Wireless communication network fault diagnosis method based on generative adversarial network

Abstract

The invention relates to the technical field of wireless mobile communication networks, and discloses a wireless communication network fault diagnosis method based on a generative adversarial network, which comprises the following steps: inputting acquired network fault real data; carrying out data processing, and extracting a key performance indicator (KPI) of the equipment; designing an NS-DCGAN, inputting a key performance indicator (KPI) of the equipment into the NS-DCGAN, and generating virtual fault data in different network states; supplementing external environment characteristic data for the virtual fault data by adopting a clustering algorithm; combining the virtual data set with the real data set, and constructing a wireless communication network fault sample data set D; dividing the data set D into an internal equipment index part D1 and an external environment feature part D2; an EAFD-GhostNet is designed, and the D1 and the D2 are input into the EAFD-GhostNet; and carrying out wireless communication network fault classification diagnosis, and outputting a diagnosis result. According to the method, the problems of insufficient fault samples and unbalanced data distribution can be solved, short-time and efficient multi-dimensional fault diagnosis analysis can be realized, and the accuracy of fault classification diagnosis can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless mobile communication networks, in particular to self-organizing network technology, and specifically discloses a wireless communication network fault diagnosis method based on a generative adversarial network, which is oriented to a cross-virtual-real combination based on a generative adversarial sample library. Background Art

[0002] With the accelerated development of 5G and 6G technologies, the structure of mobile communication networks is becoming more and more complex, and the cost of network operation and maintenance is increasing. SON (Self-Organized Networks) technology can greatly improve the automation and intelligence level of wireless mobile communication networks and effectively reduce operating costs.

[0003] As an important part of SON technology, fault diagnosis is mainly aimed at analyzing the perceived and reported network parameters, flexibly providing fault warning and diagnosis, and ensuring the QoS of communication network services. In wireless communication networks, nodes are constantly moving, network situations are complex, faults are prone to occur, and they need to be diagnosed and processed in a short time. At the same time, network equipment computing resources are in short supply, and the system must quickly perform fault diagnosis with limited resources. Therefore, the design of fault diagnosis algorithms needs to fully consider situations such as frequent network changes and limited resources. In addition, existing communication network fault samples have problems such as insufficient sample quantity and uneven distribution of fault samples, which also poses a considerable obstacle to the training design of fault diagnosis algorithms. The following defects exist in the prior art:

[0004] (1) Existing fault diagnosis designs are based on fixed-node communication networks with stable network structures and link resources, and cannot be directly adapted to wireless mobile communication networks with scarce computing resources.

[0005] (2) Existing fault diagnosis designs are difficult to implement when communication fault samples are insufficient and data distribution is unbalanced, resulting in poor results;

[0006] (3) The existing wireless mobile communication network fault diagnosis design mainly considers the system equipment itself and pays less attention to objective external environmental factors. Summary of the invention

[0007] In order to solve the above technical problems existing in the prior art, the present invention proposes a wireless communication network fault diagnosis method based on a generative adversarial network, comprising the following steps:

[0008] Step 1: Collect real network fault data from a heterogeneous wireless communication network environment, including key performance indicator (KPI) data of internal devices and external environment feature data;

[0009] Step 2: Perform data processing, including data cleaning and filling, to form a real data set, separate and extract the key performance indicator (KPI) data of the equipment;

[0010] Step 3: Design NS-DCGAN, input the key performance indicator (KPI) data of the equipment into NS-DCGAN, learn the existing data distribution based on the adversarial idea of ​​the generator and the discriminator, and generate virtual fault data under different network states;

[0011] Step 4: Use a clustering algorithm to supplement the virtual fault data with external environment feature data to obtain a virtual data set;

[0012] Step 5: Merge the virtual data set with the real data set to construct a wireless communication network fault sample data set D;

[0013] Step 6, dividing the data set D into an internal device indicator part D1 and an external environment feature part D2;

[0014] Step 7, design a dual-path environmental adaptability fault diagnosis neural network EAFD-GhostNet, process the data of the internal equipment indicator part D1, input it into Path1 of the environmental adaptability fault diagnosis neural network EAFD-GhostNet for feature extraction, process the data of the external environment feature part D2, input it into Path2 of the environmental adaptability fault diagnosis neural network EAFD-GhostNet for encoding; connect the data encoded by Path2 with the data after feature extraction by Path1, perform wireless communication network fault classification diagnosis, and output the diagnosis result.

[0015] Furthermore, the key performance indicator KPI data of the extracted equipment in step 2 includes:

[0016] Collect KPI data using different lengths of time aggregation levels;

[0017] Define S = [KPI1, KPI2, KPI3, ..., KPI m ], represents the input vector of different network states, S contains the vector of m KPIs;

[0018] Define C = [FC1, FC2, FC3, ..., FC n ], representing n different states of the network;

[0019] During time T, the network is in fault state FC i , state sample It is expressed as:

[0020]

[0021] For selected KPIsi The normalization operation is as follows:

[0022]

[0023] Among them, KPI i is the normalized i-th KPI.

[0024] Furthermore, the NS-DCGAN in step 3 includes a generator G and a discriminator D, and the activation function Nselu(x) is:

[0025]

[0026] The optimized objective function is:

[0027] Discriminator:

[0028] Generator:

[0029] Among them, P data represents the real data distribution, P z represents the distribution of the noise,

[0030] The NS-DCGAN is designed by the following steps:

[0031] Step 3.1: The generator randomly generates j noise samples Z, and generates generator samples through a deep convolutional network:

[0032] p g (z):{z (1) ,…,z (j)} (6)

[0033] Step 3.2: Randomly sample j real communication network fault samples from the real communication network fault data set to form a real fault sample:

[0034] p data (s):{s (1) ,…,s (j)} (7)

[0035] Step 3.3, fix the generator state and convert the generator sample p g (z) and the real fault sample p data (s) Input discriminator, trained through convolution operation and activation function, according to the real fault sample p data (s) and generator sample p g The difference between (z) is used to obtain the loss of the discriminator. According to the optimized objective function, the loss of the discriminator is designed as follows:

[0036]

[0037] Step 3.4: Use the stochastic gradient algorithm to update the parameters as follows:

[0038]

[0039] Among them, η is the learning rate, θ d is the discriminator weight parameter;

[0040] Step 3.5: After the discriminator is updated, fix the discriminator and randomly input the noise sample into the generator to generate the sample p. g (z):{z (1) ,…,z (j)}, according to the discriminator's discrimination result, the generator loss is obtained. Based on the optimized objective function, the generator loss is designed as follows:

[0041]

[0042] Step 3.6: Train and update the generator. The parameter update function is as follows:

[0043]

[0044] Among them, η is the learning rate, θ g is the generator weight parameter;

[0045] Step 3.7: According to Step 3.1 to Step 3.6, the discriminator and the generator are alternately iterated and trained until the network converges or reaches a preset termination condition;

[0046] Step 3.8: Save the model after training is completed.

[0047] Furthermore, step 4 specifically includes: taking the real sample data as the cluster center, performing clustering according to the Euclidean distance, and adding external environment characteristic data to the virtual fault data, wherein the external environment characteristic data includes weather conditions and ground conditions.

[0048] Furthermore, step 7 specifically includes:

[0049] Input the internal device index part D1 into path1, perform convolution operation, and extract features preliminarily;

[0050] After that, the data enters the G-bneck layer for feature extraction. G-bneck is used to replace the original convolution, and linear operations are used to replace some convolution operations. The specific formula is:

[0051] y ij =φ i,j (y′ i ) (12)

[0052] Among them, y′ i represents the features extracted by the initial convolution, φ i,j Represents the linear operation function used to generate feature maps;

[0053] Next, a 1×1 convolution kernel is used to reduce the data dimension and achieve cross-channel aggregation of features;

[0054] Input the external environment feature part D2 into Path2, and perform one-hot encoding on the text features of the external environment feature part D2;

[0055] Connect the encoded data in Path2 with the data after feature extraction in Path1;

[0056] The connected data are input into the fully connected layer FC for subsequent joint decision-making;

[0057] The fully connected data is passed through the softmax function to normalize the output results, thereby obtaining the fault diagnosis classification output;

[0058] Compare the output results with the label data, and use the multi-classification cross entropy function to calculate the classification error. The formula is:

[0059]

[0060] Among them, M represents the number of categories, N is the number of samples, and y ic represents the symbol function, if the sample is equal to the true category, it takes 1, otherwise it takes 0; p ic is the predicted probability that the observed sample belongs to category c;

[0061] Based on the classification error, the back-propagation process uses the Adam optimizer for parameter optimization;

[0062] Repeat the above process until the preset termination condition is reached, stop training and save the model;

[0063] Input the real-time wireless mobile communication network data into the saved model to obtain the fault diagnosis results.

[0064] The present invention is based on the complex jumping wireless mobile communication network environment, based on the generative adversarial idea, and innovatively designs the NS-DCGAN generative adversarial network to generate virtual communication network fault samples and construct a virtual-real combined sample library to solve the problems of insufficient fault samples and uneven data distribution.

[0065] The present invention is aimed at complex and changeable network environments, and based on artificial intelligence and deep learning technologies, can achieve short-term and efficient multi-dimensional diagnosis and analysis of faults.

[0066] The present invention fully incorporates system-related internal equipment indicators and objective external environmental characteristics, processes and analyzes the two separately, retains highly relevant features to the greatest extent, and effectively improves the accuracy of fault classification diagnosis. In addition, the present invention takes into account the shortage of communication network resources and designs a lightweight dual-path environmental adaptability fault diagnosis neural network. The previously generated virtual fault samples are merged with the real samples, and the EAFD-GhostNet algorithm is used to implement wireless communication network fault classification diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flow chart of the present invention;

[0068] Figure 2 This is the structural diagram of NS-DCGAN;

[0069] Figure 3 This is the discriminator structure diagram of NS-DCGAN;

[0070] Figure 4 This is the generator structure diagram of NS-DCGAN;

[0071] Figure 5 This is the structural diagram of EAFD-GhostNet. DETAILED DESCRIPTION

[0072] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0073] The process of the present invention is as follows Figure 1 As shown, the following steps are included:

[0074] (1) A small amount of real network fault data is collected from a heterogeneous wireless mobile communication network environment as input data, including the key performance indicator (KPI) data of internal devices and external environment feature data.

[0075] (2) Perform data processing, mainly including data cleaning and filling, to form a real data set, separate and extract the key performance indicator (KPI) data of the equipment.

[0076] In the communication network, the key performance indicator KPI data of internal equipment is the data collected during the operation of the equipment, which is related to the operation status of the equipment system program, with large data volume and high data dimension. The external environment characteristic data is the objective existence of the external environment that is close to being stable within a certain period of time and has nothing to do with the system itself. However, the external environment is highly related to network failure conditions. For example, there are many obstructions in mountainous terrain that lead to unstable link connections, and short circuit failures in links caused by thunderstorms. It is hoped that the external environment characteristic data will be retained as much as possible. The two have different characteristics, so the key performance indicator KPI data of internal equipment and the external environment data are considered separately.

[0077] (3) Design a generative adversarial network (NS-DCGAN) and input the key performance indicator (KPI) data of internal equipment into NS-DCGAN. Based on the adversarial idea of ​​the generator and the discriminator, learn the existing data distribution, thereby generating a large amount of virtual mobile communication network fault data under different network conditions, solving the problems of insufficient fault samples and unbalanced data distribution.

[0078] (4) The generated virtual fault data does not include external environment data. Based on Euclidean distance, a clustering algorithm is used to supplement the virtual data with external environment feature data to obtain a virtual data set.

[0079] (5) The virtual data set is merged with the real data set to construct a complete wireless communication network fault sample data set D that combines virtual and real data.

[0080] (6) Divide the data set D into the internal equipment indicator part D1 and the external environment characteristic part D2.

[0081] (7) A lightweight network design is adopted to design a dual-path environmental adaptive fault diagnosis neural network (EAFD-GhostNet) for resource-scarce and complex-hopping wireless mobile communication networks. The D1 and D2 data are processed and input into Path1 and Path2 of EAFD-GhostNet respectively. After feature extraction, the internal device indicators and the characteristics of the external environment are connected, and a comprehensive analysis is performed to jointly perform classification diagnosis of wireless communication network faults and output the diagnosis results.

[0082] The data used in the present invention include system internal equipment index data and objective external environment data, which are used to determine 7 different types of faults. Equipment index KPI (Key Performance Indicators) include reference signal receiving power RSRP (Reference Signal Receiving Power), reference signal receiving quality RSRQ (Reference Signal Receiving Quality), uplink packet loss rate PD_UL (Packet Drop_UpLink), downlink packet loss rate PD_DL (Packet Drop_Down Link) and other collected equipment key index data; external environment data include weather conditions, terrain conditions, etc. As shown in Table 1.

[0083] Table 1 Fault diagnosis related parameters

[0084]

[0085]

[0086] The intelligent communication network fault diagnosis method based on generative adversarial network proposed in the present invention can be roughly divided into two parts: virtual communication network fault sample generation based on NS-DCGAN generative adversarial network and communication network fault classification diagnosis based on dual-path EAFD-GhostNet. It specifically includes four stages: data sample processing, virtual fault sample generation, virtual data feature supplementation, and communication network fault classification diagnosis.

[0087] 1. Data sample processing.

[0088] In the wireless network environment, the internal device indicator data of the system is collected. These real small sample fault data constitute the input data vector, which consists of all relevant KPI device indicator data of the object under study. According to the time granularity required by the communication network fault diagnosis process, different lengths of time aggregation levels (such as seconds, minutes, hours, etc.) can be used to collect KPI data.

[0089] Define S = [KPI1, KPI2, KPI3, ..., KPI m ], represents the input vector of different network states, and S contains the vector of m KPIs. KPI values ​​that are not between the interval [0, 1] should be normalized to ensure that all variables are within the expected interval and all indicators are in the same order of magnitude.

[0090] Define C = [FC1, FC2, FC3, ..., FC n], indicating n different states of the network, such as the network is working normally or some type of failure occurs, such as link failure, interference, etc.

[0091] During a certain period of time T, the network has a fault state FC i , the state sample during this period It is expressed as follows:

[0092]

[0093] For selected KPIs i The normalization operation is as follows:

[0094]

[0095] Among them, KPI i Refers to the normalized i-th KPI.

[0096] 2. Generate virtual fault samples based on the NS-DCGAN network when there are insufficient fault samples in the wireless mobile communication network.

[0097] NS-DCGAN generative adversarial network mainly consists of two modules, one is the generator G and the other is the discriminator D. The network structure of NS-DCGAN is as follows: Figure 2 shown.

[0098] The ReLU function is often used in deep convolutional networks. When the forward propagation input is less than 0, the neuron will be in an inactive state and will "kill" the gradient during the backward propagation process. Experiments have found that with the continuous increase in the number of iterations, the training of the generative adversarial network may oscillate, non-converge or even overfit, and the sample generation effect is not good. Therefore, the present invention designs a new activation function Nselu. While solving the above problems, the samples are automatically normalized to 0 mean and unit variance to achieve self-normalization. And an upper limit threshold is set for the input x. When the input is greater than a specific value, the feature learning is stopped, effectively reducing the possibility of network oscillation in the later stage.

[0099]

[0100] In order to enhance the accuracy and robustness of the discriminator and enable the generative adversarial network to reach Nash equilibrium as quickly as possible, the discriminator objective function is regularly optimized based on the parameter gradient, and a special penalty term is added to the discrimination result. Therefore, the optimized objective function of the NS-DCGAN network is:

[0101] Discriminator:

[0102] Generator:

[0103] Among them, P datarepresents the real data distribution, P z represents the distribution that the noise obeys.

[0104] 1) The generator randomly generates j noise samples Z such as Gaussian noise input generator, and generates fault samples through the deep convolutional network:

[0105] p g (z):{z (1) ,…,z (j)} (6)

[0106] 2) Randomly sample j real communication network failure samples from the real communication network failure dataset

[0107] p data (s):{s (1) ,…,s (j)} (7)

[0108] 3) Fix the generator state and set the generator sample p g (z) and the real fault sample p data (s) Input the discriminator, train it through convolution operation and activation function, and get the discriminator loss according to the difference between the real fault sample and the generator sample. Based on the improved objective function, the discriminator loss is designed as follows:

[0109]

[0110] 4) Then use the stochastic gradient algorithm to update the parameters as follows, where η is the learning rate, θ d is the discriminator weight parameter:

[0111]

[0112] The discriminator structure is as follows Figure 3 shown.

[0113] 5) After the discriminator is updated, the discriminator is fixed and a random noise sample is input into the generator to generate a sample p g (z):{z (1) ,…,z (j)}. According to the discriminator's discrimination result, the generator loss is obtained. Based on the improved objective function, the generator loss is designed as follows:

[0114]

[0115] 6) Train and update the generator. The parameter update function is as follows, where η is the learning rate and θ is g is the generator weight parameter:

[0116]

[0117] The generator structure is as follows Figure 4 shown.

[0118] 7) According to the above process, the discriminator and the generator are trained alternately and iteratively until the network converges or reaches the pre-set termination condition.

[0119] Based on the above design, the present invention proposes a NS-DCGAN network using the proposed activation function and objective function. The specific architecture is shown in Table 2.

[0120] Table 2 NS-DCGAN network architecture

[0121]

[0122] 8) After training is completed, save the model and run the generator to generate virtual communication network fault samples.

[0123] 3. Supplement and improve external environment characteristics based on clustering.

[0124] The real communication network fault sample data includes objective external information such as weather and ground conditions. The generated virtual data only includes equipment indicator information, excluding the above external environmental characteristics. Therefore, based on the real samples, the real sample data is used as the cluster center, clustering is performed according to the Euclidean distance, and external environmental characteristics are added to the virtual fault sample annotations. The external environmental feature data in this method includes weather and ground conditions.

[0125] 4. EAFD-GhostNet communication network fault classification diagnosis based on dual paths.

[0126] The real communication network fault samples and virtual fault samples are combined into a new data set D. The data samples include internal equipment indicator data within a certain period of time and the corresponding external environmental characteristics. The internal equipment indicators of the system generally have internal correlation, have many connections with each other, and are mostly in numerical form. The external environmental characteristics are an objective existence, tend to be stable within a certain period of time, have a low correlation with each other, and exist in non-numerical form, which is different from the characteristics of internal equipment indicators. And according to historical experience, for wireless mobile communication networks, external environmental factors will have a greater impact on system failures, which are highly correlated characteristics. It is hoped that as many external environmental factors as possible will be retained. Therefore, it is necessary to adopt a dual-path processing method for the internal equipment indicators and external environmental characteristics of the system, taking both internal and external factors into consideration.

[0127] The present invention optimizes the external environment adaptability of the communication network fault dataset D that combines internal equipment indicators with external environment characteristics, increases the effective feature quantity, and thus further improves the accuracy of mobile communication network fault diagnosis. Based on the lightweight neural network, the dual-path EAFD-GhostNet (EnvironmentalAdaptive Fault Diagnosis GhostNet) is proposed, with the following structure: Figure 5 shown.

[0128] 1) Separate the internal equipment index data and external environment characteristics into D1 and D2 one by one.

[0129] 2) Input the data set D1 into path1, perform convolution operation, and extract features preliminarily;

[0130] 3) After that, the data enters the G-bneck layer for feature extraction. Using G-bneck instead of the original convolution and using simple linear operations to replace some convolution operations effectively reduces the number of network parameters and improves network operation efficiency;

[0131] y ij =φ i,j (y′ i ) (12)

[0132] y′ i represents the features extracted by the initial convolution, φ i,j Represents a linear operation function used to generate more feature maps.

[0133] 4) Then use 1×1 convolution kernel to reduce data dimension and realize cross-channel aggregation of features;

[0134] 5) Input the external environment feature data set into Path2;

[0135] 6) Perform one-hot encoding on the text features of the external environment feature dataset D2 and input it into the next step to retain the external environment features to the greatest extent;

[0136] 7) The encoded data in Path2 is concatenated with the data after feature extraction in Path1;

[0137] 8) The concatenated data is input into the fully connected layer FC for subsequent joint decision-making;

[0138] 9) The fully connected data is passed through the softmax function to normalize the output results, thereby obtaining the fault diagnosis classification output;

[0139] 10) Compare the output results with the label data and use the multi-classification cross entropy function to calculate the classification error:

[0140]

[0141] Among them, M represents the number of categories, N is the number of samples, and y ic represents the sign function (0 or 1), if the sample is equal to the true category, it takes 1, otherwise it takes 0; p ic The predicted probability that the observed sample belongs to class c.

[0142] 11) Based on the classification error, back propagation uses the Adam optimizer for parameter optimization.

[0143] 12) Repeat the above training process until the preset termination condition is reached, stop training and save the model.

[0144] 13) Input the real-time wireless mobile communication network data into the saved model to obtain the fault diagnosis results.

[0145] Aiming at the complex hopping wireless mobile communication network environment, the present invention proposes a wireless communication network fault diagnosis method based on generative adversarial network, and its beneficial effects are as follows:

[0146] The present invention takes into account the situation that there are insufficient fault samples in the wireless mobile communication network, generates fault samples based on the generative adversarial idea, constructs a virtual-real combined generative adversarial sample library, supplements the number of samples, and solves the problem of uneven distribution of fault samples.

[0147] The present invention considers the network convergence problem in the process of generating samples, proposes a new network objective function and activation function, and adjusts the generative adversarial network architecture, which is conducive to quickly reaching Nash equilibrium and thus generating more accurate virtual communication network fault samples.

[0148] The present invention fully considers the characteristics of wireless mobile communication network data, integrates internal system data and external environment characteristics, further enriches data dimensions, and retains external features that are highly correlated with faults to the greatest extent, further improving the accuracy of fault diagnosis.

[0149] The present invention takes into account the situation of insufficient resources of wireless mobile communication network, adopts lightweight network design, has fewer network parameters, consumes less computing resources, and can complete mobile communication network fault classification diagnosis more efficiently and quickly.

[0150] The present invention proposes a new nonlinear activation function Nselu, which is beneficial to reduce the oscillation phenomenon in the training of deep generative adversarial networks and accelerate network convergence; the present invention proposes an innovative method for generating fault samples of wireless mobile communication networks based on NS-DCGAN, and constructs a virtual-real combination sample library. The proposed Nselu activation function is used to improve the objective function and update function of the generative adversarial network, and the NS-DCGAN generative adversarial network is constructed to effectively improve the efficiency of communication network fault sample generation; the present invention is based on complex hopping wireless mobile communication networks, and faces the shortage of bandwidth resources and computing resources. A fault classification and diagnosis network with lightweight characteristics is designed; the present invention targets the fault characteristics of wireless mobile communication networks, comprehensively analyzes the internal equipment indicator data of the system and the external environment characteristics, adopts a dual-path architecture, and innovatively improves the lightweight network with dual paths, and proposes a lightweight dual-path environmental adaptability fault diagnosis neural network EAFD-GhostNet.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A wireless communication network fault diagnosis method based on a generative adversarial network, characterized in that: The steps include: Step 1: Collect real network fault data from a heterogeneous wireless communication network environment, including key performance indicator (KPI) data of internal devices and external environment feature data; Step 2: Perform data processing, including data cleaning and filling, to form a real data set, separate and extract the key performance indicator (KPI) data of the equipment; Step 3: Design NS-DCGAN, input the key performance indicator (KPI) data of the equipment into NS-DCGAN, learn the existing data distribution based on the adversarial idea of ​​the generator and the discriminator, and generate virtual fault data under different network states; Step 4: Use a clustering algorithm to supplement the virtual fault data with external environment feature data to obtain a virtual data set; Step 5: Merge the virtual data set with the real data set to construct a wireless communication network fault sample data set D; Step 6, dividing the data set D into an internal device indicator part D1 and an external environment feature part D2; Step 7, design a dual-path environmental adaptability fault diagnosis neural network EAFD-GhostNet, process the data of the internal equipment indicator part D1, input it into Path1 of the environmental adaptability fault diagnosis neural network EAFD-GhostNet for feature extraction, process the data of the external environment feature part D2, input it into Path2 of the environmental adaptability fault diagnosis neural network EAFD-GhostNet for encoding; connect the data encoded by Path2 with the data after feature extraction by Path1, perform wireless communication network fault classification diagnosis, and output the diagnosis result.

2. A method for diagnosing wireless communication network faults based on a generative adversarial network according to claim 1, characterized in that: The key performance indicator KPI data of the extracted equipment in step 2 includes: Collect KPI data using different lengths of time aggregation levels; Define S = [KPI1, KPI2, KPI3, ..., KPI m ], represents the input vector of different network states, S contains the vector of m KPIs; Define C = [FC1, FC2, FC3, ..., FC n ], representing n different states of the network; During time T, the network is in fault state FC i , state sample It is expressed as: For selected KPIs i The normalization operation is as follows: Among them, KPI i is the normalized i-th KPI.

3. A method for diagnosing wireless communication network faults based on a generative adversarial network according to claim 2, characterized in that: The NS-DCGAN described in step 3 includes a generator G and a discriminator D, and the activation function Nselu(x) is: The optimized objective function is: Discriminator: Generator: Among them, P data represents the real data distribution, P z represents the distribution of the noise, The NS-DCGAN is designed by the following steps: Step 3.1: The generator randomly generates j noise samples Z, and generates generator samples through a deep convolutional network: p g (from):{from (1) ,…,With (j) } (6) Step 3.2: Randomly sample j real communication network fault samples from the real communication network fault data set to form a real fault sample: p data (s):{s (1) ,…,s (j) } (7) Step 3.3, fix the generator state and convert the generator sample p g (z) and the real fault sample p data (s) Input discriminator, trained through convolution operation and activation function, according to the real fault sample p data (s) and generator sample p g The difference between (z) is used to obtain the loss of the discriminator. According to the optimized objective function, the loss of the discriminator is designed as follows: Step 3.4: Use the stochastic gradient algorithm to update the parameters as follows: Among them, η is the learning rate, θ d is the discriminator weight parameter; Step 3.5: After the discriminator is updated, fix the discriminator and randomly input the noise sample into the generator to generate the sample p. g (z):{z (1) ,…,z (j) }, according to the discriminator's discrimination result, the generator loss is obtained. Based on the optimized objective function, the generator loss is designed as follows: Step 3.6: Train and update the generator. The parameter update function is as follows: Among them, η is the learning rate, θ g is the generator weight parameter; Step 3.7: According to Step 3.1 to Step 3.6, the discriminator and the generator are alternately iterated and trained until the network converges or reaches a preset termination condition; Step 3.8: Save the model after training is completed.

4. A method for diagnosing wireless communication network faults based on a generative adversarial network according to claim 3, characterized in that: Step 4 specifically includes: taking the real sample data as the cluster center, performing clustering according to the Euclidean distance, and adding external environment characteristic data to the virtual fault data, wherein the external environment characteristic data includes weather conditions and ground conditions.

5. A method for diagnosing wireless communication network faults based on a generative adversarial network according to claim 4, characterized in that: Step 7 specifically includes: Input the internal device index part D1 into path1, perform convolution operation, and extract features preliminarily; After that, the data enters the G-bneck layer for feature extraction. G-bneck is used to replace the original convolution, and linear operations are used to replace some convolution operations. The specific formula is: and ij =φ i,j (and i ') (12) Among them, y i ' represents the features extracted by the initial convolution, φ i,j Represents the linear operation function used to generate feature maps; Next, a 1×1 convolution kernel is used to reduce the data dimension and achieve cross-channel aggregation of features; The external environment feature part D2 is input into Path2, and the text features of the external environment feature part D2 are one-hot encoded; Connect the encoded data in Path2 with the data after feature extraction in Path1; The connected data are input into the fully connected layer FC for subsequent joint decision-making; The fully connected data is passed through the softmax function to normalize the output results, thereby obtaining the fault diagnosis classification output; Compare the output results with the label data, and use the multi-classification cross entropy function to calculate the classification error. The formula is: Among them, M represents the number of categories, N is the number of samples, and y ic represents the symbol function, if the sample is equal to the true category, it takes 1, otherwise it takes 0; p ic is the predicted probability that the observed sample belongs to category c; Based on the classification error, the back-propagation process uses the Adam optimizer for parameter optimization; Repeat the above process until the preset termination condition is reached, stop training and save the model; Input the real-time wireless mobile communication network data into the saved model to obtain the fault diagnosis results.

Images