5g bearer network fault dynamic diagnosis and positioning method based on alarm correlation relationship

Abstract

The application discloses a kind of 5G carrying network fault dynamic diagnosis and positioning method based on alarm association, comprising the following steps: network topology structure of each level node in 5G carrying network is constructed using directed graph, the real connection relationship of network equipment in 5G carrying network is described;The mapping relationship between node and vector in directed graph is constructed using node2vec algorithm, generates Node Embedding;The key features of typical historical time series alarm information in 5G carrying network are encoded by fusing node embedding generated by node2vec algorithm and 5G carrying network fault data, the 5G carrying network fault dynamic diagnosis and positioning model is constructed by two-layer long short-term memory neural network model, and the 5G carrying network fault dynamic diagnosis and positioning model is trained;The trained 5G carrying network fault dynamic diagnosis and positioning model is used to dynamically diagnose 5G carrying network network fault and locate the corresponding network equipment position.The application can dynamically diagnose 5G carrying network network fault and locate the corresponding network equipment position.

Description

Technical Field

[0001] This invention belongs to the field of machine learning technology, specifically relating to a method for dynamic diagnosis and location of faults in 5G bearer networks based on alarm correlation. Background Technology

[0002] 5G, the fifth generation of mobile communication, is characterized by "ultra-high speed, ultra-low latency, and massive connectivity." The development and application of 5G networks have played a significant role in promoting the digital, networked, and intelligent transformation and upgrading of the economy and society. However, the development and application of 5G networks are inseparable from the construction of the 5G bearer network. The 5G bearer network is one of the core infrastructures of the 5G network, providing high-speed, low-latency data transmission channels and supporting the connection of large-scale devices and IoT applications. With the continuous advancement of 5G bearer network construction, faults are emerging frequently, making the operation and maintenance of the 5G bearer network a constraint on the development of 5G network applications. Currently, the mainstream operation and maintenance method for 5G bearer networks relies on experts to diagnose and locate faults through continuous experimentation and replacement of network equipment based on experience. Experts need to spend a significant amount of time diagnosing the types of faults and locating the corresponding network equipment, resulting in high time and manpower costs. Manual methods for diagnosing and locating faults have achieved some success in small-scale 5G bearer networks, but they are difficult to handle the large-scale and complex network devices in large-scale 5G bearer networks. Therefore, a dynamic fault diagnosis and location method for 5G bearer networks based on alarm correlation is proposed. This method can model large-scale and complex network devices and then use artificial intelligence to dynamically diagnose and locate faults.

[0003] In the current operation and maintenance of 5G bearer networks, network maintenance personnel dynamically collect a large amount of alarm data, with numerous alarms of varying types. During 5G bearer network operation, alarms generated by network devices often exhibit strong coupling relationships (for example, a root cause alarm generated by one network device will lead to derivative alarms being generated by multiple connected network devices). Furthermore, due to the large number of network devices and complex connections, 5G bearer network faults also exhibit dynamic changes. Therefore, by fully utilizing alarm correlations and considering the dynamic changes in 5G bearer network faults, and employing artificial intelligence to achieve dynamic diagnosis and localization of 5G bearer network faults based on alarm correlations, this approach addresses the shortcomings of traditional 5G bearer network operation and maintenance methods, thereby promoting the application and development of 5G networks. Summary of the Invention

[0004] To address the insufficient intelligence of 5G bearer network fault diagnosis and location methods, this invention provides a dynamic fault diagnosis and location method for 5G bearer networks based on alarm correlation. This method fully utilizes alarm correlation, considers the dynamic changes in 5G bearer network faults, employs the node2vec algorithm to map each network device into a finite-dimensional vector, and uses a long short-term memory neural network to extract the temporal correlation features of 5G bearer network alarm data. This accurately diagnoses 5G bearer network faults and locates the corresponding network devices, solving the problem of outdated traditional 5G bearer network operation and maintenance methods, and thus promoting the application and development of 5G networks.

[0005] The technical solution adopted in this invention is as follows:

[0006] A method for dynamic fault diagnosis and location in 5G bearer network based on alarm correlation includes the following steps:

[0007] (1) Construct the network topology of each level of nodes in the 5G bearer network using directed graphs to describe the real connection relationship of network devices in the 5G bearer network;

[0008] (2) The node2vec algorithm is used to construct the mapping relationship between nodes and vectors in the directed graph and generate NodeEmbedding;

[0009] (3) Integrate node embeddings generated by the node2vec algorithm and 5G bearer network fault data, encode the key features of typical historical time-series alarm information in the 5G bearer network, construct a dynamic diagnosis and location model for 5G bearer network faults through a two-layer long short-term memory neural network model, and train the dynamic diagnosis and location model for 5G bearer network faults.

[0010] (4) Utilize the trained 5G bearer network fault dynamic diagnosis and location model to dynamically diagnose 5G bearer network faults and locate the corresponding network equipment locations.

[0011] Furthermore, step (1) specifically includes:

[0012] ① Obtain the actual connection relationships of network devices in the 5G bearer network;

[0013] ② Constructing the network topology of each level of nodes in the 5G bearer network using a directed graph: The 5G bearer network connects network devices through unidirectional optical fibers. The DiGraph() class in the networkx package is used to create a directed graph G = network.DiGraph(). A binary data structure (v1, v2) is used to represent that network device v1 is connected to network device v2 through a unidirectional optical fiber. All binary tuples representing the actual connection relationships of network devices are put into a list, and the edges in the list are added to the directed graph G to complete the construction of the network topology of each level of nodes in the 5G bearer network.

[0014] Furthermore, step (2) specifically includes:

[0015] ① Perform random walks on a directed graph G: For each node v in the directed graph G, perform n random walks. The length of each random walk is m. The strategy for each random walk follows the following formula, with a probability of P(c i =x|c i-1 =v) biased random walk, P(c i =x|c i-1 =v) represents the probability of moving from the current node v to the next node x, E represents the set of all nodes adjacent to the current node v, Z is the normalization constant, and π vx For the transition probability function:

[0016]

[0017]

[0018] In the formula, p and q are hyperparameters, and d tx It represents the shortest path distance from the previous node t to node x;

[0019] ② Perform one-hot encoding on each node in the directed graph G;

[0020] ③ Input the generated random walks into the Skip-Gram model: a set of random walks of length m is v t-3 ,v t-2 ,v t-1 ,v t ,v t+1 ,v t+2 ,v t+3 ..., if the sliding window is set to window=3, then node v t Construct training sample pairs (v t ,v t-3 ), (v t ,v t-2 ), (v t ,v t-1 ), (v t ,v t+1 ), (v t ,v t+2 ) and (v t ,v t+3 ), where node v t The corresponding one-hot code e t As input to the Skip-Gram model, node v t-3 The corresponding one-hot code e t-3 Node v t-2 The corresponding one-hot code et-2 Node v t-1 The corresponding one-hot code e t-1 Node v t+1 The corresponding one-hot code e t+1 Node v t+2 The corresponding one-hot code e t+2 and node v t+3 The corresponding one-hot code e t+3 These are used as labels for the Skip-Gram model;

[0021] In the Skip-Gram model, the input one-hot code e t The weight matrix between the input layer and the hidden layer is W. H The hidden layer has no activation function, and the hidden unit will have the one-hot code e. t Mapped to a low-dimensional u t The weight matrix between the hidden layer and the output layer is W. O The activation function of the output layer is softmax(), and the output unit converts the low-dimensional u t Mapped to a high-dimensional o t ;

[0022] The optimization objective of the Skip-Gram model is to ensure that the position of the maximum value in the output cell is relative to the one-hot code e. t-3 Unique hot code e t-2 Unique hot code e t-1 Unique hot code e t+1 Unique hot code e t+2 and unique hot code e t+3 The positions of 1 and 2 are the same. The error is measured by the cross-entropy loss function, and the weight parameters W are optimized by stochastic gradient descent. H W O After the Skip-Gram model is trained, the weight parameters W are used to determine its performance. H The node embedding u of each node v is obtained by looking up the table.

[0023] Furthermore, step (3) specifically includes:

[0024] ① By integrating node embeddings generated by the node2vec algorithm and 5G bearer network fault data, the key features of typical historical time-series alarm information in the 5G bearer network are encoded: the key features of historical time-series alarm information are the type of alarm information, the location of the network device that issued the alarm information, and the timing of multiple network devices issuing alarm information; based on the time-series characteristics of alarms coupled and associated with the 5G bearer network, the mapping relationship between multiple time-series alarm information and faults is derived. Multiple time-series alarm information is the input of the 5G bearer network fault dynamic diagnosis and location model, and the location of the faulty network device and the fault type are the labels;

[0025] One-hot encoding is performed on each type of alarm information; the node embedding generated by the node2vec algorithm represents the mapping relationship between nodes and vectors, and each node embedding uniquely identifies the location of each network device; the node embedding u is concatenated with the one-hot encoded alarm information to generate an alarm vector, with the node embedding u on top and the one-hot encoded alarm information below; alarms based on the coupling and association of 5G bearer network have temporal sequence characteristics, the R alarm vectors are sorted in chronological order, and the mapping relationship between multiple temporal alarm information and faults is combined to create labels for the location and fault type of the faulty network device, and the labels are represented by one-hot codes; a training set is created;

[0026] ②Build a dynamic diagnosis and location model for 5G bearer network faults: The dynamic diagnosis and location model for 5G bearer network faults includes two layers of long short-term memory neural networks and a fully connected layer. Each layer of long short-term memory neural network has R LSTM recurrent units. The output of the last LSTM recurrent unit of the second layer of long short-term memory neural network is connected to the fully connected layer. The activation function of the fully connected layer is softmax().

[0027] ③ Train a dynamic diagnosis and location model for 5G bearer network faults.

[0028] Furthermore, if there are fewer than R timing alarm vectors, the missing alarm vectors will be set to 0.

[0029] Furthermore, the 5 alarm vectors are sorted in chronological order. If the m-th alarm vector among the 5 alarm vectors causes the n-th type of fault, then the 4*(m-1)+n number of the one-hot code is set to 1.

[0030] Furthermore, training the 5G bearer network fault dynamic diagnosis and localization model includes:

[0031] The error was measured using the cross-entropy loss function. The Adam optimizer was selected, with an initial learning rate of 0.01, 8 training epochs, and a batch size of 16. Dropout was introduced to prevent overfitting of the 5G bearer network fault dynamic diagnosis and localization model.

[0032] Furthermore, the dynamic diagnosis of 5G bearer network faults and the location of corresponding network devices are as follows:

[0033] Take the index of the maximum value position in the output vector as the output. If the output result is r, then the r / 4+1th alarm vector in the 5 alarm vectors causes a fault in the 5G bearer network. Find the location of the corresponding network device from the r / 4+1th alarm vector. The fault category is the r%4th fault.

[0034] An electronic device, comprising:

[0035] Memory, used to store executable computer programs;

[0036] When a processor executes an executable computer program stored in a memory, it implements the 5G bearer network fault dynamic diagnosis and location method based on alarm association as described above.

[0037] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for dynamic diagnosis and location of 5G bearer network faults based on alarm association relationships as described above.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] (1) The node2vec algorithm is used to construct the mapping relationship between nodes and vectors in the network topology, and the nodes in the complex network topology are mapped into finite-dimensional vectors.

[0040] (2) Considering that the timing of alarm information is helpful for locating faults, the long short-term memory neural network is used to extract the temporal correlation of the timing alarm information.

[0041] (3) Dynamically acquire 5G bearer network timing alarm information, combine it with the 5G bearer network fault dynamic diagnosis and location model, dynamically diagnose 5G bearer network faults and locate the corresponding network equipment locations. Attached Figure Description

[0042] Figure 1 A flowchart of a method for dynamic fault diagnosis and location in 5G bearer networks based on alarm correlation.

[0043] Figure 2 Map of townships;

[0044] Figure 3 Here is the flowchart for the node2vec algorithm;

[0045] Figure 4 This is a diagram illustrating the transition probabilities followed during a random walk in node2vec.

[0046] Figure 5 Create a flowchart for the dataset; where, Figure 5 (a) Create a flowchart for alarm vectors. Figure 5 (b) is a schematic diagram showing the five alarm vectors arranged in chronological order. Figure 5 (c) is a schematic diagram of arranging 3 alarm vectors in chronological order. Figure 5 (d) is a label example;

[0047] Figure 6 Flowchart for building and training a dynamic fault diagnosis and location model for 5G bearer networks;

[0048] Figure 7 This is a schematic diagram of the internal structure of an LSTM recurrent cell;

[0049] Figure 8 This is a flowchart for dynamic diagnosis and location of faults in the 5G bearer network. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0051] 5G transport networks consist of numerous and complexly interconnected network devices, linked together in a network topology. To address this, the node2vec algorithm can construct a mapping between nodes and vectors in the network topology. This invention employs the node2vec algorithm to map each network device into a finite-dimensional vector. Simultaneously, the coupled alarms in the 5G transport network exhibit temporal sequence characteristics, and these temporal relationships are important features reflecting the types of faults in the 5G transport network. To address this, Long Short-Term Memory (LSTM) neural networks are deep learning algorithms for processing temporal data. This invention uses LSM to extract the temporal correlation features of 5G transport network alarm data. Therefore, this invention can model the large-scale and complexly connected network devices in the 5G transport network, and then use artificial intelligence methods to dynamically diagnose and locate faults.

[0052] like Figure 1 As shown, the intelligent fault diagnosis method for 5G bearer networks of the present invention mainly includes the following steps:

[0053] (1) Construct the network topology of each level of nodes in the 5G bearer network using directed graphs to describe the real connection relationship of network devices in the 5G bearer network;

[0054] (2) The node2vec algorithm is used to construct the mapping relationship between nodes and vectors in the directed graph and generate NodeEmbedding;

[0055] (3) Integrate node embeddings generated by the node2vec algorithm and 5G bearer network fault data, encode the key features of typical historical time-series alarm information in the 5G bearer network, construct a dynamic diagnosis and location model for 5G bearer network faults through a two-layer long short-term memory neural network model, and train the dynamic diagnosis and location model for 5G bearer network faults.

[0056] (4) Use the simulated dynamic generated 5G bearer network alarm and fault data to generate alarm vectors; use the constructed 5G bearer network fault dynamic diagnosis and location model to dynamically diagnose 5G bearer network faults and locate the corresponding network equipment locations.

[0057] The above steps are the steps for intelligent diagnosis of 5G bearer network faults from data acquisition in an embodiment of the present invention.

[0058] In step (1), a directed graph is used to construct the network topology of nodes at each level in the 5G bearer network, describing the actual connection relationships of network devices in the 5G bearer network, such as... Figure 2 As shown, Figure 2 This is a map of townships under the jurisdiction of Xixiu District, Guanling County, and Zhenning County in Anshun City, Guizhou Province. The 5G bearer network equipment is distributed below the township level. The specific steps are as follows:

[0059] ① Obtain the real connection relationship of 5G bearer network devices: Obtain the real connection relationship of 5G bearer network devices in townships under the jurisdiction of Xixiu District, Guanling County and Zhenning County, Anshun City, Guizhou Province, with a total of 427 network devices.

[0060] ② Constructing the network topology of each level of nodes in the 5G bearer network using directed graphs: The 5G bearer network connects network devices through unidirectional optical fibers. Therefore, the DiGraph() class in the networkx package can be used to create a directed graph G = network.DiGraph(). The data structure (v1, v2) of tuples represents that network device v1 is connected to network device v2 through unidirectional optical fibers. All tuples representing the actual connection relationships of network devices are put into a list, and the edges in the list are added to the directed graph G to complete the construction of the network topology of each level of nodes in the 5G bearer network.

[0061] In step (2), the node2vec algorithm is used to construct the mapping relationship between nodes and vectors in the directed graph, generating Node Embedding, such as... Figure 3 As shown, Figure 3 This is the flowchart of the node2vec algorithm, and the specific steps are as follows:

[0062] ① Perform random walks on a directed graph G: For each node v in the directed graph G, perform 128 random walks. The length of each random walk is 32. The strategy for each random walk follows the following formula, with a probability of P(c i =x|c i-1 =v) biased random walk, P(c i =x|c i-1 =v) represents the probability of moving from the current node v to the next node x, E represents the set of all nodes adjacent to the current node v, Z is the normalization constant, and πvx Let π be the transition probability function. vx As shown below, Figure 4 This is a diagram illustrating the transition probabilities followed during a random walk in node2vec, where nodes x1, x2, x3, x4, and t are all neighbors of node v, and node t is also the previous node of node v. Let d be the transition probability. tx The shortest path distance from the previous node t to node x is represented by hyperparameters p and q, which control the traversal strategy. In this embodiment, p = 1 and q = 0.2.

[0063]

[0064]

[0065] ② Perform one-hot encoding on the 427 nodes in the directed graph G: the i-th node v i The unique hot code is e i = [0,…,0,1,0…,0]∈R 1*427 The i-th position is 1, and the remaining 426 positions are 0.

[0066] ③ Input the generated random walk into the Skip-Gram model: such as Figure 3 The flowchart of the node2vec algorithm is shown. A set of random walks of length 32 is v t-3 ,v t-2 ,v t-1 ,v t ,v t+1 ,v t+2 ,v t+3 ..., if the sliding window is set to window=3, then node v t Can form training sample pairs (v) t ,v t-3 ), (v t ,v t-2 ), (v t ,v t-1 ), (v t ,v t+1 ), (v t ,v t+2 ) and (v t ,v t+3 ), where node v t The corresponding one-hot code e t As input to the Skip-Gram model, node v t-3 The corresponding one-hot code e t-3 Node v t-2 The corresponding one-hot code e t-2 Node v t-1The corresponding one-hot code e t-1 Node v t+1 The corresponding one-hot code e t+1 Node v t+2 The corresponding one-hot code e t+2 and node v t+3 The corresponding one-hot code e t+3 These serve as labels for the Skip-Gram model. In the Skip-Gram model, the input one-hot code e T The weight matrix between the input layer and the hidden layer is W. H ∈R 427*16 The hidden layer has no activation function, and the 16 hidden units convert the 427-dimensional one-hot code e. t Mapped to 16-dimensional u t ∈R 1*16 The weight matrix between the hidden layer and the output layer is W. O ∈R 16*427 The activation function of the output layer is softmax(), and the 427 output units convert the 16-dimensional u t Mapped to 427-dimensional o t ∈R 1*427 The optimization objective of the Skip-Gram model is to ensure that the positions of the maximum values ​​in the 427 output units are aligned with the one-hot code e. t-3 Unique hot code e t-2 Unique hot code e t-1 Unique hot code e t+1 Unique hot code e t+2 and unique hot code e t+3 The positions of 1 and 2 are the same. The error is measured by the cross-entropy loss function, and the weight parameters W are optimized by stochastic gradient descent. H W O After the Skip-Gram model is trained, the weight parameters W are used to determine its performance. H The node embedding u∈R of each node v is obtained through a lookup table. 1*16 .

[0067] In step (3), the node embeddings generated by the node2vec algorithm and the 5G bearer network fault data are integrated, and the key features of typical historical time-series alarm information in the 5G bearer network are encoded. A dynamic diagnosis and localization model for 5G bearer network faults is constructed through a two-layer long short-term memory neural network model. The specific steps of training the dynamic diagnosis and localization model for 5G bearer network faults are as follows:

[0068] ① By integrating node embeddings generated by the node2vec algorithm and 5G bearer network fault data, the key features of typical historical time-series alarm information in the 5G bearer network are encoded: the key features of historical time-series alarm information are the type of alarm information, the location of the network device issuing the alarm information, and the timing of alarm information issued by multiple network devices. When one network device in the 5G bearer network fails (and issues an alarm information), it is often accompanied by alarm information from multiple network devices. Based on the temporal characteristics of alarms coupled and associated with the 5G bearer network, a mapping relationship between multiple time-series alarm information and faults can be derived. These multiple time-series alarm information are the input to the 5G bearer network fault dynamic diagnosis and location model, and the location and fault type of the network device that failed are the labels. The example includes four types of alarm information and four types of fault information. Four alarm messages are one-hot encoded, resulting in alarm1 = [1, 0, 0, 0], alarm2 = [0, 1, 0, 0], alarm3 = [0, 0, 1, 0], and alarm4 = [0, 0, 0, 1]. The node embeddings generated by the node2vec algorithm represent the mapping relationship between nodes and vectors, with each node embedding uniquely identifying the location of each network device. For example... Figure 5 As shown in (a), 427 nodes are embedded in u∈R 1*16 Each of the four types of alarm information, after one-hot encoding, is concatenated to generate a 20*1 dimensional alarm vector with 427*4 lines. The node embedding u is on top, and the one-hot encoded alarm information is below. Alarms coupled and associated with the 5G bearer network have temporal sequence characteristics. Multiple alarm vectors are sorted in chronological order. The embodiment specifies a maximum of 5 temporal alarm vectors. Figure 5 (b) contains exactly 5 timing alarm vectors. If there are fewer than 5 timing alarm vectors, such as Figure 5 As shown in (c), there are only 3 alarm vectors. The first 2 alarm vectors are set to 0. Combining the mapping relationship between multiple time-series alarm information and faults, the location and fault type of the faulty network device are tagged. The tag is represented by a 1*20 dimensional one-hot code. If the m-th (1≤m≤5) alarm vector among the 5 alarm vectors causes the n-th (1≤n≤4) type of fault, then the 4*(m-1)+nth number of the one-hot code is set to 1. Figure 5 As shown in (d), if Figure 5 If the first alarm message in (b) causes the second type of fault, then the second number of the one-hot code is set to 1. Figure 5 If alarm message 5 in (c) causes type 1 failure, then the 17th number of the one-hot code is set to 1. 80% of the dataset is designated as the training set, and the remainder as the validation set;

[0069] ② Establish a dynamic fault diagnosis and location model for the 5G bearer network: The established dynamic fault diagnosis and location model for the 5G bearer network is as follows: Figure 6 As shown, the 5G bearer network fault dynamic diagnosis and localization model consists of two layers of Long Short-Term Memory (LSTM) neural networks and one fully connected layer. Each LSTM neural network layer has 5 LSTM recurrent units. The output of the last LSTM recurrent unit in the second LSTM neural network layer is connected to the fully connected layer, which has 20 neurons. The activation function of the fully connected layer is softmax(). The internal structure of the LSTM recurrent unit is shown below. Figure 7 As shown, an LSTM cyclic cell contains two states, namely cell state C. t and hidden state h t Its update formula is shown below, forget gate f t Input gate i t and output gate o t The update formula for the selection and rejection of control information is as follows:

[0070] Forgotten Gate: f t =sigmoid(W f [h t-1 ;x t ]+b f )

[0071] Input gate: i t =sigmoid(W i [h t-1 ;x t ]+b i )

[0072] Output gate: o t =sigmoid(W o [h t-1 ;x t ]+b o )

[0073] Update unit state: C t =f t ×C t-1 +i t ×tanh(W C [h t-1 ;x t ]+b C )

[0074] Update hidden state: h t =o t ×tanh(C t )

[0075] ③ Training the 5G bearer network fault dynamic diagnosis and localization model: The cross-entropy loss function was used to measure the error, the Adam optimizer was selected, the initial learning rate was 0.01, the number of training times was 8 epochs, the batch size was set to 16, and Dropout was introduced to prevent the 5G bearer network fault dynamic diagnosis and localization model from overfitting. The training results of the 5G bearer network fault dynamic diagnosis and localization model are shown in Table 1.

[0076] Table 1 Training Results of 5G Bearer Network Fault Dynamic Diagnosis and Localization Model

[0077]

[0078] In step (4), considering the simulated dynamically generated 5G bearer network alarm and fault data, the generated alarm vector is fused, and the constructed 5G bearer network fault dynamic diagnosis and location model is used to dynamically diagnose 5G bearer network faults and locate the corresponding network equipment locations. The specific steps are as follows:

[0079] ① Create a test set from the simulated dynamically generated 5G bearer network alarm and fault data: There are at most 5 simulated dynamically generated 5G bearer network alarm and fault data. Find the 20*1 dimension alarm vector corresponding to each alarm data. Sort the multiple alarm vectors in chronological order. If there are less than 5 alarm vectors, set the missing part to 0. The data type of the test set is numpy.ndarray, and the size is set to (1, 5, 20).

[0080] ② Dynamically diagnose 5G bearer network faults and locate corresponding network devices: Input the test set into the 5G bearer network fault dynamic diagnosis and location model. The prediction result of the 5G bearer network fault dynamic diagnosis and location model is a 20-dimensional vector. Take the index of the maximum value position in the 20-dimensional vector as the output. If the output result is r (1≤r≤20), then the r / 4+1th alarm vector among the 5 alarm vectors causes a 5G bearer network fault (if there are less than 5 alarm data, the missing part is set to 0). The location of the corresponding network device can be found from the r / 4+1th alarm vector, and the fault category is the r%4th fault. The 5G bearer network fault dynamic diagnosis and location process is as follows: Figure 8 As shown, the numbers 10109, 10101, 10102, etc. are custom identifiers for 427 network devices. Multiple alarm vectors are input into the 5G bearer network fault dynamic diagnosis and location model to accurately diagnose 5G bearer network faults and locate the corresponding device locations.

[0081] The present invention also provides an electronic device, comprising:

[0082] Memory, used to store executable computer programs;

[0083] When a processor executes an executable computer program stored in a memory, it implements the 5G bearer network fault dynamic diagnosis and location method based on alarm association as described above.

[0084] The present invention also provides a computer-readable storage medium storing a computer program for implementing, when executed by a processor, the method for dynamic diagnosis and location of 5G bearer network faults based on alarm association as described above.

[0085] In summary, this invention discloses a dynamic fault diagnosis and location method for 5G bearer networks based on alarm correlation, comprising the following steps: constructing the network topology of nodes at each level in the 5G bearer network using a directed graph to describe the actual connection relationships of network devices in the 5G bearer network; constructing the mapping relationship between nodes and vectors in the directed graph using the node2vec algorithm to generate Node Embeddings; fusing the node embeddings generated by the node2vec algorithm and 5G bearer network fault data to encode the key features of typical historical time-series alarm information in the 5G bearer network; constructing a dynamic fault diagnosis and location model for the 5G bearer network using a two-layer long short-term memory neural network model; and training the dynamic fault diagnosis and location model for the 5G bearer network; and using the trained dynamic fault diagnosis and location model for the 5G bearer network to dynamically diagnose 5G bearer network faults and locate the corresponding network device locations. This invention can dynamically diagnose 5G bearer network faults and locate the corresponding network device locations.

[0086] It should be noted that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0087] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0088] Those skilled in the art will readily understand 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 within the scope of protection of the present invention.

Claims

1. A method for dynamic fault diagnosis and location in 5G bearer networks based on alarm correlation, characterized in that, Includes the following steps: (1) Construct the network topology of each level of nodes in the 5G bearer network using directed graphs to describe the real connection relationship of network devices in the 5G bearer network; (2) The node2vec algorithm is used to construct the mapping relationship between nodes and vectors in the directed graph and generate NodeEmbedding; (3) Integrating node embeddings generated by the node2vec algorithm and 5G bearer network fault data, the key features of typical historical time-series alarm information in the 5G bearer network are encoded: the key features of historical time-series alarm information are the type of alarm information, the location of the network device that issued the alarm information, and the timing of multiple network devices issuing alarm information; based on the time-series characteristics of alarms coupled and associated with the 5G bearer network, the mapping relationship between multiple time-series alarm information and faults is derived. Multiple time-series alarm information is the input of the 5G bearer network fault dynamic diagnosis and location model, and the location and fault type of the faulty network device are the labels; one-hot encoding is performed on each type of alarm information; the node embedding generated by the node2vec algorithm represents the mapping relationship between nodes and vectors, and each node embedding uniquely identifies the location of each network device; the node embedding is then used to encode the key features of the alarm information. u Each alarm is concatenated with the one-hot encoded alarm information to generate an alarm vector, and the nodes are embedded. u The alarm information after unique hot coding is above; Alarms based on the coupling and association of 5G bearer network have time-series characteristics. R The alarm vectors are sorted in chronological order. By combining the mapping relationship between multiple time-series alarm information and the fault, the location and fault type of the faulty network device are labeled, and the labels are represented by one-hot codes; a training set is created. A dynamic fault diagnosis and localization model for 5G bearer network is constructed using a two-layer long short-term memory neural network model, and the model is then trained. (4) Using the trained 5G bearer network fault dynamic diagnosis and location model, dynamically diagnose 5G bearer network faults and locate the corresponding network equipment locations.

2. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, Step (1) specifically includes: ① Obtain the actual connection relationships of network devices in the 5G bearer network; ② Constructing the network topology of nodes at each level in the 5G bearer network using a directed graph: The 5G bearer network connects network devices through unidirectional optical fibers. The directed graph is created using the DiGraph() class in the networkx package: G = network.DiGraph(), using a binary tuple data structure (…). 1, 2) Characterizing network devices 1. Connect to network devices via unidirectional fiber optic cable.

2. Place all the pairs representing the actual connection relationships of network devices into a list, and add the edges in the list to the directed graph G to complete the construction of the network topology of each level of nodes in the 5G bearer network.

3. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, Step (2) specifically includes: ① Perform a random walk on a directed graph G: For each node in the directed graph G... Do n n random walks, each random walk having a length of n. m Each random walk follows the strategy described in the following formula, with a probability of . P ( =x| = Biased random walk, P ( =x| = ) indicates starting from the current node The probability of reaching node x at the next time step. E Indicates the relationship with the current node The set of all adjacent nodes, Z It is a normalization constant. For the transition probability function: P ( =x| = ) = = In the formula, p and q For hyperparameters, It represents the shortest path distance from the previous node t to node x; ② Perform one-hot encoding on each node in the directed graph G; ③ Input the generated random walks into the Skip-Gram model: a set of random walks of length... m random walk for , , , , , , ..., if the sliding window is set to window=3, then the node... Construct training sample pairs ( , ), ( , ), ( , ), ( , ), ( , )as well as( , ), where nodes Corresponding one-hot code As input to the Skip-Gram model, nodes Corresponding one-hot code ,node Corresponding one-hot code ,node Corresponding one-hot code ,node Corresponding one-hot code ,node Corresponding one-hot code and nodes Corresponding one-hot code These are used as labels for the Skip-Gram model; In the Skip-Gram model, input a one-hot code. The weight matrix between the input layer and the hidden layer is: The hidden layer has no activation function, and the hidden units will be one-hot codes. Mapped to a lower dimension The weight matrix between the hidden layer and the output layer is: The activation function of the output layer is softmax(), and the output unit will convert the low-dimensional... Mapped to a higher dimension ; The optimization objective of the Skip-Gram model is to ensure that the position of the maximum value in the output cell is relative to the one-hot code. Unique hot code Unique hot code Unique hot code Unique hot code and unique hot code The positions of '1' and '2' are the same. The error is measured using the cross-entropy loss function, and the weight parameters are optimized using stochastic gradient descent. , After the Skip-Gram model is trained, the weight parameters are used to determine its performance. Each node is obtained through a lookup table. v Node embedding u .

4. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, Step (3) of building a dynamic fault diagnosis and location model for the 5G bearer network specifically includes: Building a dynamic fault diagnosis and localization model for 5G bearer networks: The dynamic fault diagnosis and localization model for 5G bearer networks includes two layers of long short-term memory neural networks and one fully connected layer. Each layer of long short-term memory neural networks has... R The output of the last LSTM recurrent unit in the second layer of the Long Short-Term Memory neural network is fed into a fully connected layer, and the activation function of the fully connected layer is softmax().

5. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, If the timing alarm vector is insufficient R If the previous alarm vector is missing, then the missing alarm vector will be set to 0.

6. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, R is 5, the number of fault types is 4, and the 5 alarm vectors are sorted in chronological order. If the m-th alarm vector among the 5 alarm vectors causes the n-th fault, then the 4th one-hot code... Set (m-1)+n numbers to 1.

7. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 1, characterized in that, Training the 5G bearer network fault dynamic diagnosis and localization model includes: The error was measured using the cross-entropy loss function. The Adam optimizer was selected, with an initial learning rate of 0.01, 8 training epochs, and a batch size of 16. Dropout was introduced to prevent overfitting of the 5G bearer network fault dynamic diagnosis and localization model.

8. The method for dynamic diagnosis and location of 5G bearer network faults based on alarm correlation as described in claim 6, characterized in that, The following describes how to dynamically diagnose 5G bearer network faults and locate the corresponding network devices: Take the index of the maximum value position in the output vector as the output. If the output result is r, then the r / 4+1th alarm vector among the 5 alarm vectors causes a fault in the 5G bearer network. Find the location of the corresponding network device from the r / 4+1th alarm vector. The fault category is the r%4th fault.

9. An electronic device, characterized in that, include: Memory, used to store executable computer programs; The processor, when executing an executable computer program stored in a memory, implements the 5G bearer network fault dynamic diagnosis and location method based on alarm association as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the method for dynamic diagnosis and location of 5G bearer network faults based on alarm association as described in any one of claims 1 to 8.

Images