generator;

[0077] S202, the network traffic pattern generator simulates a network traffic pattern of the mobile user based on the extracted behavior features;

[0078] S203, the network traffic mode switching behavior generator learns mode switching based on the simulated network traffic pattern of the mobile user, and generates a network traffic sequence of the mobile user.

[0079] In generating mobile user network traffic patterns, both long-term and short-term temporal changes in network traffic need to be considered, thus the generator adopts bidirectional long short term memory network (BiLSTM) and self-attention mechanism, as shown in Figure 10 Long short term memory (LSTM) is a kind of RNN structure commonly used to remember historical information, and the output sequence is obtained through step-by-step circulation. In a typical LSTM, the output of each step is affected by the input of this step and the previous state, while the BiLSTM network can make the output of each step contain the information of the previous state and the subsequent state in the sequence by reversing the circulation of the LSTM network. In specific applications, first, the feature vector obtained is mapped to the initial cell state c0 of the BiLSTM network through a linear layer, and the noise vector is mapped to the initial hidden layer state h0 of the BiLSTM network through another linear layer, and the BiLSTM network is initialized. Next, the above-mentioned feature vector and noise vector are mapped to the input of each step of the BiLSTM network through a linear layer. Since the influence of the time sequence change that the BiLSTM network can capture decreases with the increase of the number of cycles, a segment of network traffic sequence is generated at each step of the BiLSTM network cycle to reduce the decline in the performance of the BiLSTM network.

[0080] The self-attention mechanism is adopted to capture the temporal association within the corresponding segment of network traffic sequence at each step of the BiLSTM network cycle as follows:

[0081]

[0082] where h k is the hidden layer state corresponding to the kth step of the BiLSTM network cycle, R K , R Q and R V are the key, query and value representation vectors obtained through linear layer and ReLU activation function projection, P k ^ is a segment of network traffic sequence generated at the kth step through the softmax activation function. After Q steps of circulation, the complete network traffic sequence generated can be obtained as where P k ^ = {p t} k·Q t＝(k-1)·Q+1 . Finally, N simulated mobile user network traffic patterns can be obtained by using N mobile user network traffic pattern generators

[0083] As Figure 11As shown, the present application proposes a linear flow mode switching behavior learning unit to memorize and update the mobile user network flow mode switching behavior, and outputs a one-hot encoding from the continuous state vector saved in the switching behavior learning unit through a Gumbel-Softmax layer, while ensuring the continuity of the model. In specific applications, first, the feature vector obtained is mapped to the initial state of the switching behavior learning unit through a linear layer, and then the Gumbel-Softmax layer outputs the one-hot encoding of the flow mode followed by the user in the first step. The output of each step will be used as the input of the next step, and the mode switching line is updated through a loop. The complete mobile user network flow mode switching behavior generated for the learning unit is as follows:

[0084]

[0085] Z represents a noise vector, condition represents the obtained feature vector, m0 represents the initial state of the switching behavior learning unit, and m t represents the state of the switching behavior learning unit at the t-th step, represents the one-hot encoding of the flow mode followed by the user generated at the t-th step, and SL represents the linear flow mode switching behavior learning unit.

[0086] In the linear flow mode switching behavior learning unit, the flow mode switching state followed by the user is represented by a matrix, where each element records the two-way switching between each flow mode occurring in the early stage. Gumbel-Softmax is a continuous distribution that can be used to approximate the sampling of a multinomial distribution. By applying the softmax function to the Gumbel-Max method

[23] implementation. The Gumbel-Max method can sample t from a multinomial distribution subject to probability as follows:

[0087]

[0088] m i t is the i-th element of m t , g i is an independent and identically distributed sample from Gumbel(0, 1), and u i is an independent and identically distributed sample from Uniform(0, 1). After replacing the non-differentiable argmax function in the above formula with the differentiable softmax function, the Gumbel-Max method can be represented as:

[0089]

[0090] where is the i-th element of N represents the dimension of N is equal to the number of user network traffic patterns. Through the above formula, by approaching the softmax temperature parameter τ to 0, the becomes one-hot encoding, which facilitates the gradient update of the model through reparameterization.

[0091] Finally, the generated N simulated mobile user network traffic patterns {P^i}Ni=1 and the current generated mobile user network traffic pattern switching behavior are element-wise multiplied to obtain the generated user network traffic sequence where ⊙ denotes the Hadamard product.

[0092] The authenticity of the mobile user network traffic sequence is judged by the pre-stored discriminator to generate a judgment result, which specifically includes:

[0093] S301, judging the authenticity of the mobile user network traffic sequence by the discriminator;

[0094] S302, if the mobile user network traffic sequence is judged as real data, then performing adversarial training

[0095] There are large differences between the aggregated traffic of different types of user groups at the macro level. In this section, a discriminator is designed based on the Temporal Convolutional Networks (TCN), which can give the true or false discrimination result of the traffic sequence sample to be discriminated and the category of the sample at the same time, thereby breaking the barrier between individual and group traffic.

[0096] TCN is a one-dimensional convolutional neural network that uses a widened causal convolution to calculate features and transmit them to the next layer at each layer. Each time, the subsequence used to calculate the features is determined by the configuration of the convolution kernel. Therefore, using a TCN unit with a convolution kernel of size in time units of days, weeks, etc. can help the model capture the time sequence correlation of the subsequence within the previous day or week between layers. As shown in Figure 12 The real user network traffic sequence X and the generated user network traffic sequence are first input into two TCN units, and then the discrimination and clustering results are given via a linear layer and a softmax layer, as follows:

[0097] D(X)=(Df(X),Dc(X)),Dc(X)=[Dc1(X),Dc2(X),...,DcM(X)],

[0098] D f (X) represents the probability that the traffic sequence sample X is judged as a false sample, and D ci (X) represents the probability that X is judged to be a real sample and belongs to the i-th category. At the same time, each traffic sequence sample is either judged to be a fake sample or a real sample of any one category, that is, D f (X)+∑ M i＝1 D c i (X) = 1.

[0099] According to the judgment result, the preset adversarial network model is trained by using the loss function of the discriminator and the loss function of the generator, and specifically includes:

[0100] S401, the loss function of the discriminator and the loss function of the generator both include an adversarial loss function, a clustering loss function and a regularization term;

[0101] S402, the adversarial network model is alternately trained by the loss function of the generator and the loss function of the discriminator.

[0102] The adversarial network model is trained to converge to generate a normalized network traffic sequence dataset, specifically including:

[0103] S501, after the adversarial network model is trained to converge, a real traffic sequence sample is input into the adversarial network model;

[0104] S502, the adversarial network model outputs a normalized network traffic sequence dataset.

[0105] The discriminator loss function includes three parts: an adversarial loss function, a clustering loss function and a regularization term. As follows:

[0106]

[0107] Where X represents a real traffic sequence sample, pr represents the probability distribution of real data, represents a generated traffic sequence sample, G represents a generator, Df represents the probability that the sample given by the discriminator is a fake sample, D c (X) represents the category of the traffic sequence sample X given by the discriminator, C i represents the i-th category, O i represents the clustering center of C i , N represents the total number of categories, D w represents the weight matrix of the linear layer for output in the discriminator, D w ′ is the transpose matrix of D w , I represents an identity matrix, AE represents an absolute error, and βc and βr are parameters for adjusting the weights of the clustering loss function and the regularization term.

[0108] The generator loss function includes three parts: an adversarial loss function, an aggregation loss function and a regularization term, as follows:

[0109]

[0110] where D c (X)^ is the class given by the discriminator when the generated traffic sequence sample is a real sample, A i indicates the aggregated traffic of the real traffic sequence sample {X} belonging to the class C i, indicates the aggregated traffic of the generated traffic sequence sample {X}^ belonging to the class C i, indicates the generated traffic mode switching behavior, indicates the one-hot encoding of the traffic mode followed by the user in the t th time period, diff indicates the first-order difference, ||·|| 2 indicates the two-norm, and β a and β s are parameters for adjusting the weights of the aggregation loss function and the regularization term.

[0111] In the adversarial training process, the generator and the discriminator are alternately trained according to the corresponding loss functions. In each training, first, the real traffic sequence sample {X} is input into the discriminator to give the class to which each sample belongs. Next, according to the sample class given by the discriminator, the corresponding traffic mode switching behavior {S}^ and traffic mode {P}^ are generated by multiple generators. After taking the Hadamard product of the traffic mode switching behavior {S}^ and the traffic mode {P}^, the generated user network traffic sequence can be obtained, and input into the discriminator to obtain the discrimination result of the generated traffic sequence sample, which is used to calculate the loss function and update the model.

[0112] The application provides a mobile user network traffic generation method based on an adversarial network, which models multiple network traffic modes of a mobile user and mode switching behaviors thereof through multiple generators, clusters the mobile user traffic while discriminating the authenticity of the mobile user traffic by using a discriminator, and finally supervises the model from the micro and macro levels at the same time by using multiple loss functions to generate more real and effective mobile user network traffic, so that the spatial coverage range is large enough and the resolution is improved.

[0113] Reference Figure 7 The application further discloses a mobile user network traffic generation system based on an adversarial network, which comprises:

[0114] The behavior feature acquisition module 110 is configured to acquire mobile user behavior feature data and extract behavior features from the mobile user behavior feature data.

[0115] The mobile user network traffic sequence generation module 120 generates a mobile user network traffic sequence by simulating multiple network traffic modes and mode switching behaviors of a mobile user based on the extracted behavior features through the pre-stored multiple generators.

[0116] A judgment module 130 is configured to judge the authenticity of the mobile user network traffic sequence by using a pre-stored discriminator, and generate a judgment result.

[0117] A training module 140 is configured to train a preset adversarial network model by using a loss function of the discriminator and a loss function of a generator according to the judgment result.

[0118] A data generation module 150 is configured to train the adversarial network model to convergence, and generate a normalized network traffic sequence dataset.

[0119] A behavior feature acquisition module 110 is configured to extract a mobile application usage feature and a time mobile behavior feature of each user from mobile user behavior feature data.

[0120] The mobile application usage feature is obtained from network data packets, and the time mobile behavior feature is obtained from a base station address accessed by the user when the user performs mobile network connection.

[0121] A mobile user network traffic sequence generation module 120 is configured to generate a mobile user network traffic sequence by using a generator, and the generator includes a network traffic pattern generator and a network traffic pattern switching behavior generator.

[0122] The network traffic pattern generator simulates a mobile user network traffic pattern based on the extracted behavior features.

[0123] The network traffic pattern switching behavior generator performs pattern switching learning based on the simulated mobile user network traffic pattern, and generates a mobile user network traffic sequence.

[0124] The judgment module 130 judges the authenticity of the mobile user network traffic sequence by using the discriminator.

[0125] If the mobile user network traffic sequence is judged as real data, the adversarial training is performed.

[0126] The loss function of the discriminator and the loss function of the generator both include an adversarial loss function, a clustering loss function and a regularization term.

[0127] The adversarial network model is alternately trained by using the loss function of the generator and the loss function of the discriminator.

[0128] The data generation module 150 inputs a real traffic sequence sample into the adversarial network model after the adversarial network model is trained to convergence.

[0129] The adversarial network model outputs a normalized network traffic sequence dataset.

[0130] The application provides a mobile user network traffic generation system based on an adversarial network, multiple generators are used to model multiple network traffic modes of mobile users and mode switching behaviors of the mobile users, a discriminator is used to cluster the mobile user traffic while judging the authenticity of the mobile user traffic, multiple loss functions are used to supervise the model from micro and macro levels to generate more real and effective mobile user network traffic, and the spatial coverage range is large enough to improve the resolution.

[0131] Figure 8 An example of an entity structure diagram of an electronic device is shown in Figure 8 The electronic device can include a processor 810, a communications interface 820, a memory 830 and a communications bus 840, wherein the processor 810, the communications interface 820 and the memory 830 can communicate with each other through the communications bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the mobile user network traffic generation method based on the adversarial network, and the method includes: obtaining mobile user behavior feature data, extracting behavior features from the mobile user behavior feature data;

[0132] Based on the extracted behavior features, multiple generators are used to simulate multiple network traffic modes and mode switching behaviors of mobile users, and mobile user network traffic sequences are generated;

[0133] The authenticity of the mobile user network traffic sequences is judged by a pre-stored discriminator, and a judgment result is generated;

[0134] According to the judgment result, the loss function of the discriminator and the loss function of the generator are used to train a preset adversarial network model;

[0135] The adversarial network model is trained to convergence, and a normalized network traffic sequence dataset is generated.

[0136] Further, the logic instructions in the memory 830 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0137] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the mobile user network traffic generation method based on the adversarial network provided by the above-mentioned methods, the method comprising: obtaining mobile user behavior feature data, and extracting behavior features from the mobile user behavior feature data;

[0138] Based on the extracted behavior features, a plurality of generators are used to simulate a plurality of network traffic modes and mode switching behaviors of the mobile user, and a mobile user network traffic sequence is generated;

[0139] A discriminator is used to judge the authenticity of the mobile user network traffic sequence, and a judgment result is generated;

[0140] According to the judgment result, the loss function of the discriminator and the loss function of the generator are used to train a preset adversarial network model;

[0141] The adversarial network model is trained to convergence, and a normalized network traffic sequence dataset is generated.

[0142] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the mobile user network traffic generation method based on the adversarial network provided by the above-mentioned methods, the method comprising: obtaining mobile user behavior feature data, and extracting behavior features from the mobile user behavior feature data;

[0143] Based on the extracted behavior features, a plurality of generators are used to simulate a plurality of network traffic modes and mode switching behaviors of the mobile user, and a mobile user network traffic sequence is generated;

[0144] The authenticity of the mobile user network traffic sequence is judged by a pre-stored discriminator, and a judgment result is generated;

[0145] According to the judgment result, a preset adversarial network model is trained by using a loss function of the discriminator and a loss function of the generator;

[0146] The adversarial network model is trained to convergence, and a normalized network traffic sequence dataset is generated.

[0147] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0148] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course, they can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0149] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating mobile user network traffic based on adversarial networks, characterized in that, The method comprises the following steps: obtaining mobile user behavior feature data, and extracting behavior features from the mobile user behavior feature data; based on the extracted behavior features, simulating multiple network traffic patterns and mode switching behaviors of mobile users by using multiple pre-stored generators to generate mobile user network traffic sequences; judging the authenticity of the mobile user network traffic sequences by using a pre-stored discriminator to generate a judgment result; training a preset adversarial network model according to the judgment result by using a loss function of the discriminator and a loss function of the generator; training the adversarial network model to convergence to generate a normalized network traffic sequence dataset; wherein the step of obtaining mobile user behavior feature data and extracting behavior features from the mobile user behavior feature data comprises: extracting the use of mobile application features and time mobile behavior features of each user from the mobile user behavior feature data; the use of mobile application features is obtained from network data packets, and the time mobile behavior features are obtained from the base station addresses accessed by the user when connecting to the mobile network. 2.The mobile user network traffic generation method based on the adversarial network according to claim 1, wherein, the step of simulating multiple network traffic patterns and mode switching behaviors of mobile users by using multiple pre-stored generators based on the extracted behavior features to generate mobile user network traffic sequences comprises: the generator comprises a network traffic pattern generator and a network traffic mode switching behavior generator; the network traffic pattern generator simulates the network traffic pattern of the mobile user based on the extracted behavior features; the network traffic mode switching behavior generator learns mode switching based on the simulated network traffic pattern of the mobile user to generate the mobile user network traffic sequence. 3.The mobile user network traffic generation method based on the adversarial network according to claim 1, wherein, the step of judging the authenticity of the mobile user network traffic sequences by using a pre-stored discriminator to generate a judgment result comprises: judging the authenticity of the mobile user network traffic sequences by using the discriminator; if the mobile user network traffic sequence is judged as real data, adversarial training is performed. 4.The mobile user network traffic generation method based on the adversarial network according to claim 1, wherein, the step of training a preset adversarial network model according to the judgment result by using a loss function of the discriminator and a loss function of the generator comprises: the loss function of the discriminator and the loss function of the generator both comprise an adversarial loss function, a clustering loss function and a regularization term; the adversarial network model is alternately trained by using the loss function of the generator and the loss function of the discriminator. 5.The mobile user network traffic generation method based on the adversarial network according to claim 1, wherein, the step of training the adversarial network model to convergence to generate a normalized network traffic sequence dataset comprises: after the adversarial network model is trained to convergence, a real traffic sequence sample is input into the adversarial network model; the adversarial network model outputs a normalized network traffic sequence dataset.

6. A mobile user network traffic generation system based on adversarial networks, characterized in that, The system comprises: a behavior feature acquisition module for obtaining mobile user behavior feature data and extracting behavior features from the mobile user behavior feature data; a mobile user network traffic sequence generation module for simulating multiple network traffic patterns and mode switching behaviors of mobile users by using multiple pre-stored generators based on the extracted behavior features to generate mobile user network traffic sequences; a judgment module for judging the authenticity of the mobile user network traffic sequences by using a pre-stored discriminator to generate a judgment result; The training module is configured to train a preset adversarial network model according to the judgment result and using a loss function of the discriminator and a loss function of the generator. The data generation module is configured to train the adversarial network model to convergence and generate a normalized network traffic sequence data set. The mobile user behavior feature data is obtained, and behavior features are extracted from the mobile user behavior feature data, and the process specifically includes: The use mobile application features and time mobile behavior features of each user are extracted from the mobile user behavior feature data. The use mobile application features are obtained from network data packets, and the time mobile behavior features are obtained from base station addresses accessed by the user when the user connects to a mobile network.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the mobile user network traffic generation method based on the adversarial network according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the mobile user network traffic generation method based on the adversarial network according to any one of claims 1 to 5.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the mobile user network traffic generation method based on the adversarial network according to any one of claims 1 to 5.

Images