5G bearer network fault diagnosis method and equipment based on graph convolutional neural network

Abstract

The present invention discloses a 5G bearer network fault diagnosis method and device based on a graph convolutional neural network. The method includes: obtaining historical time-series operation and maintenance data and corresponding labels of network equipment in the 5G bearer network; constructing a network structure diagram based on the 5G bearer network; obtaining historical and existing time-series operation and maintenance data after dimensionality reduction; constructing a 5G bearer network fault intelligent diagnosis model based on the network structure diagram through a graph convolutional neural network algorithm; using the reduced-dimensional historical time-series operation and maintenance data and labels as well as the reduced-dimensional existing time-series operation and maintenance data to train the 5G bearer network fault intelligent diagnosis model, and obtaining the fault type of the existing time-series operation and maintenance data, thereby realizing the diagnosis of 5G bearer network faults. The present invention constructs a 5G bearer network fault intelligent diagnosis model based on a graph convolutional neural network, performs semi-supervised classification on the time-series operation and maintenance data, can accurately identify the fault type of the 5G bearer network, and realizes the intelligent diagnosis of 5G bearer network faults.

Description

Technical Field

[0001] The present invention belongs to the technical field of 5G bearer network fault diagnosis, and specifically relates to a 5G bearer network fault diagnosis method and device based on graph convolutional neural network. Background Art

[0002] With the commercialization of 5G, driven by various research and applications, the number of 5G terminal connections has increased rapidly. Various new services based on 5G deployment and applications have accelerated the development of digital technology in life and played a significant role in the transformation from the consumer field to the production field and from the virtual economy to the real economy.

[0003] The infrastructure of a 5G mobile communications network consists of an access network, a bearer network, and a core network. The 5G bearer network, located between the core and access networks, carries all 5G mobile services. With the research and application of 5G, the scale of the bearer network has rapidly expanded, and network faults have also increased, necessitating more intelligent fault diagnosis technologies. The current mainstream 5G bearer network operation and maintenance approach relies on semi-automatic tools to assist experts in manual analysis and location of fault causes. However, in large-scale projects, fault diagnosis, location, and equipment repair and replacement are time-consuming. Timely replacement of equipment requires the availability of pre-assembled spare parts for network equipment, resulting in extremely low time and cost-effectiveness. To alleviate the lack of intelligent fault diagnosis in 5G bearer networks, the integration of artificial intelligence theories and methods into the operation and maintenance process can be considered to enable automated diagnosis of faults and security risks. Summary of the Invention

[0004] In response to the technical problem that the means of 5G bearer network fault diagnosis are not intelligent enough, the present invention provides a 5G bearer network fault diagnosis method and equipment based on graph convolutional neural network. By making full use of graph convolutional neural network theory and technology, an intelligent diagnosis method for 5G bearer network faults is proposed. The existing time series operation and maintenance data of network equipment in the 5G bearer network is dynamically acquired. By combining historical time series operation and maintenance data with historical fault data, an intelligent diagnosis model for 5G bearer network faults based on graph convolutional neural network is constructed. Semi-supervised classification is performed on the time series operation and maintenance data, and the fault type of the 5G bearer network is accurately identified, thereby realizing intelligent diagnosis of 5G bearer network faults.

[0005] In order to achieve the above-mentioned purpose, the technical solution provided by the present invention is as follows:

[0006] A 5G bearer network fault diagnosis method based on graph convolutional neural network is constructed to accurately identify the fault type of the 5G bearer network. The method includes the following steps:

[0007] (1) Obtain historical time-series operation and maintenance data and historical fault data of network equipment in the 5G bearer network, and quantify and encode the historical fault data;

[0008] (2) Considering that historical time series operation and maintenance data contains a large amount of redundant data, the principal component analysis (PCA) method is used to extract features from the historical time series operation and maintenance data;

[0009] (3) Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network, and use PCA to extract features. Combine the processed historical time series operation and maintenance data and historical fault data to divide the training set, validation set, test set and label;

[0010] (4) A 5G bearer network fault intelligent diagnosis model is constructed through the graph convolutional neural network algorithm to identify the fault type of the existing time series operation and maintenance data and realize the intelligent diagnosis of 5G bearer network faults.

[0011] Furthermore, in step (1), the historical time series operation and maintenance data and historical fault data of the network equipment in the 5G bearer network are obtained by the following method, and the historical fault data is quantized and encoded:

[0012] ① Obtain historical time-series operation and maintenance data and historical fault data: Obtain network equipment status data, network equipment environment data, network equipment operation data, and corresponding fault types in the 5G bearer network, and organize them to obtain N historical time-series operation and maintenance data with k feature dimensions and n corresponding fault types.

[0013] ② Quantitative encoding of historical fault data: Map n fault types to n one-hot code vectors, namely [1,0,…,0,0], [0,1,…,0,0], …, [0,0,…,1,0], [0,0,…,0,1], and then convert the n one-hot code vectors into natural numbers, that is, use 0 to represent [1,0,…,0,0], use 1 to represent [0,1,…,0,0], …, use n-2 to represent [0,0,…,1,0], and use n-1 to represent [0,0,…,0,1]. At the same time, a dictionary with natural numbers as keys and fault types as values ​​is established to facilitate subsequent models to predict fault types. The natural numbers representing the fault types are used as labels, corresponding to the historical time series operation and maintenance data.

[0014] ③ Constructing a network structure diagram based on the 5G bearer network: Modeling the network nodes and their connections within the 5G bearer network creates a network structure diagram with a non-Euclidean spatial structure. Network nodes form the nodes in the network structure diagram, and the connections between them form the edges, resulting in a complete, connected, undirected graph. Each network node has a unique digital ID representing its identity. N historical time-series operation and maintenance data with a feature dimension of k, containing fault data for all network nodes, is randomly generated.

[0015] Furthermore, in step (2), considering that the historical time series operation and maintenance data contains a large amount of redundant data, the historical time series operation and maintenance data is extracted by PCA, and the historical time series operation and maintenance data A with N feature dimensions of k is converted into (a1, a2, ..., k ) Dimensionality reduction to l dimension:

[0016] ① Centralize historical time series operation and maintenance data. Centralization can be calculated using the following formula:

[0017]

[0018] ② Using the centralized historical time series operation and maintenance data, calculate the covariance matrix of the historical time series operation and maintenance data: The covariance matrix of the historical time series operation and maintenance data is given by The covariance matrix is ​​obtained It is a k*k square matrix.

[0019] ③ For the covariance matrix A T A performs eigenvalue decomposition: Use Jacobi method to find the covariance matrix A T The eigenvalues ​​of A and the covariance matrix A T The eigenvalues ​​of A are sorted by their size, i.e. λ1≥λ2≥…≥λ k ≥0, and then find the eigenvalue λ respectively i The corresponding unit eigenvector e i (i=1,2,…,k).

[0020] ④ Calculate the first l principal component values ​​of the historical time series operation and maintenance data: take out the first l eigenvalues ​​λ j The corresponding unit eigenvector e j , j = 1, 2, ..., l, the first l principal component values ​​can be calculated as follows:

[0021] P=(Ae1,Ae2,…,Ae l )=(p1,p2,…,p l )

[0022] Where P is the historical time series operation and maintenance data after dimensionality reduction, P∈R N*l .

[0023] Furthermore, in step (3), the existing time series operation and maintenance data of network equipment in the 5G bearer network is dynamically obtained, and PCA is used for feature extraction. The processed historical time series operation and maintenance data and historical fault data are combined to divide the training set, validation set, test set and labels:

[0024] ① Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network: obtain the existing network equipment status data, network equipment environment data, and network equipment operation data in the 5G bearer network, organize and obtain M existing time series operation and maintenance data with a feature dimension of k, and randomly generate M existing time series operation and maintenance data.

[0025] ② Use PCA to remove redundant data in the existing time series operation and maintenance data: Refer to the PCA dimensionality reduction method in step (2) and reduce the existing time series operation and maintenance data B with M feature dimensions k = (b1, b2, ..., k ) dimension is reduced to l dimension, and the existing time series operation data Q∈R M*l .

[0026] ③ Divide the training set, validation set, test set and labels of the graph convolutional neural network algorithm: the training set contains Historical time series operation and maintenance data, the validation set contains The test set contains M historical time series operation and maintenance data. The historical time series operation and maintenance data have quantized and encoded fault type labels.

[0027] Furthermore, in step (4), a 5G bearer network fault intelligent diagnosis model is constructed through a graph convolutional neural network algorithm to identify the fault type of the existing time series operation and maintenance data and realize intelligent diagnosis of 5G bearer network faults:

[0028] ① Construct the adjacency matrix A of the time series operation and maintenance data and perform normalization processing: the historical time series operation and maintenance data P after dimensionality reduction and the existing time series operation and maintenance data Q form a matrix Take the last set of randomly shuffled digital IDs of all network nodes in P, record the positions of the digital IDs, and build the adjacency matrix A∈R of the time series operation and maintenance data based on the connection relationship between the network nodes represented by the digital IDs. (M+N)*(M+N) , the adjacency matrix A is a sparse real symmetric square matrix. The adjacency matrix A plus the self-loop I N Prevent repeated calls to this operator in graph convolutional neural networks from causing numerical instability, gradient explosion, and gradient disappearance, that is, Among them I N ∈R (M+N)*(M+N) is the identity matrix. The degree matrix for The normalized adjacency matrix A is calculated as follows, where is the normalized adjacency matrix.

[0029]

[0030] ② Build an intelligent fault diagnosis model for the 5G bearer network using the graph convolutional neural network algorithm: Based on the theory and technology of graph convolutional neural networks, a two-layer graph convolutional neural network model is constructed. The forward propagation formula of the two-layer graph convolutional neural network model is as follows:

[0031]

[0032] in, is the normalized adjacency matrix, X∈R (+N)* is (M+N) time series operation and maintenance data with feature dimension l, W (0) ∈R l*t is the weight matrix from the input layer to the hidden layer. After the hidden layer, the time series data changes from l dimension to t dimension. W (1) ∈R t*n is the weight matrix from the hidden layer to the output layer. After the output layer, the time series data changes from t dimension to n dimension. Z∈R (+N)* is the output matrix, the ReLU function is the activation function, and the softmax function is the normalized exponential function, which acts on each row of the output matrix Z.

[0033] ③Calculate the accuracy of the fault type of the training set: get the output result H of the training set from the output matrix Z, Take the index of the maximum position in each row vector of H to form The matrix of Compare the corresponding labels of the time series data, if the values ​​are the same, record them as 1, if the values ​​are different, record them as 0. Add the comparison results and divide by The accuracy of the fault type in the training set can be obtained.

[0034] ④Training a two-layer graph convolutional neural network model: weight matrix W (0) and W (1) The values ​​are randomly initialized and then iteratively optimized by back-propagation gradient descent according to the maximum likelihood loss function. The learning rate and number of training times are set, and L2 regularization and Dropout are introduced to prevent overfitting of the two-layer graph convolutional neural network model.

[0035] ⑤ Predict the fault type of the existing time series operation and maintenance data: The prediction result of the fault type of the existing time series operation and maintenance data is an n-dimensional vector. Take the index of the maximum value position in the n-dimensional vector, call the dictionary established in step (1) with natural numbers as keys and fault types as values, and identify the fault type of the time series data test set.

[0036] A diagnostic device that uses any of the above-mentioned 5G bearer network fault diagnosis methods based on graph convolutional neural networks.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] (1) Considering that time series operation and maintenance data contains a large amount of redundant data, PCA is used to map high-dimensional data to low-dimensional data to achieve feature extraction of time series operation and maintenance data of 5G bearer networks;

[0039] (2) Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network, combine it with historical time series operation and maintenance data and historical fault data, and realize real-time prediction of 5G bearer network fault types;

[0040] (3) Construct an intelligent diagnosis model for 5G bearer network faults based on graph convolutional neural networks to accurately identify 5G bearer network faults and improve the intelligence of 5G bearer network fault diagnosis methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flowchart of a 5G bearer network fault diagnosis method based on graph convolutional neural network;

[0042] Figure 2 This is a network structure diagram of the interconnectedness of the main prefecture-level cities in Hubei Province;

[0043] Figure 3 Flowchart for PCA feature extraction of historical time series operation and maintenance data;

[0044] Figure 4 A flowchart for building a 5G bearer network fault intelligent diagnosis model based on graph convolutional neural networks;

[0045] Figure 5 This is a graph showing the changes in training set loss and validation set accuracy over epochs during graph convolutional neural network algorithm training.

[0046] Figure 6 This is a graph showing the changes in training set loss and test set accuracy over epochs during the training of the graph convolutional neural network algorithm. DETAILED DESCRIPTION

[0047] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0048] Graph convolutional neural network is a technology that uses edge information to aggregate node information to generate new node representations. It uses graph convolutional neural network to extract spatial features from the relationship topology graph established by nodes and edges for machine learning. By extracting the spatial features of the topology graph through graph convolutional neural network, the present invention proposes a 5G bearer network fault intelligent diagnosis method based on graph convolutional neural network, which can be applied to the fault type judgment and identification process in the operation and maintenance of 5G bearer network: according to the fault characteristics of 5G bearer network, combined with the network equipment status, environment and operation data, the common fault types of 5G bearer network are analyzed and quantized, and the principal component analysis method is used to remove redundant data in the historical time series operation and maintenance data; the historical time series operation and maintenance data and historical fault data are used as the training set, verification set and label of the graph convolutional neural network algorithm respectively, and the existing time series operation and maintenance data of network equipment in the 5G bearer network are dynamically obtained as the test set of the graph convolutional neural network algorithm, and a 5G bearer network fault intelligent diagnosis model is constructed to accurately identify the fault type of the existing time series operation and maintenance data, thereby realizing intelligent 5G bearer network fault diagnosis.

[0049] like Figure 1 As shown, the 5G bearer network fault intelligent diagnosis method of the embodiment of the present invention mainly includes the following steps:

[0050] (1) Obtain historical time-series operation and maintenance data and historical fault data of network equipment in the 5G bearer network, and quantify and encode the historical fault data;

[0051] (2) Considering that historical time series operation and maintenance data contains a large amount of redundant data, the principal component analysis (PCA) method is used to extract features from the historical time series operation and maintenance data;

[0052] (3) Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network, and use PCA to extract features. Combine the processed historical time series operation and maintenance data and historical fault data to divide the training set, validation set, test set and label;

[0053] (4) A 5G bearer network fault intelligent diagnosis model is constructed through the graph convolutional neural network algorithm to identify the fault type of the existing time series operation and maintenance data and realize the intelligent diagnosis of 5G bearer network faults.

[0054] The above steps are the steps from data acquisition to intelligent diagnosis of 5G bearer network faults in an embodiment of the present invention.

[0055] Wherein, in step (1), Figure 2 As shown in the figure, based on the geographical location relationship, a network structure diagram of the interconnected major prefecture-level cities in Hubei Province is constructed, the historical time-series operation and maintenance data and historical fault data of network equipment in Hubei Province's 5G bearer network are obtained, and the historical fault data is quantified and encoded. The specific steps are as follows:

[0056] ① Obtain historical time-series operation and maintenance data and historical fault data: Imitating the network device status data, network device environment data, network device operation data in the 5G bearer network and the corresponding fault types, construct 2088 historical time-series operation and maintenance data with 60 feature dimensions and the corresponding 6 fault types;

[0057] ② Quantitatively encode historical fault data: Map the 6 fault types to 6 one-hot code vectors, namely [1,0,0,0,0,0], [0,1,0,0,0,0], [0,0,1,0,0,0], [0,0,0,1,0,0], [0, and then convert the 6 one-hot code vectors into natural numbers, that is, use 0 to represent [1,0,0,0,0,0], use 1 to represent [0,1,0,0,0,0], use 2 to represent [0,0,1,0,0,0], use 3 to represent [0,0,0,1,0,0], use 4 to represent [0,0,0,0,1,0], use 5 to represent [0,0,0,0,0,1]. At the same time, establish a dictionary with natural numbers as keys and fault types as values to facilitate the subsequent model to predict fault types. The natural numbers representing fault types are used as labels and correspond to the historical time-series operation and maintenance data; <000020, establish a dictionary with natural numbers as keys and fault types as values to facilitate the subsequent model to predict fault types. The natural numbers representing fault types are used as labels and correspond to the historical time-series operation and maintenance data;

[0058] ③ Introduction to the network structure diagram of the embodiment: In view of the problem of insufficient intelligence in the fault diagnosis means of the 5G bearer network, therefore, it is necessary to model the geographical location relationship of many cities and counties in Hubei Province to obtain a network structure diagram with a non-Euclidean space structure. The main prefecture-level cities in Hubei Province and the subordinate cities and counties form 116 nodes of the network structure diagram. The connection relationships of the main prefecture-level cities are as Figure 2 shown. The cities and counties under the main prefecture-level cities are directly connected to the prefecture-level cities they belong to respectively, forming a network structure diagram with 115 edges. This diagram is an undirected graph. The 116 cities and counties have unique digital ids and Chinese ids. For example, the digital id of Wuhan City is 31 and the Chinese id is Wuhan City. The digital id is used to represent the connection relationships of the 115 edges. Each of the 116 cities and counties has 3 groups of 60-dimensional historical time-series operation and maintenance data containing 6 fault types and the corresponding fault type labels. The 2088 = 116 * 3 * 6 historical time-series operation and maintenance data are randomly generated.

[0059] In step (2), as Figure 3 shown, considering that there is a large amount of redundant data in the historical time-series operation and maintenance data, perform feature extraction on the historical time-series operation and maintenance data through PCA. Reduce the 2088 historical time-series operation and maintenance data A = (a1, a2,... 60 ) with 60 feature dimensions to 40 dimensions. The specific steps are as follows:

[0060] ① Centralize the historical time-series operation and maintenance data: Centralization can be calculated through the following formula;

[0061]

[0062] ② Using the centralized historical time series operation and maintenance data, calculate the covariance matrix of the historical time series operation and maintenance data: The covariance matrix of the historical time series operation and maintenance data is given by The covariance matrix is ​​obtained It is a 60*60 square matrix;

[0063] ③ For the covariance matrix A T A performs eigenvalue decomposition: Use Jacobi method to find the covariance matrix A T The eigenvalues ​​of A and the covariance matrix A T The eigenvalues ​​of A are sorted by their size, i.e. λ1≥λ2≥…≥λ 60 ≥0, and then find the eigenvalue λ respectively i The corresponding unit eigenvector e i (i=1,2,…,60);

[0064] ④ Calculate the first 40 principal component values ​​of the historical time series operation and maintenance data: extract the first 40 eigenvalues ​​λ j The corresponding unit eigenvector e j (j=1,2,…,40), the first 40 principal component values ​​can be calculated by the following formula, P is the historical time series operation and maintenance data after dimensionality reduction, P∈R 2088* .

[0065] P=(Ae1,Ae2,…,Ae 40 )=(p1,p2,…,p 40 )

[0066] In step (3), the existing time series operation and maintenance data of network equipment in the 5G bearer network is dynamically obtained, and PCA is used for feature extraction. The processed historical time series operation and maintenance data and historical fault data are combined to divide the training set, validation set, test set and label. The specific steps are as follows:

[0067] ① Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network: Based on the network equipment status data, network equipment environment data, network equipment operation data and the corresponding fault types in the 5G bearer network, 2088 existing time series operation and maintenance data with a feature dimension of 60 are constructed. The 2088 existing time series operation and maintenance data are randomly generated. In order to subsequently verify the accuracy of the existing time series operation and maintenance data, 6 fault types corresponding to the existing time series operation and maintenance data are constructed. The fault types are quantized and encoded with reference to step (1);

[0068] ② Use PCA to remove redundant data in the existing time series operation and maintenance data: Refer to the PCA dimensionality reduction method in step (2) to reduce the 2088 existing time series operation and maintenance data B with a feature dimension of 60 = (b1, b2, ..., 60 ) is reduced to 40 dimensions, and the existing time series operation data Q∈R 2088* ;

[0069] ③ Divide the graph convolutional neural network algorithm into training set, validation set, test set and labels: the training set contains 1,392 historical time series operation and maintenance data, the validation set contains 696 historical time series operation and maintenance data, and the test set contains 2,088 existing time series operation and maintenance data. The label consists of quantized encoded fault data.

[0070] In step (4), if Figure 4 As shown in the figure, a 5G bearer network fault intelligent diagnosis model is constructed through the graph convolutional neural network algorithm to identify the fault type of existing time series operation and maintenance data and realize intelligent diagnosis of 5G bearer network faults. The specific steps are as follows:

[0071] ① Construct the adjacency matrix A of the time series operation and maintenance data and perform normalization processing: the historical time series operation and maintenance data P after dimensionality reduction and the existing time series operation and maintenance data Q form a matrix Take the last set of 116 randomly shuffled digital IDs of cities and counties in P, record the positions of the digital IDs, and build the adjacency matrix A∈R of the time series operation and maintenance data based on the network structure diagram of the 116 cities and counties represented by the digital IDs. 4176*4176 , the adjacency matrix A is a sparse real symmetric square matrix. In this embodiment, there are 2088 digital IDs in P, namely: 116 counties and cities, but each county and city appears 3*6 times. Only the digital IDs of the last 116 counties and cities are taken, the positions of the digital IDs are recorded, and the adjacency matrix is ​​constructed according to the network structure diagram of the 116 counties and cities. Take the adjacency matrix A plus the self-loop I N Prevent repeated calls to this operator in graph convolutional neural networks from causing numerical instability, gradient explosion, and gradient disappearance, that is, Among them I N ∈R 4176*4176 is the identity matrix. The degree matrix for The normalized adjacency matrix A is calculated as follows, where is the normalized adjacency matrix;

[0072]

[0073] ② Use the graph convolutional neural network algorithm to build an intelligent fault diagnosis model for Hubei Province's 5G bearer network: Based on the theory and technology of graph convolutional neural networks, a two-layer graph convolutional neural network model is constructed. The forward propagation formula of the two-layer graph convolutional neural network model is as follows:

[0074]

[0075] in is the normalized adjacency matrix, X∈R 4176*40 It is 4176 time series operation and maintenance data with a feature dimension of 40, W (0) ∈R 40*t is the weight matrix from the input layer to the hidden layer. After the hidden layer, the time series data changes from 40 dimensions to t dimensions. W (1) ∈R t*6 is the weight matrix from the hidden layer to the output layer. After the output layer, the time series data changes from t dimensions to 6 dimensions. Z∈R 4176 *6 is the output matrix, the ReLU function is the activation function, and the softmax function is the normalized exponential function, which acts on each row of the output matrix Z;

[0076] ③Calculate the accuracy of the fault type of the training set: get the output result H of the training set from the output matrix Z, H∈R 1392 *6 , take the index of the maximum value position in each row vector of H to form a 1392*1 matrix, and compare it with the corresponding labels of the 1392 time series data. If the values ​​are the same, it is recorded as 1, and if the values ​​are different, it is recorded as 0. Add the 1392 comparison results and divide them by 1392 to get the accuracy of the fault type;

[0077] The accuracy calculation method is pre-set here, and the accuracy can be obtained according to this method after training is completed;

[0078] ④Training a two-layer graph convolutional neural network model: weight matrix W (0) and W (1) Randomly initialize the assignment and then perform iterative optimization through back-propagation gradient descent according to the maximum likelihood loss function. The number of training epochs is 100. The learning rate is set, and L2 regularization and Dropout are introduced to prevent overfitting of the two-layer graph convolutional neural network model.

[0079] ⑤ Calculate the accuracy of the fault type of the existing time series operation and maintenance data and predict the fault type: refer to the method for calculating the accuracy of the fault type of the training set to obtain the accuracy of the fault type of the existing time series operation and maintenance data. The prediction result of the fault type of the existing time series operation and maintenance data is a 6-dimensional vector. Take the index of the maximum value position in the 6-dimensional vector, call the dictionary established in step (1) with natural numbers as keys and fault types as values, and identify the fault type of the time series data test set.

[0080] Based on the actual fault types in the 5G bearer network, the present invention constructs 2088 historical time series operation and maintenance data with 60 feature dimensions, including 6 types of faults. The specific situation is shown in Table 1, and the method proposed in the present invention is tested and verified.

[0081] Table 1 Sequential operation and maintenance data and corresponding fault types

[0082]

[0083] Similarly, existing time series operation and maintenance data can be randomly generated according to Table 1.

[0084] Based on the PyCharm integrated development environment, we preprocessed historical time-series O&M data and fault data. Combined with dynamically acquired existing time-series O&M data, we built a two-layer graph convolutional neural network framework. We then performed semi-supervised learning on the two-layer graph convolutional neural network model to identify fault types in existing time-series O&M data and enable intelligent diagnosis of 5G bearer network faults. Table 2 shows the training results of the graph convolutional neural network model under different hyperparameters. Figure 5 and Figure 6 It is a curve chart showing the changes of training set loss and validation set accuracy with epoch during the training process of the graph convolutional neural network algorithm, as well as a curve chart showing the changes of training set loss and test set accuracy with epoch.

[0085] Table 2 Training results of graph convolutional neural network model under different hyperparameters

[0086]

[0087] The software platform and environment configuration information implemented in this invention are: Pycharm (2021.3.3), torch (1.12.1), cuda (12.0), matplotlib (3.5.1) and scipy (1.9.3).

[0088] Assuming that the existing time series operation and maintenance data have the same fault types as the historical time series operation and maintenance data, we construct 2088 existing time series operation and maintenance data with a feature dimension of 60, which contains 6 fault types. We also provide fault labels for the existing time series operation and maintenance data. We use the trained graph convolutional neural network model to perform semi-supervised classification of fault types on the existing time series operation and maintenance data, and predict the fault types of the existing time series operation and maintenance data. The results are shown in Table 3.

[0089] Table 3 Comparative analysis results

[0090]

[0091]

[0092] As can be seen from the preceding table, the accuracy of the test set, i.e., existing time-series O&M data, is as high as 93.63%. The graph convolutional neural network model can intelligently diagnose 5G bearer network faults based on the fault types in historical time-series O&M data. This method improves the efficiency of 5G bearer network fault diagnosis and addresses the issue of insufficient intelligence in 5G bearer network fault diagnosis.

[0093] In summary, the present invention provides a 5G bearer network fault intelligent diagnosis method and device based on graph convolutional neural network, which is used to accurately identify the fault type of the 5G bearer network. The method mainly includes: obtaining historical time series operation and maintenance data and historical fault data of network equipment in the 5G bearer network, and quantizing and encoding the historical fault data; considering that the historical time series operation and maintenance data contains a large amount of redundant data, the historical time series operation and maintenance data is subjected to feature extraction through principal component analysis (PCA); dynamically obtaining the existing time series operation and maintenance data of network equipment in the 5G bearer network, and performing feature extraction with PCA, and dividing the processed historical time series operation and maintenance data and historical fault data into training sets, verification sets, test sets and labels; constructing a 5G bearer network fault intelligent diagnosis model through the graph convolutional neural network algorithm, identifying the fault type of the existing time series operation and maintenance data, and improving the intelligence level of the 5G bearer network fault diagnosis method. Compared with the existing technology, the present invention addresses the problem of insufficient intelligence in 5G bearer network fault diagnosis methods. It makes full use of graph convolutional neural network theory and technology, proposes a 5G bearer network fault intelligent diagnosis method, dynamically obtains the existing time series operation and maintenance data of network equipment in the 5G bearer network, combines historical time series operation and maintenance data with historical fault data, and constructs a 5G bearer network fault intelligent diagnosis model based on graph convolutional neural network. It performs semi-supervised classification on the time series operation and maintenance data, accurately identifies the fault type of the 5G bearer network, and realizes intelligent 5G bearer network fault diagnosis.

[0094] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0095] It will be easily understood by those skilled in the art that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A 5G bearer network fault diagnosis method based on graph convolutional neural network, characterized in that: This method is used to identify the fault type of a 5G bearer network and includes the following steps: (1) Obtain the historical time-series operation and maintenance data and the corresponding historical fault data of network equipment in the 5G bearer network, and quantize and encode the historical fault data to obtain a natural number representing the fault type as a label, and the label corresponds to the historical time-series operation and maintenance data; Constructing a network structure diagram based on the 5G bearer network: Modeling the network nodes and their connection relationships in the 5G bearer network. Network nodes constitute nodes in the network structure diagram, and the connections between network nodes constitute edges in the network structure diagram. (2) Extract features from historical time series operation and maintenance data through principal component analysis (PCA) to obtain the historical time series operation and maintenance data after dimensionality reduction; (3) Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network, and use principal component analysis (PCA) to extract features to obtain the existing time series operation and maintenance data after dimensionality reduction; (4) Based on the constructed network structure diagram, a 5G bearer network fault intelligent diagnosis model is constructed through the graph convolutional neural network algorithm; the 5G bearer network fault intelligent diagnosis model is trained using the reduced-dimensional historical time series operation and maintenance data and labels as well as the reduced-dimensional existing time series operation and maintenance data, and the fault type of the existing time series operation and maintenance data is obtained to realize the diagnosis of 5G bearer network faults; Wherein, step (4) specifically includes: ① Construct the adjacency matrix A of the time series operation and maintenance data and perform normalization processing: The historical time series operation and maintenance data after dimensionality reduction is , the existing time series operation and maintenance data after dimension reduction is ;in, is the number of historical time series operation and maintenance data after dimensionality reduction, is the number of existing time series operation and maintenance data after dimensionality reduction, is the dimension; the historical time series operation and maintenance data P after dimensionality reduction and the existing time series operation and maintenance data Q after dimensionality reduction constitute the matrix X = ; Each network node has a unique digital ID. We take the last set of randomly shuffled digital IDs of all network nodes from the historical time series operation and maintenance data P after dimensionality reduction, record the positions of these digital IDs, and build the adjacency matrix of the time series operation and maintenance data based on the connection relationship between the network nodes represented by the digital IDs. ; Calculate the normalized adjacency matrix. The calculation formula is as follows: Where, is the normalized adjacency matrix, is the adjacency matrix The degree matrix , is the identity matrix; ②Build an intelligent fault diagnosis model for 5G bearer networks using a graph convolutional neural network algorithm: Construct a two-layer graph convolutional neural network model. The forward propagation formula of the two-layer graph convolutional neural network model is as follows: Z = 𝑓(X, A) = softmax( ReLU( X ) ) Where, is the normalized adjacency matrix, X The feature dimensions are (M+N) Time series operation and maintenance data, ∈ is the weight matrix from the input layer to the hidden layer, and the time series data through the hidden layer is Dimension becomes t dimension, ∈ is the weight matrix from the hidden layer to the output layer, and the time series data through the output layer is t Dimension becomes n dimension, n is the total number of fault types, Z∈ is the output matrix, the ReLU function is the activation function, and the softmax function is the normalized exponential function, which acts on each row of the output matrix Z; ③Calculate the accuracy of fault types in historical time series operation and maintenance data: ④Train a two-layer graph convolutional neural network model; ⑤ Predict the fault type of existing time series operation and maintenance data and identify the fault type of existing time series operation and maintenance data.

2. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 1 is characterized in that: Step (1) specifically includes: ① Obtain historical time series operation and maintenance data and historical fault data: Obtain network equipment status data, network equipment environment data, network equipment operation data and corresponding fault types in the 5G bearer network, and organize them N The feature dimension is k Historical time series operation and maintenance data and its corresponding n Types of faults; ② Quantitative coding of historical fault data: n Fault types are mapped to n In the one-hot code vectors, they are [1, 0,…,0, 0], [0, 1, …, 0, 0], …, [0, 0, …, 1, 0], [0, 0, …, 0, 1], and then n The one-hot code vectors are converted into natural numbers, that is, 0 represents [1, 0, …, 0, 0], 1 represents [0, 1, …, 0, 0], …, n -2 means [0, 0, …, 1, 0], use n -1 represents [0, 0, …, 0, 1]. A dictionary is created with natural numbers as keys and fault types as values, so that natural numbers correspond one-to-one with fault types. The natural numbers representing the fault types are used as labels and correspond to historical time series operation and maintenance data. ③ Construct a network structure diagram based on the 5G bearer network: Model the network nodes and the connection relationships of the network nodes in the 5G bearer network to obtain a network structure diagram with a non-Euclidean spatial structure, in which the network nodes constitute the nodes in the network structure diagram, and the connection relationships between the network nodes constitute the edges in the network structure diagram, thus obtaining a complete connected undirected graph; each network node has a unique digital ID to represent its identity, N The feature dimension is k The historical time series operation and maintenance data includes the fault data of all network nodes.

3. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 2 is characterized in that: Random Generation N Historical time series operation and maintenance data.

4. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 1 is characterized in that: Step (2) N The feature dimension is k The historical time series operation and maintenance data A = ( , ,…, ) Dimensionality reduction to l Dimensions, specifically including: ① Centralize historical time series operation and maintenance data. Centralization is calculated using the following formula: = - Where, The first i elements, For all historical time series operation and maintenance data i The average value of the elements; ② Based on the centralized historical time series operation and maintenance data, calculate the covariance matrix of the historical time series operation and maintenance data: the covariance matrix of the historical time series operation and maintenance data is given by The covariance matrix is ​​obtained yes k × k phalanx; ③ Covariance matrix Perform eigenvalue decomposition: Use Jacobi method to find the covariance matrix The eigenvalues ​​of and the covariance matrix The eigenvalues ​​of are sorted by their size, that is, ≥ ≥… …≥ ≥0, and then find the eigenvalues ​​respectively The corresponding unit eigenvector ; ④ Before calculating historical time series operation and maintenance data l Principal component values: before extraction l eigenvalues The corresponding unit eigenvector (j = 1, 2, …, l ),forward l The principal component values ​​are calculated as follows: P = (A , A ,…, A ) = ( , ,…, ) Where P is the historical time series operation and maintenance data after dimensionality reduction, .

5. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 4 is characterized in that: Step (3) specifically includes: ① Dynamically obtain the existing time series operation and maintenance data of network equipment in the 5G bearer network: obtain the existing network equipment status data, network equipment environment data and network equipment operation data in the 5G bearer network, and organize them M The feature dimension is k Existing time series operation and maintenance data; ② Use PCA to remove redundant data in existing time series operation and maintenance data: Refer to the PCA dimensionality reduction method of historical time series operation and maintenance data, and M The feature dimension is k The existing time series operation and maintenance data B = ( , ,…, ) Dimensionality reduction to l Dimension, the existing time series operation and maintenance data after dimensionality reduction is .

6. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 5 is characterized in that: Random Generation M Historical time series operation and maintenance data.

7. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 1 is characterized in that: Step (4) specifically also includes: ③ The accuracy of calculating the fault type of historical time series operation and maintenance data is as follows: Will n Fault types are mapped to n In the one-hot code vectors, they are [1, 0, …,0, 0], [0, 1, …, 0,0], …, [0, 0, …, 1, 0], [0, 0, …, 0, 1], and then n The one-hot code vectors are converted into natural numbers, that is, 0 represents [1, 0, …, 0, 0], 1 represents [0, 1, …, 0, 0], …, n -2 means [0, 0, …, 1, 0], use n -1 represents [0, 0, …, 0, 1]. A dictionary is created with natural numbers as keys and fault types as values, so that natural numbers correspond one-to-one with fault types. The natural numbers representing the fault types are used as labels and correspond to historical time series operation and maintenance data. The output result of historical time series operation and maintenance data is obtained from the output matrix Z , x is the number of selected historical time series operation and maintenance data, x ≤ N ; Take the index of the maximum position in each row vector of H to form x ×1 matrix, and x ×1 historical time series data, if the values ​​are the same, they are recorded as 1, if the values ​​are different, they are recorded as 0. x Add the comparison results and divide by x The accuracy of the selected historical time series operation and maintenance data fault type can be obtained; ⑤ Predict the fault type of the existing time series operation and maintenance data. The specific identification of the fault type of the existing time series operation and maintenance data is as follows: The prediction result of the fault type of the existing time series operation and maintenance data is n dimensional vector, take n The index of the maximum value position in the dimensional vector is used to call the established dictionary with natural numbers as keys and fault types as values ​​to identify the fault types of the existing time series operation and maintenance data.

8. The 5G bearer network fault diagnosis method based on graph convolutional neural network according to claim 1, characterized in that: Training a two-layer graph convolutional neural network model specifically involves: Weight matrix and Randomly initialize the assignment, then perform iterative optimization through back-propagation gradient descent according to the maximum likelihood loss function, set the learning rate and number of training times, and introduce L2 regularization and Dropout to prevent overfitting of the two-layer graph convolutional neural network model.

9. A diagnostic device, characterized in that The diagnostic device adopts the 5G bearer network fault diagnosis method based on graph convolutional neural network as described in any one of claims 1 to 8.

Images