Communication fault prediction diagnosis method and device based on AI, equipment and medium

Abstract

The invention relates to an AI-based communication fault prediction diagnosis method and device, equipment and a medium. The method comprises the following steps: acquiring multi-source communication data, and performing space-time alignment fusion feature processing on the multi-source communication data to obtain fault fusion features; performing feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault association feature matrix; independently checking the fault causal feature matrix and the fault association feature matrix to obtain a checking result, and constructing a fault causal graph according to the checking result; according to the fault causal graph, calculating a node PageRank value in the fault causal graph, determining a root cause fault node, and obtaining a fault diagnosis result; the fault diagnosis result comprises a fault type and a fault position. By adopting the method, the causal and incidence relation between the faults can be accurately analyzed, fuzziness of a traditional diagnosis method is avoided, the efficiency and precision of fault diagnosis are improved, and the troubleshooting time and labor cost are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of communication fault diagnosis, and in particular relates to an AI-based communication fault prediction and diagnosis method, device, equipment and medium. Background Art

[0002] With the rapid development and large-scale deployment of next-generation information and communication technologies such as 5G, the Internet of Things, and cloud computing, modern communication networks have become unprecedentedly complex and large-scale. The surge in the number of network devices, increasingly dense connections, and highly diverse service types have posed significant challenges to network operations and maintenance. Traditional communication network fault diagnosis relies primarily on rule bases built from expert experience, alarm correlation engines, and threshold-based monitoring and alarm systems. These approaches typically identify and locate faults based on pre-set logical rules and single-dimensional thresholds for key performance indicators.

[0003] Traditionally, operations personnel manually analyze alarms, check device logs and performance data, and perform fault correlation and root cause analysis based on empirical rules. This approach relies heavily on the completeness of expert knowledge and the accuracy of pre-defined rules.

[0004] However, current diagnostic methods have significant limitations. First, they fail to effectively integrate multi-dimensional, heterogeneous data source information such as network topology, traffic load, device logs, and environmental parameters, making it difficult to capture subtle, cross-domain correlation changes before a fault occurs. More critically, when faced with multiple concurrent fault scenarios, traditional correlation analysis based on fixed rules is easily ineffective. The rule base is difficult to cover all complex concurrent fault combination patterns, and the alarms generated by different faults may interfere with or mask each other, resulting in confusing diagnostic logic, significantly increasing the risk of misjudging the fault type, and extending fault recovery time. Summary of the Invention

[0005] Based on this, it is necessary to provide an AI-based communication fault prediction and diagnosis method, device, equipment and medium that can improve the accuracy of fault type judgment in response to the above technical problems.

[0006] In a first aspect, the present application provides an AI-based communication fault prediction and diagnosis method, comprising:

[0007] Acquire multi-source communication data and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features;

[0008] Perform feature decoupling on the fault fusion features to obtain the fault causal feature matrix and the fault correlation feature matrix;

[0009] Conduct independence condition tests on the fault causal feature matrix and the fault correlation feature matrix respectively, obtain the test results, and construct a fault causal graph based on the test results;

[0010] According to the fault cause-effect graph, the PageRank values ​​of the nodes in the fault cause-effect graph are calculated, the root cause fault node is determined, and the fault diagnosis result is obtained; the fault diagnosis result includes the fault type and fault location.

[0011] In one embodiment, the multi-source communication data includes network traffic data, device log data, and environmental parameter data;

[0012] Perform spatiotemporal alignment fusion feature processing on multi-source communication data to obtain spatiotemporal fusion features, including:

[0013] The time series analysis method based on wavelet transform is used to decompose network traffic data into subsequences of different frequencies;

[0014] By transforming the subsequence at different scales and translation parameters, the network traffic data features at different scales are extracted to obtain the network traffic feature vector;

[0015] Convert device log data into vector representation to obtain device log feature vector;

[0016] Extract the time series characteristics of environmental parameter data to obtain the environmental parameter feature vector;

[0017] The network traffic feature vector, device log feature vector and environmental parameter feature vector are temporally and spatially aligned and fused to obtain the fault fusion feature.

[0018] In one embodiment, feature decoupling is performed on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix, including:

[0019] The fault fusion features are input into the feature decoupling network model for feature decoupling processing to obtain the fault causal feature matrix and the fault correlation feature matrix;

[0020] The feature decoupling network model includes an input layer, an encoder, a first decoder, a second decoder, a discriminator, and an output layer;

[0021] The input layer is used to input fault fusion features;

[0022] The encoder is used to map the fault fusion features into the latent space to obtain the latent representation;

[0023] The first decoder is used to learn the fault causal features and obtain the fault causal matrix;

[0024] The second decoder is used to learn fault correlation features and obtain a fault correlation matrix;

[0025] The discriminator is used to distinguish causal features from correlation features;

[0026] The output layer is used to output the fault causal feature matrix and the fault correlation matrix.

[0027] In one embodiment, an independence condition test is performed on the fault causal feature matrix and the fault correlation feature matrix respectively to obtain the test results, and a fault causal graph is constructed based on the test results, including:

[0028] For the fault causal feature matrix, calculate the Hilbert-Schmidt independence index between the feature vectors of two fault nodes;

[0029] If the Hilbert-Schmidt independence index is greater than the preset causal threshold, a causal judgment result is obtained; the causal judgment result is used to indicate that there is a causal relationship between the two fault nodes;

[0030] For the fault correlation feature matrix, calculate the mutual information value between the feature vectors of two fault nodes;

[0031] If the mutual information value is greater than the preset correlation threshold, the correlation judgment result is obtained; the correlation judgment result is used to indicate that there is a correlation relationship between the two fault nodes;

[0032] A fault causality graph is constructed based on the causal judgment results and the correlation judgment results; the nodes in the fault causality graph represent fault nodes; and the edges in the fault causality graph represent causal relationships or correlation relationships.

[0033] In one embodiment, according to the fault cause-effect graph, calculating the PageRank value of the node in the fault cause-effect graph, determining the root cause fault node, and obtaining the fault diagnosis result include:

[0034] Preprocess the fault cause-effect graph, remove isolated nodes and self-loop edges, and obtain the preprocessed fault cause-effect graph;

[0035] The edge weights of the pre-processed fault causal graph are adjusted to obtain a fault enhanced causal graph; the edge weight adjustment is performed by assigning weights based on the Hilbert-Schmidt independence index and the mutual information value;

[0036] According to the fault reinforcement causal graph, the PageRank value of each fault node in the fault reinforcement causal graph is calculated, and the fault node with the highest PageRank value is taken as the root cause fault node;

[0037] The fault diagnosis result is obtained according to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault causality graph.

[0038] In one embodiment, the fault diagnosis result includes the fault type and the fault location;

[0039] According to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault reinforcement causal graph, the fault diagnosis results are obtained, including:

[0040] Perform node representation extraction on the root cause failure node to obtain a unique node identifier;

[0041] Query the preset graph database based on the unique node identifier to obtain the corresponding fault type;

[0042] Perform fault type mapping based on the fault type and the preset fault knowledge base to obtain a standardized fault description;

[0043] Obtain fault diagnosis results based on standardized fault descriptions.

[0044] In one embodiment, the method further comprises:

[0045] Match the corresponding fault repair measures according to the fault diagnosis results;

[0046] Generate a fault repair order based on the fault diagnosis results and fault repair measures;

[0047] Send a fault repair ticket to a preset device; the preset device is used to display the fault repair ticket to prompt maintenance personnel to perform fault diagnosis and repair measures.

[0048] In a second aspect, the present application also provides an AI-based communication fault prediction and diagnosis device, comprising:

[0049] The fault feature module is used to obtain multi-source communication data and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features;

[0050] The fault causal association module is used to perform feature decoupling processing on the fault fusion features to obtain the fault causal feature matrix and the fault correlation feature matrix;

[0051] The fault cause-effect diagram module is used to perform independence condition tests on the fault cause-effect feature matrix and the fault correlation feature matrix respectively, obtain the test results, and construct the fault cause-effect diagram based on the test results;

[0052] The fault diagnosis module is used to calculate the PageRank value of the nodes in the fault cause-effect graph according to the fault cause-effect graph, determine the root cause fault node, and obtain the fault diagnosis result; the fault diagnosis includes the fault type and fault location.

[0053] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the first aspect.

[0054] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the first aspect when the computer program is executed by a processor.

[0055] The above-mentioned AI-based communication fault prediction and diagnosis method, device, computer equipment and storage medium effectively integrate the scattered communication data through the spatiotemporal alignment and fusion feature processing of multi-source data, extract comprehensive and representative fault features, and solve the problem of insufficient information from a single data source; feature decoupling and independence condition testing realize in-depth analysis and causal relationship modeling of fault features, clarify the key factors and related factors of the fault, and avoid the diagnostic ambiguity caused by feature mixing in traditional methods; the root cause fault node determination based on the PageRank algorithm can quickly and accurately locate the root cause of the fault, improve the efficiency and accuracy of fault diagnosis, and reduce troubleshooting time and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 1 is a flow chart of an AI-based communication fault prediction and diagnosis method in one embodiment;

[0058] Figure 2 1 is a schematic structural diagram of an AI-based communication fault prediction and diagnosis device in one embodiment;

[0059] Figure 3 FIG. 1 is a schematic diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0061] In one embodiment, Figure 1 As shown, a communication fault prediction and diagnosis method based on AI is provided. This embodiment uses the method applied to a communication fault diagnosis terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0062] S101: Acquire multi-source communication data, and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features.

[0063] Exemplarily, the communication fault diagnosis terminal acquires multi-source communication data through sensors and interfaces deployed in different communication devices and network nodes. Multi-source communication data includes multi-dimensional information such as network traffic, device status parameters, signal strength, and transmission delay. The acquired multi-source communication data may come from collection points in different time and space, resulting in inconsistent timestamps and inconsistent spatial coordinates. The communication fault diagnosis terminal uses a time synchronization algorithm (such as the Network Time Protocol (NTP)) to time-align the multi-source communication data, unifying the data collected at different times to the same time reference. At the same time, it calibrates the spatially related data through spatial coordinate system transformation to achieve spatiotemporal alignment and obtain spatiotemporal aligned multi-source data. The communication fault diagnosis terminal fuses the spatiotemporal aligned multi-source data to obtain multi-source fused data. The communication fault diagnosis terminal extracts features from the multi-source fused data to obtain fault fusion features.

[0064] S102 , performing feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix.

[0065] Exemplarily, the communication fault diagnosis terminal uses feature decoupling technology to perform feature decoupling processing on the fault fusion features, and decomposes the fault fusion features into a set of sub-features with specific semantics based on the causal relationship and correlation relationship between the features; the communication fault diagnosis terminal identifies the features corresponding to the key factors that directly cause the fault, such as equipment hardware failure, software vulnerabilities, etc., as fault causal features, and constructs a fault causal feature matrix; the communication fault diagnosis terminal classifies the features corresponding to factors that are indirectly related to the fault, such as ambient temperature changes, network load fluctuations, etc., as fault correlation features, and constructs a fault correlation matrix.

[0066] S103 , performing independence condition tests on the fault causal feature matrix and the fault correlation feature matrix respectively, obtaining test results, and constructing a fault causal graph based on the test results.

[0067] For example, the communication fault diagnosis terminal performs an independence condition test on the fault causal feature matrix and the fault correlation feature matrix, testing the independence between each feature variable in the matrix. During the test, the correlation coefficient and mutual information value between the feature variables are calculated to determine whether the independence condition is met. If there is a significant dependency relationship between two feature variables, it is considered that there is a causal relationship or correlation between the two feature variables. Based on the test results, a fault causal graph is constructed with the feature variables as nodes and the dependency relationships as edges. The fault causal graph clearly displays the causal relationship and correlation structure between each fault feature. Among them, the independence condition test methods include chi-square test and mutual information test.

[0068] S104: Calculate the PageRank values ​​of the nodes in the fault cause-effect graph according to the fault cause-effect graph, determine the root cause fault node, and obtain the fault diagnosis result.

[0069] Among them, the PageRank value reflects the influence and importance of the node in the graph; the fault diagnosis results include the fault type and fault location.

[0070] For example, in the fault causal graph, nodes with high PageRank values ​​have a stronger influence on the occurrence and propagation of faults and are more likely to be the root cause of the faults. The communication fault diagnosis terminal calculates the PageRank value of each node and selects the node with the highest PageRank value as the root cause fault node. Based on the corresponding feature information of the root cause fault node, the specific type of fault (such as equipment failure, link interruption and software abnormality) and the location of the fault (such as the specific device number and network node address) are determined to obtain the fault diagnosis result.

[0071] The above-mentioned AI-based communication fault prediction and diagnosis method effectively integrates scattered communication data through spatiotemporal alignment and fusion feature processing of multi-source data, extracts comprehensive and representative fault features, and solves the problem of insufficient information from a single data source; feature decoupling and independence condition testing realize in-depth analysis and causal relationship modeling of fault features, clarify the key factors and correlation factors of the fault, and avoid the diagnostic ambiguity caused by feature mixing in traditional methods; the root cause fault node determination based on the PageRank algorithm can quickly and accurately locate the root cause of the fault, improve the efficiency and accuracy of fault diagnosis, and can timely and accurately predict and diagnose faults in complex communication network environments, reducing troubleshooting time and maintenance costs.

[0072] Optionally, the multi-source communication data includes network traffic data, device log data, and environmental parameter data; performing spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain spatiotemporal fusion features includes the following steps:

[0073] S201, using a time series analysis method based on wavelet transform to decompose network traffic data into subsequences of different frequencies.

[0074] Among them, time series analysis methods based on wavelet transform include Daubechies wavelet and Haar wavelet.

[0075] For example, the communication fault diagnosis terminal uses a wavelet-based time series analysis method to perform a discrete wavelet transform on network traffic data, breaking it down into subsequences of varying frequencies. These subsequences consist of low-frequency components and high-frequency components. The low-frequency components reflect the long-term trends and overall characteristics of network traffic, while the high-frequency components capture short-term fluctuations and sudden changes in network traffic. This separation of frequency components within network traffic data lays the foundation for subsequent feature extraction at different scales.

[0076] S202 , extracting features of network traffic data at different scales by transforming the subsequences at different scales and translation parameters, and obtaining a network traffic feature vector.

[0077] Specifically, after obtaining subsequences of different frequencies, the communication fault diagnosis terminal performs wavelet transforms on the subsequences using different scale and translation parameters. The scale parameter controls the degree of scaling of the wavelet function, with different scales corresponding to different frequency ranges and time resolutions. The translation parameter controls the position of the wavelet function on the time axis. By adjusting the scale and translation parameters, network traffic data is analyzed at multiple scales, capturing its characteristics at different time scales and frequency ranges. Specifically, at larger scales, cyclical changes and long-term trends in network traffic can be analyzed; at smaller scales, sudden anomalies and transient changes in network traffic can be detected. The communication fault diagnosis terminal extracts network traffic features from the multi-scale transformation results. Network traffic features reflect the characteristic information of network traffic at different scales, providing a rich feature representation for subsequent fault diagnosis.

[0078] S203: Convert the device log data into a vector representation to obtain a device log feature vector.

[0079] Exemplarily, the communication fault diagnosis terminal preprocesses the device log, including operations such as character denoising, word segmentation, and part-of-speech tagging, to convert the log text into a structured text sequence. Using word embedding technology in natural language processing (such as Word2Vec (word vector model), GloVe (Global Vectors for Word Representation, a word representation tool for global word frequency statistics), BERT (Bidirectional Encoder Representations from Transformers, a bidirectional encoder representation method), etc.), each word in the device log is mapped to a low-dimensional vector representation, and an average pooling aggregation method is used to combine the vector representations of each word in the log text into an overall vector, namely the device log feature vector. The device log vector retains the semantic information and key features in the device log and can reflect the device operating status and fault information.

[0080] S204: extracting time series features of the environmental parameter data to obtain an environmental parameter feature vector.

[0081] For example, the communication fault diagnosis terminal preprocesses environmental parameter data, including operations such as missing value filling, outlier processing, and normalization. The preprocessed environmental parameter data is analyzed using the Autoregressive Integrated Moving Average (ARIMA) model, which combines the temporal and statistical features to form an environmental parameter feature vector. This vector reflects the temporal variation patterns and characteristics of the environmental parameters, providing a basis for subsequent analysis of the impact of environmental factors on communication faults.

[0082] S205 , performing spatiotemporal alignment and fusion processing on the network traffic feature vector, the device log feature vector, and the environmental parameter feature vector to obtain a fault fusion feature.

[0083] Exemplarily, the communication fault diagnosis terminal performs timestamp and spatial coordinate mapping processing on the network traffic feature vector, the device log feature vector and the environmental parameter feature vector to obtain three time-space aligned feature vectors; the three time-space aligned feature vectors are spliced ​​and fused to obtain the fault fusion feature.

[0084] Optionally, performing feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix includes the following steps:

[0085] S301 , inputting the fault fusion features into a feature decoupling network model for feature decoupling processing to obtain a fault causal feature matrix and a fault correlation feature matrix.

[0086] Exemplarily, the communication fault diagnosis terminal inputs the fault fusion features into a trained feature decoupling network model for feature decoupling processing to obtain a fault causal feature matrix and a fault correlation feature matrix.

[0087] Among them, the feature decoupling network model includes an input layer, an encoder, a first decoder, a second decoder, a discriminator and an output layer.

[0088] Among them, the feature decoupling network model is a multi-branch architecture based on deep learning, including an input layer, an encoder, a first decoder, a second decoder, a discriminator and an output layer.

[0089] The input layer is used to input fault fusion features.

[0090] The input layer is used to convert and preprocess the input fault fusion features to obtain the fault fusion feature input representation for subsequent input.

[0091] The encoder is used to map the fault fusion features into the latent space to obtain the latent representation.

[0092] Exemplarily, the encoder performs a nonlinear transformation on the input fault fusion features, maps the fault fusion features into a latent space, and obtains a latent representation; wherein the latent space contains a compressed representation of the fault fusion features; the encoder learns the intrinsic structure and pattern of the fault fusion features through a multi-layer neural network and converts them into a more abstract and essential representation.

[0093] The first decoder is used to learn the fault causal features and obtain the fault causal matrix.

[0094] Exemplarily, the first decoder learns fault causal features from the latent representation, and reconstructs a feature matrix closely related to the fault causal relationship, namely, the fault causal matrix, by performing a decoding operation on the latent representation. Each element in the fault causal matrix represents the correlation strength and direction between different fault causal features.

[0095] The second decoder is used to learn the fault correlation features and obtain the fault correlation matrix.

[0096] Exemplarily, the second decoder learns the fault correlation features from the latent representation, and obtains a fault correlation matrix by performing a decoding operation on the latent representation. The fault correlation matrix reflects the relationship between the fault correlation features.

[0097] The discriminator is used to distinguish causal features from correlation features.

[0098] Exemplarily, the discriminator distinguishes the causal features output by the first decoder and the correlation features output by the second decoder through adversarial training, prompting the first decoder and the second decoder to focus on learning the causal features and the correlation features respectively, thereby achieving feature decoupling.

[0099] The output layer is used to output the fault causal feature matrix and the fault correlation matrix.

[0100] Exemplarily, the output layer organizes and formats the fault causality matrix and the fault correlation matrix generated by the first decoder and the second decoder, and outputs the final fault causality feature matrix and the fault correlation feature matrix.

[0101] Optionally, performing independence condition tests on the fault causal feature matrix and the fault correlation feature matrix respectively to obtain test results, and constructing a fault causal graph based on the test results, including the following steps:

[0102] S401, for the fault causal characteristic matrix, calculate the Hilbert-Schmidt independence index between the characteristic vectors of two fault nodes.

[0103] Exemplarily, the communication fault diagnosis terminal selects the eigenvectors corresponding to any two fault nodes in the fault causal feature matrix and measures the independence between the eigenvectors of the two fault nodes by calculating the Hilbert-Schmidt independence index. Specifically, the communication fault diagnosis terminal maps the eigenvectors of any two fault nodes to a reproducing kernel Hilbert space and obtains the Hilbert-Schmidt independence index by calculating the Hilbert-Schmidt norm of the covariance operator of the eigenvectors of the two fault nodes in the reproducing kernel Hilbert space. The Hilbert-Schmidt independence index is used to quantify the degree of correlation between the eigenvectors of the two fault nodes. The larger the value of the Hilbert-Schmidt independence index, the stronger the dependence between the eigenvectors of the two fault nodes. The smaller the value of the Hilbert-Schmidt independence index, the stronger the independence between the eigenvectors of the two fault nodes.

[0104] S402: If the Hilbert-Schmidt independence index is greater than a preset causal threshold, a causal judgment result is obtained.

[0105] Among them, the preset causal threshold is a critical value determined through statistical analysis and models based on a large amount of historical communication failure data and diagnostic experience. It is used to determine whether there is a causal relationship between two fault nodes; the causal judgment result is used to characterize the existence of a causal relationship between the two fault nodes.

[0106] Exemplarily, the communication fault diagnosis terminal compares the calculated Hilbert-Schmidt independence index with the preset causal threshold. When the Hilbert-Schmidt independence index is greater than the preset causal threshold, it indicates that the characteristic vectors of the two fault nodes are highly dependent, satisfying the judgment condition of the existence of a causal relationship, and obtaining a causal judgment result, that is, there is a causal relationship between the two fault nodes.

[0107] S403: Calculate the mutual information value between the feature vectors of two fault nodes for the fault correlation feature matrix.

[0108] Among them, the mutual information value is used to measure the dependency between two fault nodes.

[0109] Exemplarily, the communication fault diagnosis terminal selects the eigenvectors of any two faulty nodes in the fault correlation feature matrix and calculates the mutual information value between the eigenvectors of the two faulty nodes. Specifically, the communication fault diagnosis terminal calculates the mutual information value by calculating the logarithmic expectation of the product of the joint probability distribution and the marginal probability distribution of the random variables represented by the eigenvectors of the two faulty nodes. A larger mutual information value indicates a higher degree of correlation between the eigenvectors of the two faulty nodes; a smaller mutual information value indicates a lower degree of correlation between the eigenvectors of the two faulty nodes.

[0110] S404: If the mutual information value is greater than the preset association threshold, an association determination result is obtained.

[0111] The preset correlation threshold is determined based on the statistical characteristics of the correlation relationship in the historical fault data; the correlation judgment result is used to characterize whether there is a correlation relationship between two fault nodes.

[0112] Exemplarily, the communication fault diagnosis terminal compares the calculated mutual information value with the preset correlation threshold. When the mutual information value is greater than the preset correlation threshold, it indicates that the degree of correlation between the feature vectors of the two fault nodes is high, which meets the judgment criteria for the existence of a correlation relationship, and obtains a correlation judgment result, that is, there is a correlation relationship between the two fault nodes.

[0113] S405: Construct a fault cause-effect diagram based on the cause-effect judgment result and the association judgment result.

[0114] The nodes in the fault causality graph represent fault nodes; the edges in the fault causality graph represent causal relationships or association relationships.

[0115] Exemplarily, the communication fault diagnosis terminal uses the fault node as a node in the fault causal graph and determines the edges in the fault causal graph based on the causal judgment results and the association judgment results. Specifically, if there is a causal relationship between two fault nodes, an edge representing the causal relationship is drawn between the corresponding nodes, and the causal direction is indicated by an arrow; if there is an association relationship between two fault nodes, an edge representing the association relationship is drawn between the corresponding nodes. The causal and association relationships between all fault nodes are presented in a graphical form to construct a complete fault causal graph. The fault causal graph clearly and intuitively presents the causal and association relationships between fault nodes, improving the accuracy and efficiency of communication fault diagnosis and enhancing the reliability of communication fault diagnosis.

[0116] Optionally, according to the fault cause-effect graph, calculating the PageRank values ​​of nodes in the fault cause-effect graph, determining the root cause fault node, and obtaining the fault diagnosis result includes the following steps:

[0117] S501 , preprocessing the fault cause and effect graph, removing isolated nodes and self-loop edges, and obtaining a preprocessed fault cause and effect graph.

[0118] Among them, an isolated node is a node that is not connected to any other node; a self-loop edge is an edge pointing from a node to itself.

[0119] Exemplarily, the communication fault diagnosis terminal preprocesses the constructed fault causal graph, scans the entire fault causal graph, identifies isolated nodes and self-loop edges, directly deletes the isolated nodes and their related data, and removes the self-loop edges to obtain the preprocessed fault causal graph.

[0120] S502 , adjusting edge weights of the pre-processed fault causal graph to obtain a fault reinforcement causal graph.

[0121] Among them, the edge weight adjustment is adjusted by assigning weights based on the Hilbert-Schmidt independence index and the mutual information value.

[0122] Exemplarily, the communication fault diagnosis terminal adjusts the edge weights of the preprocessed fault causal graph based on the calculated Hilbert-Schmidt independence index and mutual information value. Specifically, for causal edges, the communication fault diagnosis terminal assigns weights based on the Hilbert-Schmidt independence index between the feature vectors of the two fault nodes. A larger Hilbert-Schmidt independence index indicates a stronger causal relationship and a higher weight is assigned. For associative edges, the communication fault diagnosis terminal assigns weights based on the mutual information value between the feature vectors of the two fault nodes. A larger mutual information value indicates a higher degree of association and a correspondingly larger edge weight. The communication fault diagnosis terminal reassigns and adjusts the weights of all edges in the fault causal graph to obtain a fault-enhanced causal graph. The fault-enhanced causal graph allows the strength of the relationship between fault nodes to be more intuitively reflected in the graph.

[0123] S503 , calculating the PageRank value of each fault node in the fault reinforcement causal graph according to the fault reinforcement causal graph, and taking the fault node with the highest PageRank value as the root cause fault node.

[0124] Among them, the PageRank value depends on the PageRank values ​​of other nodes pointing to the faulty node and the weight of the edge. The higher the edge weight, the more "importance" is transferred. After multiple iterations, when the PageRank values ​​of all nodes converge, the final PageRank value is obtained.

[0125] Exemplarily, the communication fault diagnosis terminal calculates the PageRank value of the fault node in the fault reinforcement causal graph, and the fault node with the highest PageRank value has the highest influence and importance in the fault reinforcement causal graph, and is most likely to be the root cause of the fault. Therefore, the fault node with the highest PageRank value is determined to be the root cause fault node.

[0126] S504 , obtaining a fault diagnosis result according to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault reinforcement causal graph.

[0127] For example, after determining the root cause fault node, the communication fault diagnosis terminal obtains the fault type information corresponding to the root cause fault node from a pre-established mapping relationship between fault type and node characteristics. Simultaneously, the node's specific location or device number, etc., are located in the fault reinforcement causal graph using the node identifier. The fault type and node identifier information are then integrated to ultimately obtain a fault diagnosis result that includes both the fault type and location. This avoids misjudgments caused by complex information in traditional fault diagnosis, achieves precise location of the fault root cause, and improves the accuracy of communication fault diagnosis.

[0128] Optionally, the fault diagnosis result includes the fault type and the fault location;

[0129] Obtaining a fault diagnosis result based on the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault reinforcement causal graph includes the following steps:

[0130] S601: Perform node representation extraction processing on the root cause failure node to obtain a unique node identifier.

[0131] Exemplarily, the communication fault diagnosis terminal extracts the node identifier of the root cause node. Specifically, the terminal parses the data structure of the root cause node in the fault reinforcement causal graph, extracts the unique identifier information contained in the root cause node, and uses the unique identifier information as a unique node identifier to accurately locate the root cause node. The unique identifier information includes a specific code, device number, and network address.

[0132] S602: Perform query processing in a preset graph database according to the unique node identifier to obtain a corresponding fault type.

[0133] Among them, the preset graph database stores a large amount of correspondence data between node identifiers and fault types.

[0134] Exemplarily, the communication fault diagnosis terminal uses the acquired unique node identifier to perform a query in a pre-set graph database. Using the unique node identifier as a query condition, the pre-set graph database performs an accurate match to find the fault type corresponding to the unique node identifier. Specifically, if the unique node identifier corresponds to a specific device ID, the pre-set graph database returns the possible fault type, preliminarily determining the fault type of the node that is the root cause of the fault.

[0135] S603: Perform fault type mapping processing according to the fault type and a preset fault knowledge base to obtain a standardized fault description.

[0136] Among them, the fault knowledge base contains detailed information of various fault types and standardized description templates.

[0137] Exemplarily, the communication fault diagnosis terminal performs fault type mapping processing in a preset fault knowledge base based on the acquired fault type. Specifically, the communication fault diagnosis terminal matches the acquired fault type with the information in the fault knowledge base and converts the original fault type into a standardized fault description according to unified standards and specifications.

[0138] S604: Obtain a fault diagnosis result based on the standardized fault description.

[0139] For example, the communication fault diagnosis terminal combines the standardized fault description with relevant information of the root cause node to generate a final fault diagnosis result. The fault diagnosis result clearly includes a standardized description of the fault type and the fault location.

[0140] Optionally, the method further comprises the following steps:

[0141] S701, matching corresponding fault repair measures according to the fault diagnosis result.

[0142] For example, after receiving the fault diagnosis results, the communication fault diagnosis terminal performs keyword matching and feature comparison on the results in a pre-set fault repair measure library, retrieving the most appropriate repair measure for the current fault from the library. The library contains a large amount of standard repair procedures, operating steps, required tools, and precautions for different fault types and corresponding locations.

[0143] S702: Generate a fault repair order based on the fault diagnosis result and fault repair measures.

[0144] Among them, the fault repair order contains basic information such as the fault type, fault location, and discovery time, and lists the corresponding repair measures in detail.

[0145] Illustratively, the communication fault diagnosis terminal generates a fault repair order based on the obtained fault diagnosis result and the matched fault repair measures; the fault repair order also adds a unique order number identifier to facilitate subsequent tracking and management.

[0146] S703: Send a fault repair order to the preset device.

[0147] Among them, the preset equipment is used to display the fault repair order to prompt the maintenance personnel to perform fault diagnosis and repair measures.

[0148] For example, after generating a fault repair order, the communication fault diagnosis terminal sends the fault repair order to a preset device through a network communication protocol. After receiving the fault repair order, the preset device will immediately display the contents of the fault repair order in the form of a pop-up window, sound prompt or message push, clearly presenting the fault information and repair requirements, and promptly prompting maintenance personnel to perform fault repair operations in accordance with the measures in the fault repair order.

[0149] The above-mentioned AI-based communication fault prediction and diagnosis method comprehensively integrates multi-dimensional information such as network traffic, equipment logs and environmental parameters through in-depth analysis and fusion of multi-source communication data, overcoming the limitations of a single data source and improving the completeness and accuracy of fault feature extraction; it uses a feature decoupling network model to effectively separate fault features, and combines independence condition tests to construct an accurate fault cause-effect graph, accurately analyzing the cause and effect and correlation relationships between faults, avoiding the ambiguity of traditional diagnostic methods; based on the PageRank algorithm and graph processing technology, it locates the root cause fault node, and improves the efficiency and accuracy of fault diagnosis through standardized fault description and maintenance measure matching; the process from fault diagnosis to repair order generation and sending reduces troubleshooting time and labor costs.

[0150] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0151] Based on the same inventive concept, the embodiments of the present application also provide an AI-based communication fault prediction and diagnosis device for implementing the aforementioned AI-based communication fault prediction and diagnosis method. The implementation solution provided by this device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more AI-based communication fault prediction and diagnosis device embodiments provided below can be found in the limitations of the AI-based communication fault prediction and diagnosis method above, and will not be repeated here.

[0152] In an exemplary embodiment, Figure 2 As shown, a communication fault prediction and diagnosis device 200 based on AI is provided, comprising:

[0153] The fault feature module 201 is used to obtain multi-source communication data and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features;

[0154] The fault causal correlation module 202 is used to perform feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix;

[0155] The fault cause-effect diagram module 203 is used to perform independence condition tests on the fault cause-effect feature matrix and the fault correlation feature matrix respectively, obtain test results, and construct a fault cause-effect diagram based on the test results;

[0156] The fault diagnosis module 204 is used to calculate the PageRank value of the nodes in the fault cause and effect graph according to the fault cause and effect graph, determine the root cause fault node, and obtain the fault diagnosis result; the fault diagnosis includes the fault type and fault location.

[0157] Furthermore, the multi-source communication data includes network traffic data, device log data, and environmental parameter data;

[0158] The fault characteristic module 201 is further configured to:

[0159] The time series analysis method based on wavelet transform is used to decompose network traffic data into subsequences of different frequencies;

[0160] By transforming the subsequence at different scales and translation parameters, the network traffic data features at different scales are extracted to obtain the network traffic feature vector;

[0161] Convert device log data into vector representation to obtain device log feature vector;

[0162] Extract the time series characteristics of environmental parameter data to obtain the environmental parameter feature vector;

[0163] The network traffic feature vector, device log feature vector and environmental parameter feature vector are temporally and spatially aligned and fused to obtain the fault fusion feature.

[0164] Furthermore, the fault causal association module 202 is further configured to:

[0165] The fault fusion features are input into the feature decoupling network model for feature decoupling processing to obtain the fault causal feature matrix and the fault correlation feature matrix;

[0166] The feature decoupling network model includes an input layer, an encoder, a first decoder, a second decoder, a discriminator, and an output layer;

[0167] The input layer is used to input fault fusion features;

[0168] The encoder is used to map the fault fusion features into the latent space to obtain the latent representation;

[0169] The first decoder is used to learn the fault causal features and obtain the fault causal matrix;

[0170] The second decoder is used to learn fault correlation features and obtain a fault correlation matrix;

[0171] The discriminator is used to distinguish causal features from correlation features;

[0172] The output layer is used to output the fault causal feature matrix and the fault correlation matrix.

[0173] Furthermore, the fault cause and effect diagram module 203 is further configured to:

[0174] For the fault causal feature matrix, calculate the Hilbert-Schmidt independence index between the feature vectors of two fault nodes;

[0175] If the Hilbert-Schmidt independence index is greater than the preset causal threshold, a causal judgment result is obtained; the causal judgment result is used to indicate that there is a causal relationship between the two fault nodes;

[0176] For the fault correlation feature matrix, calculate the mutual information value between the feature vectors of two fault nodes;

[0177] If the mutual information value is greater than the preset correlation threshold, the correlation judgment result is obtained; the correlation judgment result is used to indicate that there is a correlation relationship between the two fault nodes;

[0178] A fault causality graph is constructed based on the causal judgment results and the correlation judgment results; the nodes in the fault causality graph represent fault nodes; and the edges in the fault causality graph represent causal relationships or correlation relationships.

[0179] Furthermore, the fault diagnosis module 204 is further configured to:

[0180] Preprocess the fault cause-effect graph, remove isolated nodes and self-loop edges, and obtain the preprocessed fault cause-effect graph;

[0181] The edge weights of the pre-processed fault causal graph are adjusted to obtain a fault enhanced causal graph; the edge weight adjustment is performed by assigning weights based on the Hilbert-Schmidt independence index and the mutual information value;

[0182] According to the fault reinforcement causal graph, the PageRank value of each fault node in the fault causal graph is calculated, and the fault node with the highest PageRank value is taken as the root cause fault node;

[0183] The fault diagnosis result is obtained according to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault causality graph.

[0184] Furthermore, the fault diagnosis module 204 is further configured to:

[0185] Perform node representation extraction on the root cause failure node to obtain a unique node identifier;

[0186] Query the preset graph database based on the unique node identifier to obtain the corresponding fault type;

[0187] Perform fault type mapping based on the fault type and the preset fault knowledge base to obtain a standardized fault description;

[0188] Obtain fault diagnosis results based on standardized fault descriptions.

[0189] Furthermore, the device further comprises:

[0190] Maintenance measure matching module, used to match corresponding fault maintenance measures according to fault diagnosis results;

[0191] The maintenance order generation module is used to generate a fault maintenance order based on the fault diagnosis results and fault maintenance measures;

[0192] The maintenance order sending module is used to send a fault maintenance order to a preset device; the preset device is used to display the fault maintenance order to prompt maintenance personnel to perform fault diagnosis and maintenance measures.

[0193] In one embodiment, Figure 3 A computer device 300 is provided, comprising:

[0194] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301; the memory storing application code executable by the at least one processor, the application code being executed by the at least one processor to enable the at least one processor to perform the steps of the aforementioned AI-based communication fault prediction and diagnosis method;

[0195] The computer device may further include: a sensor 303;

[0196] The processor 301, the memory 301 and the sensor 303 may be connected via a bus 304 or other means. In the figure, the bus 304 is used as an example. Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0197] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0198] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0199] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A communication fault prediction and diagnosis method based on AI, characterized in that: The method comprises: Acquire multi-source communication data, and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features; Performing feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix; Performing independence condition tests on the fault causal feature matrix and the fault correlation feature matrix respectively to obtain test results, and constructing a fault causal graph based on the test results; According to the fault cause-effect graph, the PageRank values ​​of the nodes in the fault cause-effect graph are calculated, the root cause fault node is determined, and a fault diagnosis result is obtained; the fault diagnosis result includes the fault type and the fault location.

2. The method according to claim 1, characterized in that The multi-source communication data includes network traffic data, device log data and environmental parameter data; The performing spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain spatiotemporal fusion features includes: Using a time series analysis method based on wavelet transform, the network traffic data is decomposed into subsequences of different frequencies; By transforming the subsequence at different scales and translation parameters, the features of the network traffic data at different scales are extracted to obtain a network traffic feature vector; Converting the device log data into a vector representation to obtain a device log feature vector; Extracting time series features of the environmental parameter data to obtain an environmental parameter feature vector; The network traffic feature vector, the device log feature vector, and the environmental parameter feature vector are subjected to spatiotemporal alignment and fusion processing to obtain a fault fusion feature.

3. The method according to claim 1, characterized in that The performing feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix includes: Inputting the fault fusion features into a feature decoupling network model for feature decoupling processing to obtain a fault causal feature matrix and a fault correlation feature matrix; The feature decoupling network model includes an input layer, an encoder, a first decoder, a second decoder, a discriminator and an output layer; The input layer is used to input fault fusion features; The encoder is used to map the fault fusion features to a latent space to obtain a latent representation; The first decoder is used to learn fault causal features to obtain a fault causal matrix; The second decoder is used to learn fault correlation features to obtain a fault correlation matrix; The discriminator is used to distinguish causal features from correlation features; The output layer is used to output a fault causal feature matrix and a fault correlation matrix.

4. The method according to claim 1, wherein The independence condition test is performed on the fault causal feature matrix and the fault correlation feature matrix respectively to obtain the test results, and a fault causal graph is constructed according to the test results, including: For the fault causal characteristic matrix, calculating the Hilbert-Schmidt independence index between the characteristic vectors of two fault nodes; If the Hilbert-Schmidt independence index is greater than a preset causal threshold, a causal judgment result is obtained; the causal judgment result is used to indicate that there is a causal relationship between the two fault nodes; For the fault correlation feature matrix, calculating the mutual information value between the feature vectors of two fault nodes; If the mutual information value is greater than a preset correlation threshold, an association judgment result is obtained; the association judgment result is used to indicate that there is an association relationship between the two fault nodes; A fault cause-effect graph is constructed based on the causal judgment result and the association judgment result; the nodes in the fault cause-effect graph represent fault nodes; and the edges in the fault cause-effect graph represent causal relationships or association relationships.

5. The method according to claim 4, characterized in that The step of calculating the PageRank values ​​of nodes in the fault cause-effect graph according to the fault cause-effect graph, determining the root cause fault node, and obtaining the fault diagnosis result includes: Preprocessing the fault cause-effect graph to remove isolated nodes and self-loop edges, thereby obtaining a preprocessed fault cause-effect graph; Adjusting edge weights on the pre-processed fault causal graph to obtain a fault reinforcement causal graph; the edge weight adjustment is performed by assigning weights based on the Hilbert-Schmidt independence index and the mutual information value; According to the fault enhanced causal graph, the PageRank value of each fault node in the enhanced fault causal graph is calculated, and the fault node with the highest PageRank value is taken as the root cause fault node; A fault diagnosis result is obtained according to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault reinforcement causal graph.

6. The method according to claim 5, characterized in that The fault diagnosis result includes the fault type and fault location; Obtaining a fault diagnosis result according to the fault type corresponding to the root fault node and the node identifier of the root fault node in the fault reinforcement causal graph includes: Performing node representation extraction processing on the root cause failure node to obtain a unique node identifier; Perform query processing in a preset graph database according to the unique node identifier to obtain a corresponding fault type; Perform fault type mapping processing according to the fault type and a preset fault knowledge base to obtain a standardized fault description; A fault diagnosis result is obtained according to the standardized fault description.

7. The method according to claim 1, characterized in that The method further comprises: Matching corresponding fault repair measures according to the fault diagnosis results; generating a fault repair order based on the fault diagnosis result and the fault repair measures; The fault repair sheet is sent to a preset device; the preset device is used to display the fault repair sheet to prompt the maintenance personnel to perform the fault diagnosis and maintenance measures.

8. A communication fault prediction and diagnosis device based on AI, characterized in that: The device comprises: A fault feature module is used to obtain multi-source communication data and perform spatiotemporal alignment fusion feature processing on the multi-source communication data to obtain fault fusion features; A fault causal association module is used to perform feature decoupling processing on the fault fusion features to obtain a fault causal feature matrix and a fault correlation feature matrix; A fault cause-effect diagram module is used to perform independence condition tests on the fault cause-effect feature matrix and the fault correlation feature matrix respectively, obtain test results, and construct a fault cause-effect diagram according to the test results; The fault diagnosis module is used to calculate the PageRank value of the node in the fault cause and effect diagram according to the fault cause and effect diagram, determine the root cause fault node, and obtain the fault diagnosis result; the fault diagnosis includes the fault type and fault location.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images