Artificial intelligence based optical network fault detection system

Abstract

The application discloses an optical network fault detection system based on artificial intelligence, and belongs to the technical field of optical network intelligent operation and maintenance, and comprises a data acquisition module, an intelligent fault detection module, an influence range evaluation module and a detection report generation module.The application adopts a multi-modal adaptive fault detection method to perform intelligent fault detection, comprehensively links operation data, equipment operation logs and network structure information, dynamically identifies potential abnormal nodes, reduces false positive rates, and discovers possible faults that may affect network operation in advance, and provides quantifiable abnormal credibility scores;an optical network fault multi-dimensional influence range evaluation method is used for influence range evaluation, on the basis of considering network structure, node abnormality degree and business traffic, dynamically predicting nodes and links that may be affected by the fault, quantifying the influence range and severity of the fault, and providing intuitive and operable decision basis for operation and maintenance personnel, thereby effectively reducing the risk of fault diffusion and optimizing network maintenance strategies.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of optical network intelligent operation and maintenance, and particularly relates to an optical network fault detection system based on artificial intelligence. BACKGROUND

[0002] The optical network fault detection system based on artificial intelligence utilizes multi-modal data fusion, graph neural networks and dynamic anomaly analysis technology to intelligently monitor and identify faults of the optical network running state, aims to accurately detect faults and quantify the influence of the faults on the network structure and services, and provides scientific and reliable decision-making basis for the operation and maintenance personnel, so that efficient maintenance and stable operation of the optical network are realized. However, in the existing optical network fault detection process, there are technical problems that the device state information is scattered and various in type, and single data source monitoring is difficult to accurately capture faults, leading to false positives; the fault may quickly spread through the network to affect a large number of nodes and services, and single node abnormal information is difficult to accurately determine the overall influence range, and the importance of different nodes and links to services is quite different, and traditional methods are difficult to quantify the actual influence degree of the fault. SUMMARY

[0003] In view of the above problems, in order to overcome the defects of the prior art, the optical network fault detection system based on artificial intelligence is provided. In the existing optical network fault detection process, there are technical problems that the device state information is scattered and various in type, and single data source monitoring is difficult to accurately capture faults, leading to false positives. The present solution creatively adopts a multi-modal adaptive fault detection method for intelligent fault detection, comprehensively considers link running data, device running logs and network structure information, dynamically identifies potential abnormal nodes, reduces the false positive rate and discovers possible faults affecting network operation in advance, and provides quantifiable abnormal confidence scores. In the existing optical network fault detection process, there are technical problems that the fault may quickly spread through the network to affect a large number of nodes and services, and single node abnormal information is difficult to accurately determine the overall influence range, and the importance of different nodes and links to services is quite different, and traditional methods are difficult to quantify the actual influence degree of the fault. The present solution creatively adopts an optical network fault multi-dimensional influence range evaluation method for influence range evaluation, dynamically predicts nodes and links possibly affected by the fault on the basis of considering network structure, node abnormality degree and service traffic, quantifies the influence range and severity of the fault, and provides intuitive and operable decision-making basis for the operation and maintenance personnel, thereby effectively reducing the fault diffusion risk and optimizing the network maintenance strategy.

[0004] The technical solutions adopted by the present application are as follows: The optical network fault detection system based on artificial intelligence provided by the present application comprises a data acquisition module, an intelligent fault detection module, an influence range evaluation module and a detection report generation module.

[0005] The data acquisition module is configured to acquire data, obtain optical network operation data through data acquisition, and send the optical network operation data to the intelligent fault detection module.

[0006] The intelligent fault detection module is configured to perform intelligent fault detection based on the optical network operation data, adopt a multi-modal adaptive fault detection method to obtain optical network fault detection information, and send the optical network fault detection information to the influence range evaluation module and the detection report generation module.

[0007] The influence range evaluation module is configured to evaluate the influence range based on the optical network operation data and the optical network fault detection information, adopt an optical network fault multi-dimensional influence range evaluation method to obtain fault influence range evaluation information, and send the fault influence range evaluation information to the detection report generation module.

[0008] The detection report generation module is configured to generate a detection report.

[0009] Further, the data acquisition specifically includes acquiring optical network operation data and performing preprocessing, and the optical network operation data includes link operation data, device operation logs, and optical network topology structure information.

[0010] The link operation data includes OSNR, BER, optical power, link delay, link traffic, link bandwidth capacity, link bandwidth occupancy, link physical length, and optical power loss.

[0011] Further, the intelligent fault detection specifically includes performing intelligent fault detection based on the optical network operation data, adopting a multi-modal adaptive fault detection method to obtain optical network fault detection information, and includes the following steps: feature cross-modal fusion, graph time series anomaly detection, false alarm correction, and optical network fault detection information generation.

[0012] The feature cross-modal fusion specifically includes constructing an optical network topology graph by taking each physical device in the optical network as a graph node and each link as an edge based on the optical network topology structure information, calculating time series signal features, log features, and topology features, and splicing to obtain cross-modal fusion features as node features of the optical network topology graph to construct a node feature matrix.

[0013] The time series signal feature specifically includes normalizing the link operation data corresponding to each edge according to time steps, performing one-dimensional convolution to extract link time series patterns, and then performing average pooling on the link time series patterns corresponding to all edges connected to the same node to obtain the time series signal feature.

[0014] The log feature, specifically, by BERT encoding the device running log corresponding to each node, obtaining the log feature, when the device running log corresponding to the node is missing, using the neighbor log feature mean to fill in;

[0015] The topology feature, specifically, extracting the topology feature from the optical network topology structure information, the topology feature including node level, neighbor level and global topology statistical feature;

[0016] The graph time series anomaly detection, specifically, taking the optical network topology graph and node feature matrix as input, modeling the spatial dependence through the graph attention network model to obtain the spatial dependence enhanced feature matrix; modeling the time series through the time series Transformer encoder on the spatial dependence enhanced feature matrix to obtain the time series prediction feature matrix; calculating the prediction error and the domain consistency error, and generating the node comprehensive anomaly score by weighting;

[0017] The false alarm correction, specifically, by constructing a standard self-supervised mask reconstruction model, randomly masking and predicting reconstruction on the spatial dependence enhanced feature matrix to obtain the reconstruction feature matrix, and then calculating the reconstruction error; combining the reconstruction error and the node comprehensive anomaly score to generate the node anomaly credibility;

[0018] When the node comprehensive anomaly score is higher than the anomaly score threshold, the corresponding node is marked as abnormal; when the node comprehensive anomaly score is higher than the anomaly score threshold, but the reconstruction error is lower than the reconstruction error threshold, it is determined that there is a false alarm, and the corresponding node is marked as normal, and the abnormal node set is screened out;

[0019] The optical network fault detection information generation, specifically, through the feature cross-modal fusion, the graph time series anomaly detection and the false alarm correction, obtaining the optical network fault detection information, the optical network fault detection information including the optical network topology graph, the node comprehensive anomaly score, the node anomaly credibility and the abnormal node set.

[0020] Further, the influence range evaluation, specifically, according to the optical network operation data and the optical network fault detection information, using the optical network fault multi-dimensional influence range evaluation method to evaluate the influence range, obtaining the fault influence range evaluation information, including the following steps: fault propagation path modeling, diffusion simulation, influence measurement and evaluation information generation;

[0021] The fault propagation path modeling, specifically, by setting the static edge weight in the optical network topology graph, and combining the node anomaly credibility and the link delay, calculating the fault propagation probability of each edge as the dynamic edge weight to construct the fault propagation path probability graph;

[0022] The diffusion simulation, specifically according to the fault propagation path probability graph, calculates a propagation cumulative abnormal probability for each node, and then screens out nodes with a propagation cumulative abnormal probability exceeding an adaptive threshold to obtain a set of nodes affected by the fault;

[0023] The influence metric, specifically introducing a node topology centrality coefficient and a link traffic sensitivity, combines a node abnormal credibility, a static edge weight and a total amount of affected traffic to perform multi-dimensional quantitative analysis on the nodes affected by the fault and calculate a fault comprehensive influence score;

[0024] The evaluation information generation, specifically by performing the fault propagation path modeling, the diffusion simulation and the influence metric, obtains fault influence range evaluation information including a fault propagation path probability graph, a set of nodes affected by the fault and a fault comprehensive influence score.

[0025] Further, the detection report generation, specifically combining optical network fault detection information and fault influence range evaluation information, automatically generates an optical network fault detection report to provide a reference for optical network operation and decision-making.

[0026] The above scheme has the following beneficial effects:

[0027] (1) In view of the technical problem that in the existing optical network fault detection process, the device state information is scattered and has various types, and a single data source monitoring cannot accurately capture the fault, which leads to false positives, the present scheme creatively uses a multi-modal adaptive fault detection method for intelligent fault detection, comprehensively considers link operation data, device operation logs and network structure information, dynamically identifies potential abnormal nodes, reduces the false positive rate and discovers possible faults affecting network operation in advance, and provides a quantifiable abnormal credibility score;

[0028] (2) In view of the technical problem that in the existing optical network fault detection process, the fault may quickly spread through the network to affect a large number of nodes and services, a single node abnormal information cannot accurately determine the overall influence range, and different nodes and links have different importance to services, and the traditional method cannot quantitatively analyze the actual influence degree of the fault, the present scheme creatively uses an optical network fault multi-dimensional influence range evaluation method to evaluate the influence range, dynamically predicts nodes and links that may be affected by the fault on the basis of considering network structure, node abnormal degree and traffic, quantifies the fault influence range and severity, and provides intuitive and operable decision-making basis for operation and maintenance personnel, thereby effectively reducing the fault diffusion risk and optimizing the network maintenance strategy. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 A module schematic diagram of the optical network fault detection system based on artificial intelligence provided by the present application is shown in the figure.

[0030] Figure 2 Flowchart for the intelligent fault detection module;

[0031] Figure 3 Flowchart for the impact range evaluation module.

[0032] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and are used to explain the application, but are not intended to limit the application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0034] In the description of the application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the application.

[0035] Embodiment one, refer to Figure 1 The application provides an optical network fault detection system based on artificial intelligence, which comprises a data acquisition module, an intelligent fault detection module, an impact range evaluation module and a detection report generation module.

[0036] The data acquisition module is used for data acquisition, and through data acquisition, optical network operation data is obtained, and the optical network operation data is sent to the intelligent fault detection module.

[0037] The intelligent fault detection module is used for intelligent fault detection, and according to the optical network operation data, a multi-modal adaptive fault detection method is used for intelligent fault detection, optical network fault detection information is obtained, and the optical network fault detection information is sent to the impact range evaluation module and the detection report generation module.

[0038] The impact range evaluation module is used for impact range evaluation, and according to the optical network operation data and the optical network fault detection information, an optical network fault multi-dimensional impact range evaluation method is used for impact range evaluation, fault impact range evaluation information is obtained, and the fault impact range evaluation information is sent to the detection report generation module.

[0039] The detection report generation module is used to generate detection reports, resulting in optical network fault detection reports.

[0040] Example 2, see Figure 1 This embodiment is based on the above embodiment. The data acquisition specifically involves acquiring optical network operation data and performing preprocessing. The optical network operation data includes link operation data, device operation logs, and optical network topology information.

[0041] The link operation data includes OSNR, BER, optical power, link latency, link service traffic, link bandwidth capacity, link bandwidth utilization, link physical length, and optical power loss.

[0042] The preprocessing includes data cleaning and timestamp alignment.

[0043] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The intelligent fault detection specifically involves using a multimodal adaptive fault detection method based on optical network operation data to perform intelligent fault detection and obtain optical network fault detection information. The steps include: feature cross-modal fusion, graph temporal anomaly detection, false alarm correction, and optical network fault detection information generation.

[0044] The cross-modal feature fusion is specifically based on the optical network topology information. Each physical device in the optical network is treated as a graph node, and each link is treated as an edge to construct an optical network topology graph. By calculating time-series signal features, log features, and topology features, and then splicing them together, cross-modal fusion features are obtained as node features of the optical network topology graph, and a node feature matrix is ​​constructed.

[0045] The time-series signal features are specifically obtained by normalizing the link operation data corresponding to each edge according to time steps, then performing one-dimensional convolution to extract the link time-series pattern, and then, for each node, performing average pooling on the link time-series patterns corresponding to all edges connecting the same node to obtain the time-series signal features.

[0046] The log features are specifically obtained by encoding the device operation logs corresponding to each node using BERT. When the device operation logs corresponding to a node are missing, the average value of the neighboring log features is used to fill the gap.

[0047] The topology features are specifically extracted from optical network topology information, and the topology features include node-level, neighbor-level, and global topology statistical features.

[0048] The formula for calculating the node features is as follows:

[0049] ;

[0050] In the formula, v is the first index of the node, and H v These are node features, specifically the cross-modal fusion features of node v. It is the time-series signal characteristic of node v. It is the log characteristic of node v. It represents the topological features of node v, and [·||·] represents the concatenation operation;

[0051] The formula for calculating the mean of the neighbor log features is as follows:

[0052] ;

[0053] In the formula, Let N(v) be the mean of the neighbor log features of node v, N(v) be the set of neighbor nodes of node v, |N(v)| be the number of neighbor nodes of node v, and u be the second index of the node, used to represent the neighbor node. These are the log characteristics of node u;

[0054] The graph temporal anomaly detection method specifically involves taking the optical network topology graph and node feature matrix as input, performing spatial dependency modeling through a graph attention network model to obtain a spatial dependency enhanced feature matrix, performing temporal modeling on the spatial dependency enhanced feature matrix through a temporal Transformer encoder to obtain a temporal prediction feature matrix, and generating a weighted comprehensive node anomaly score by calculating the prediction error and the neighborhood consistency error.

[0055] The formula for calculating the prediction error is as follows:

[0056] ;

[0057] In the formula, It represents the prediction error of node v at time step t, where t is the time step index, and ||·||2 is the L2 norm sign. It is a spatial dependency enhancement feature of node v at time step t. It is the temporal prediction feature of node v at time step t;

[0058] The neighborhood consistency error is used to evaluate the degree of difference between a node and its neighboring nodes within the topological neighborhood, and is calculated using the following formula:

[0059] ;

[0060] In the formula, It is the neighborhood consistency error of node v at time step t. It is a spatial dependency enhancement feature of node u at time step t;

[0061] The formula for calculating the node's overall anomaly score is as follows:

[0062] ;

[0063] In the formula, S v (t) is the node's overall anomaly score, specifically the overall anomaly score of node v at time step t. It is the prediction error weight. It is the domain consistency error weight;

[0064] Preferably, the graph attention network model includes two layers of GATv2. Each layer of GATv2 adopts a multi-head attention mechanism with four attention heads. The output of each attention head is concatenated and first activated by LeakyReLU, and then randomly deactivated by Dropout.

[0065] The false alarm correction specifically involves constructing a standard self-supervised mask reconstruction model, randomly masking the spatially dependent enhanced feature matrix and predicting reconstruction to obtain the reconstructed feature matrix, and then calculating the reconstruction error; combining the reconstruction error and the node comprehensive anomaly score, a node anomaly confidence level is generated.

[0066] When the overall anomaly score of a node is higher than the anomaly score threshold, the corresponding node is marked as an anomaly; when the overall anomaly score of a node is higher than the anomaly score threshold, but the reconstruction error is lower than the reconstruction error threshold, it is determined that there is a false alarm, the corresponding node is marked as normal, and the set of anomaly nodes is obtained by filtering.

[0067] The formula for calculating the reconstruction error is as follows:

[0068] ;

[0069] In the formula, It is the reconstruction error of node v at time step t. It is the reconstruction feature of node v at time step t;

[0070] The formula for calculating the confidence level of node anomalies is as follows:

[0071] ;

[0072] In the formula, C v (t) represents the anomalous confidence of node v at time step t. It is a control factor with a value range of [0,1].

[0073] Preferably, the anomaly score threshold is set based on the historical node comprehensive anomaly score distribution to distinguish node status, and the reconstruction error threshold is set based on the historical reconstruction error distribution to assist in identifying false alarms;

[0074] The generation of optical network fault detection information specifically involves obtaining optical network fault detection information through the feature cross-modal fusion, the graph temporal anomaly detection, and the false alarm correction. The optical network fault detection information includes an optical network topology graph, a node comprehensive anomaly score, a node anomaly confidence level, and a set of anomaly nodes.

[0075] By performing the above operations, this solution addresses the technical problem in existing optical network fault detection processes where device status information is scattered and diverse, and single data source monitoring is insufficient to accurately capture faults, leading to a high likelihood of false alarms. It creatively employs a multimodal adaptive fault detection method for intelligent fault detection, integrating link operation data, device operation logs, and network structure information to dynamically identify potential abnormal nodes, reduce false alarm rates, and proactively detect faults that may affect network operation, providing a quantifiable anomaly confidence score.

[0076] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The impact range assessment specifically involves using an optical network fault multidimensional impact range assessment method to assess the impact range based on optical network operation data and optical network fault detection information, and obtaining fault impact range assessment information. The steps include: fault propagation path modeling, diffusion simulation, impact measurement, and assessment information generation.

[0077] The fault propagation path modeling specifically involves setting static edge weights in the optical network topology graph and combining node anomaly confidence and link delay to calculate the fault propagation probability of each edge as a dynamic edge weight, thereby constructing a fault propagation path probability graph.

[0078] The formula for calculating the static edge weight is:

[0079] ;

[0080] In the formula, w uv It is the static edge weight between node u and node v. It is to adjust the first coefficient, W BW This refers to the link bandwidth utilization rate, specifically the link bandwidth utilization rate corresponding to the edge between node u and node v. It is the adjustment of the second coefficient, W loss It is optical power loss. It is the adjustment of the third coefficient, W length It is the physical length of the link;

[0081] The formula for calculating the fault propagation probability is:

[0082] ;

[0083] In the formula, It represents the probability of fault propagation, specifically the probability that a fault in node u will propagate to node v at time step t. Sigmoid(·) is the Sigmoid function. It is the anomaly credibility weight, C u (t) represents the anomalous confidence of node u at time step t. It is a static edge weight influence factor. It is the time decay factor. It refers to the link delay, specifically the link delay corresponding to the edge between node u and node v;

[0084] The diffusion simulation specifically involves calculating the cumulative anomaly probability of propagation for each node based on the fault propagation path probability map, and then selecting nodes whose cumulative anomaly probability of propagation exceeds an adaptive threshold to obtain a set of nodes affected by the fault.

[0085] The formula for calculating the cumulative anomaly probability during propagation is as follows:

[0086] ;

[0087] In the formula, It is the cumulative anomaly probability of node v propagating at time step t. The overall representation represents the effective probability that node u has an abnormal impact on node v at time point t;

[0088] The formula for calculating the set of nodes affected by the fault is:

[0089] ;

[0090] In the formula, R t V is the set of nodes affected by the fault at time step t, and V is the set of nodes in the fault propagation path probability graph. It is the adaptive threshold of node v;

[0091] Preferably, the adaptive threshold is set based on the historical propagation cumulative anomaly probability distribution of the node, and is used to filter out nodes affected by the fault;

[0092] The impact metric specifically involves introducing the node topology centrality coefficient and link service sensitivity, combined with node anomaly credibility, static edge weights, and total affected service traffic, to conduct a multi-dimensional quantitative analysis of nodes affected by faults and calculate a comprehensive fault impact score.

[0093] Preferably, the formula for calculating the node topological centrality coefficient is as follows:

[0094] ;

[0095] In the formula, It is the topological centrality coefficient of node v. It is the degree centrality weight, d v It represents the degree of node v, specifically the number of edges connecting node v. `max` indicates the maximum value, and `d` represents the degree of node v. u It is the degree of node u. It is the betweenness centrality weight, b v b is the betweenness centrality of node v, used to represent the importance of a node in the shortest path. u It is the betweenness centrality of node u;

[0096] Preferably, the formula for calculating the link service sensitivity is:

[0097] ;

[0098] In the formula, It is the service sensitivity of the link corresponding to the edge between node u and node v. It is bandwidth weight, B uv Here, is the link bandwidth capacity corresponding to the edge between node u and node v, i is the third index of the node, j is the fourth index of the node, E is the set of edges in the fault propagation path probability graph, and B is the edge set. ij It represents the link bandwidth capacity corresponding to the edge between node i and node j. It is the business traffic weight, F uv F is the link traffic corresponding to the edge between node u and node v. ij It represents the link traffic corresponding to the edge between node i and node j;

[0099] The formula for calculating the total affected service traffic is as follows:

[0100] ;

[0101] In the formula, L(R) t E(R) represents the total affected business traffic. t () is the set of edges of nodes affected by the fault;

[0102] The formula for calculating the comprehensive impact score of the fault is:

[0103] ;

[0104] In the formula, I t It is a comprehensive fault impact score at time step t, used to quantify the degree of fault impact. It is the node influence factor. It is the link impact factor. It is a business influencing factor;

[0105] The assessment information is generated by performing the fault propagation path modeling, the diffusion simulation, and the impact measurement to obtain fault impact range assessment information, which includes a fault propagation path probability map, a set of nodes affected by the fault, and a comprehensive fault impact score.

[0106] By performing the above operations, this solution addresses the technical challenges in existing optical network fault detection processes. These challenges include the potential for faults to spread rapidly across the network, impacting numerous nodes and services. Furthermore, the difficulty in accurately assessing the overall impact of a single node's anomaly information, and the significant differences in the importance of different nodes and links to services, making it challenging to quantify the actual impact of faults using traditional methods. This solution creatively employs a multi-dimensional impact assessment method for optical network faults. Considering network structure, node anomaly levels, and service traffic, it dynamically predicts the nodes and links that may be affected by the fault, quantifying the scope and severity of the fault's impact. This provides operations and maintenance personnel with an intuitive and actionable basis for decision-making, effectively reducing the risk of fault propagation and optimizing network maintenance strategies.

[0107] Example 5, see Figure 1 This embodiment is based on the above embodiment. Specifically, the generation of the detection report involves automatically generating an optical network fault detection report by combining optical network fault detection information and fault impact range assessment information, providing a reference for optical network operation and maintenance decisions.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0110] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based optical network fault detection system, characterized in that: It includes a data acquisition module, an intelligent fault detection module, an impact range assessment module, and a test report generation module; The data acquisition module is used for data acquisition, obtaining optical network operation data through data acquisition, and sending the optical network operation data to the intelligent fault detection module; The intelligent fault detection module is used for intelligent fault detection. Based on the optical network operation data, it uses a multimodal adaptive fault detection method to perform intelligent fault detection, obtain optical network fault detection information, and send the optical network fault detection information to the impact range assessment module and the detection report generation module. The intelligent fault detection includes the following steps: feature cross-modal fusion, graph temporal anomaly detection, false alarm correction, and optical network fault detection information generation; The cross-modal feature fusion is specifically based on the optical network topology information. Each physical device in the optical network is treated as a graph node, and each link is treated as an edge to construct an optical network topology graph. By calculating time-series signal features, log features, and topology features, and then splicing them together, cross-modal fusion features are obtained as node features of the optical network topology graph, and a node feature matrix is ​​constructed. The graph temporal anomaly detection method specifically involves taking the optical network topology graph and node feature matrix as input, performing spatial dependency modeling through a graph attention network model to obtain a spatial dependency enhanced feature matrix, performing temporal modeling on the spatial dependency enhanced feature matrix through a temporal Transformer encoder to obtain a temporal prediction feature matrix, and generating a weighted comprehensive node anomaly score by calculating the prediction error and the neighborhood consistency error. The false alarm correction specifically involves constructing a standard self-supervised mask reconstruction model, randomly masking the spatially dependent enhanced feature matrix and predicting reconstruction to obtain the reconstructed feature matrix, and then calculating the reconstruction error; combining the reconstruction error and the node comprehensive anomaly score, a node anomaly confidence level is generated. When the overall anomaly score of a node is higher than the anomaly score threshold, the corresponding node is marked as an anomaly; when the overall anomaly score of a node is higher than the anomaly score threshold, but the reconstruction error is lower than the reconstruction error threshold, it is determined that there is a false alarm, the corresponding node is marked as normal, and the set of anomaly nodes is obtained by filtering. The generation of optical network fault detection information specifically involves obtaining optical network fault detection information through the feature cross-modal fusion, the graph temporal anomaly detection, and the false alarm correction. The impact range assessment module is used to assess the impact range. It uses a multi-dimensional impact range assessment method for optical network faults based on optical network operation data and optical network fault detection information to assess the impact range, obtain fault impact range assessment information, and send the fault impact range assessment information to the detection report generation module. The impact range assessment includes the following steps: fault propagation path modeling, diffusion simulation, impact measurement, and assessment information generation; The fault propagation path modeling specifically involves setting static edge weights in the optical network topology graph and combining node anomaly confidence and link delay to calculate the fault propagation probability of each edge as a dynamic edge weight, thereby constructing a fault propagation path probability graph. The diffusion simulation specifically involves calculating the cumulative anomaly probability of propagation for each node based on the fault propagation path probability map, and then selecting nodes whose cumulative anomaly probability of propagation exceeds an adaptive threshold to obtain a set of nodes affected by the fault. The impact metric specifically involves introducing the node topology centrality coefficient and link service sensitivity, combined with node anomaly credibility, static edge weights, and total affected service traffic, to conduct a multi-dimensional quantitative analysis of nodes affected by faults and calculate a comprehensive fault impact score. The assessment information is generated specifically by performing the fault propagation path modeling, the diffusion simulation, and the impact measurement to obtain fault impact range assessment information; The detection report generation module is used to generate detection reports, resulting in optical network fault detection reports.

2. The artificial intelligence-based optical network fault detection system according to claim 1, characterized in that: The optical network fault detection information includes an optical network topology diagram, a comprehensive node anomaly score, a node anomaly confidence level, and a set of anomaly nodes.

3. The artificial intelligence-based optical network fault detection system according to claim 2, characterized in that: In the cross-modal feature fusion, the time-series signal features are specifically obtained by normalizing the link operation data corresponding to each edge according to time steps, then performing one-dimensional convolution to extract the link time-series pattern, and then, for each node, performing average pooling on the link time-series patterns corresponding to all edges connecting the same node to obtain the time-series signal features.

4. The artificial intelligence-based optical network fault detection system according to claim 3, characterized in that: The log features are specifically obtained by encoding the device operation logs corresponding to each node using BERT. When the device operation logs corresponding to a node are missing, the average value of the neighboring log features is used to fill the gap. The topology features are specifically extracted from the optical network topology information, and these topology features include node-level, neighbor-level, and global topology statistical features.

5. The artificial intelligence-based optical network fault detection system according to claim 4, characterized in that: The fault impact range assessment information includes a fault propagation path probability diagram, a set of nodes affected by the fault, and a comprehensive fault impact score.

6. The artificial intelligence-based optical network fault detection system according to claim 5, characterized in that: The data acquisition specifically involves collecting optical network operation data and preprocessing it. The optical network operation data includes link operation data, device operation logs, and optical network topology information. The link operation data includes OSNR, BER, optical power, link latency, link service traffic, link bandwidth capacity, link bandwidth utilization, link physical length, and optical power loss.

7. The artificial intelligence-based optical network fault detection system according to claim 6, characterized in that: The generation of the detection report specifically involves automatically generating an optical network fault detection report by combining optical network fault detection information and fault impact range assessment information, providing a reference for optical network operation and maintenance decisions.

Images