Network fault diagnosis method and system, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a network fault diagnosis method, a network fault diagnosis system, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining multi-dimensional data of network equipment, the multi-dimensional data comprises equipment configuration information, network topology information, performance index data and historical fault records; constructing a knowledge graph according to the multi-dimensional data, wherein the knowledge graph comprises nodes representing the network equipment, equipment links, network events and equipment configuration items, and edges representing a physical connection relationship, a logic dependency relationship, an equipment dependency relationship, network configuration information and a causal relationship among the nodes; performing map reasoning based on the knowledge map to obtain a map reasoning result; and generating diagnosis information for a target fault according to the map reasoning result. According to the embodiment of the invention, through multi-dimensional data integration, knowledge graph construction and automatic reasoning, the diagnosis efficiency, accuracy and adaptability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network security, in particular to a network fault diagnosis method, a network fault diagnosis system, an electronic device and a computer readable storage medium. BACKGROUND

[0002] With the development of information technology and the popularity of the Internet, the scale and complexity of the network are increasing, and the frequent occurrence of network faults seriously affects the continuity of business and user experience. Traditional broadband fault diagnosis mainly relies on manual experience and static rules, and when facing massive data, it is inefficient and easy to miss. SUMMARY

[0003] In view of the above problems, the present application is proposed to provide a network fault diagnosis method, a network fault diagnosis system, an electronic device and a computer readable storage medium which overcome the above problems or at least partially solve the above problems.

[0004] In order to solve the above problems, the present application discloses a network fault diagnosis method, which comprises:

[0005] Obtaining multi-dimensional data of network equipment, the multi-dimensional data comprising device configuration information, network topology information, performance index data and historical fault records;

[0006] Constructing a knowledge graph according to the multi-dimensional data, the knowledge graph comprising nodes representing the network equipment, device links, network events and device configuration items, and edges representing physical connection relationships, logical dependency relationships, device dependency relationships, network configuration information and causal relationships between the nodes;

[0007] Performing graph reasoning based on the knowledge graph to obtain a graph reasoning result;

[0008] Generating diagnosis information for a target fault according to the graph reasoning result.

[0009] Optionally, the graph reasoning based on the knowledge graph to obtain a graph reasoning result comprises:

[0010] Analyzing the device dependency relationships and the network configuration information by a graph algorithm to generate an abnormal pattern associated with the target fault;

[0011] Determining an affected range of the target fault according to the logical dependency relationships and the causal relationships;

[0012] Generating the graph reasoning result based on the abnormal pattern and the affected range, the graph reasoning result comprising a fault cause and an associated path.

[0013] Optionally, generating diagnostic information for the target fault based on the graph inference result includes:

[0014] The machine learning model is used to extract features and detect anomalies in the multidimensional data, generating anomaly detection results.

[0015] The anomaly detection results are fused with the graph inference results to generate the diagnostic information, which includes the fault type, fault location, and fault root cause.

[0016] Optionally, the step of using a machine learning model to extract features and detect anomalies in the multidimensional data includes:

[0017] Key features were extracted from the multidimensional data using the random forest algorithm. These key features included bandwidth utilization, network latency, packet loss rate, and optical power.

[0018] An anomaly detection model is trained based on the key features and the historical fault records;

[0019] The anomaly detection model is used to analyze the performance index data in real time and generate the anomaly detection results, which include the probability of failure and the fault type label.

[0020] Optionally, after generating diagnostic information for the target fault based on the atlas inference result, the method further includes:

[0021] Based on the diagnostic information, a repair strategy is generated for the target fault, and the repair strategy includes configuration adjustment, device restart, or traffic redirection;

[0022] The repair strategy is executed automatically, and the repair effect is verified using the performance metrics data.

[0023] The parameters of the knowledge graph and the machine learning model are updated based on the repair results.

[0024] Optionally, the automatic execution of the repair strategy includes:

[0025] Based on the network topology information, the physical connection relationships, and the device dependencies, a repair path is dynamically generated.

[0026] Perform predefined basic repair operations or composite repair operations according to the repair path. The basic repair operations include disk cleanup or service restart, and the composite repair operations include performing optical power calibration after switching to a backup port.

[0027] Optionally, the machine learning model includes a fault prediction model, and the method further includes:

[0028] Based on the historical fault records, the device dependencies, the logical dependencies, and the network configuration information, supervised training is performed to generate an initial fault prediction model.

[0029] The fault diagnosis process is optimized using reinforcement learning algorithms, and the parameters of the initial fault prediction model are adjusted based on the optimization results of the fault diagnosis process.

[0030] The parameters of the initial fault prediction model are periodically fine-tuned based on newly collected multidimensional data to obtain the trained fault prediction model.

[0031] Optionally, acquiring multidimensional data from the network device includes:

[0032] The network device is subjected to fine-grained testing through an atomic diagnostic module, which includes an optical power monitoring module, a broadband account status detection module, a routing and bridging mode verification module, a VLAN configuration check module, and a LAN port status monitoring module.

[0033] The multidimensional data is formed based on the detection results of the atomic diagnostic module and the historical fault records.

[0034] This invention also discloses a network fault diagnosis system, the system comprising:

[0035] The multidimensional data acquisition module is used to acquire multidimensional data of network devices, including device configuration information, network topology information, performance index data, and historical fault records.

[0036] The knowledge graph construction module is used to construct a knowledge graph based on the multidimensional data. The knowledge graph includes nodes representing network devices, device links, network events, and device configuration items, as well as edges representing physical connection relationships, logical dependencies, device dependencies, network configuration information, and causal relationships between the nodes.

[0037] The knowledge graph reasoning module is used to perform graph reasoning based on the knowledge graph to obtain graph reasoning results.

[0038] The diagnostic information generation module is used to generate diagnostic information for the target fault based on the inference results of the graph.

[0039] Optionally, the knowledge graph reasoning module includes:

[0040] An anomaly pattern generation module is used to analyze the device dependencies and network configuration information using graph algorithms to generate anomaly patterns associated with the target fault.

[0041] An affected scope determination module is used to determine the affected scope of the target fault based on the logical dependency and the causal relationship;

[0042] The graph inference result generation module generates the graph inference result based on the abnormal pattern and the affected range. The graph inference result includes the cause of the fault and the associated path.

[0043] Optionally, the diagnostic information generation module includes:

[0044] An anomaly detection result generation module is used to perform feature extraction and anomaly detection on the multidimensional data using a machine learning model, and generate anomaly detection results.

[0045] The result fusion module is used to fuse the anomaly detection results with the graph inference results to generate the diagnostic information, which includes the fault type, fault location, and fault root cause.

[0046] Optionally, the anomaly detection result generation module includes:

[0047] The key feature extraction module is used to extract key features from the multidimensional data using the random forest algorithm. The key features include bandwidth utilization, network latency, packet loss rate, and optical power value.

[0048] An anomaly detection model training module is used to train an anomaly detection model based on the key features and the historical fault records.

[0049] The performance index data analysis module is used to analyze the performance index data in real time using the anomaly detection model and generate the anomaly detection results, which include the probability of failure and the fault type label.

[0050] Optionally, the system further includes:

[0051] The repair strategy generation module is used to generate a repair strategy for the target fault based on the diagnostic information generated by the diagnostic information generation module according to the graph inference result. The repair strategy includes configuration adjustment, device restart or traffic redirection.

[0052] The repair strategy execution module is used to automatically execute the repair strategy and verify the repair effect through the performance indicator data.

[0053] The knowledge graph parameter update module is used to update the parameters of the knowledge graph and the machine learning model based on the repair effect.

[0054] Optionally, the repair strategy execution module includes:

[0055] The repair path generation module is used to dynamically generate repair paths based on the network topology information, the physical connection relationships, and the device dependency relationships.

[0056] The repair operation execution module is used to perform predefined basic repair operations or composite repair operations according to the repair path. The basic repair operations include disk cleanup or service restart, and the composite repair operations include performing optical power calibration after switching to a backup port.

[0057] Optionally, the machine learning model includes a fault prediction model, and the system further includes:

[0058] The initial prediction model generation module is used to perform supervised training based on the historical fault records, the device dependencies, the logical dependencies, and the network configuration information to generate an initial fault prediction model.

[0059] The initial model parameter adjustment module is used to optimize the fault diagnosis process using reinforcement learning algorithms and adjust the parameters of the initial fault prediction model based on the optimization results of the fault diagnosis process.

[0060] The initial model parameter fine-tuning module is used to periodically fine-tune the parameters of the initial fault prediction model based on newly collected multidimensional data to obtain the trained fault prediction model.

[0061] Optionally, the multidimensional data acquisition module includes:

[0062] The fine-grained detection module is used to perform fine-grained detection on the network device through the atomic diagnostic module. The atomic diagnostic module includes an optical power monitoring module, a broadband account status detection module, a routing and bridging mode verification module, a VLAN configuration check module, and a LAN port status monitoring module.

[0063] A multidimensional data forming module is used to form the multidimensional data based on the detection results of the atomic diagnostic module and the historical fault records.

[0064] This invention also discloses an electronic device, comprising: one or more processors; and one or more machine-readable media storing instructions thereon, which, when executed by the one or more processors, cause the electronic device to perform a network fault diagnosis method as described above.

[0065] This invention also discloses a computer-readable storage medium storing a computer program that causes a processor to execute a network fault diagnosis method as described above.

[0066] The embodiments of the present invention have the following advantages:

[0067] This invention provides a network fault diagnosis scheme that acquires multidimensional data of network devices, including device configuration information, network topology information, performance index data, and historical fault records; constructs a knowledge graph based on the multidimensional data, the knowledge graph containing nodes representing network devices, device links, network events, and device configuration items, as well as edges representing physical connection relationships, logical dependencies, device dependencies, network configuration information, and causal relationships between nodes; performs graph reasoning based on the knowledge graph to obtain graph reasoning results; and generates diagnostic information for the target fault based on the graph reasoning results.

[0068] Compared to threshold judgment based on single-dimensional data, the comprehensive collection of multi-dimensional data in this invention can fully capture the network's operational status, significantly improving the detection accuracy and root cause localization accuracy of complex faults (such as multi-point faults or configuration conflicts). Secondly, the knowledge graph, through the structured representation of nodes and edges, systematically analyzes the physical and logical relationships between devices, overcoming the deficiency of lacking relationship analysis, effectively identifying fault propagation paths and impact ranges, thereby shortening fault recovery time and improving user experience. Furthermore, graph reasoning based on the knowledge graph enables automated diagnosis, replacing rule bases that rely on manual configuration, adapting to frequent network configuration changes and emerging fault modes, and reducing maintenance costs.

[0069] In summary, the embodiments of the present invention significantly improve diagnostic efficiency, accuracy, and adaptability through multi-dimensional data integration, knowledge graph construction, and automated reasoning. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the steps of a network fault diagnosis method according to an embodiment of the present invention;

[0071] Figure 2 This is a flowchart illustrating a broadband fault diagnosis and handling method based on knowledge graphs and machine learning according to an embodiment of the present invention.

[0072] Figure 3 This is a diagnostic orchestration flowchart of a broadband fault diagnosis and processing method based on knowledge graphs and machine learning according to an embodiment of the present invention.

[0073] Figure 4 This is a structural block diagram of a network fault diagnosis system according to an embodiment of the present invention. Detailed Implementation

[0074] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0075] This invention proposes a network fault diagnosis scheme that acquires multi-dimensional data from network devices (including device configuration information, network topology information, performance index data, and historical fault records). It constructs a knowledge graph containing nodes such as network devices, device links, network events, and device configuration items, as well as edges representing physical connections, logical dependencies, device dependencies, network configuration information, and causal relationships. Based on graph algorithms, it generates graph inference results indicating fault causes and their impact range. It also integrates anomaly detection results from machine learning models (extracting key features such as bandwidth utilization and network latency through random forest algorithms) to generate diagnostic information including fault type, location, and root cause. Furthermore, it automatically executes repair operation strategies (such as configuration adjustments and traffic redirection) based on dynamically generated repair paths from the knowledge graph. The scheme optimizes diagnostic efficiency through a distributed architecture and real-time monitoring, and continuously optimizes the fault prediction model using reinforcement learning and periodic fine-tuning. Finally, it achieves efficient and automated fault diagnosis and self-healing through fine-grained detection in atomic diagnostic modules and dynamic diagnostic process scripts, significantly improving diagnostic accuracy, efficiency, and adaptability in complex network environments.

[0076] Reference Figure 1 The diagram illustrates a flowchart of a network fault diagnosis method according to an embodiment of the present invention. This network fault diagnosis method can be applied to systems such as network fault diagnosis systems or intelligent diagnosis systems (hereinafter referred to as "systems"). Specifically, the network fault diagnosis method may include the following steps:

[0077] Step 101: Obtain multidimensional data of the network device.

[0078] Multidimensional data includes, but is not limited to, device configuration information, network topology information, performance metrics, and historical fault records to comprehensively characterize the network's operational status. Device configuration information covers basic device parameters, such as Virtual Local Area Network (VLAN) settings, Broadband Remote Access Server (BRAS) authentication parameters, and port configurations, ensuring the capture of static characteristics of device operation. Network topology information describes the physical and logical connections between network devices, such as the link status between Optical Network Units (ONUs) and Optical Line Terminals (OLTs), as well as the connection layers between switches, used to analyze fault propagation paths. Performance metrics include real-time collected dynamic indicators such as bandwidth utilization, network latency, packet loss rate, and optical power values, reflecting the network's operational health through periodic sampling. Historical fault records store the occurrence time, type, and impact range of past faults for subsequent fault mode analysis.

[0079] Step 102: Construct a knowledge graph based on multidimensional data.

[0080] A knowledge graph comprises nodes and edges. Nodes represent network devices (such as ONUs, OLTs, and switches), device links (such as fiber optic links), network events (such as authentication failures and abnormal optical power), and device configuration items (such as VLAN parameters and port rates). Edges represent physical connections between nodes (such as the physical link between an ONU and an OLT), logical dependencies (such as the logical constraints between VLAN configuration and port status), device dependencies (such as the BRAS depending on the ONU's operating status), network configuration information (such as IP address allocation rules), and causal relationships (such as optical power degradation leading to disconnection). The construction process first extracts entity information from multidimensional data. For example, VLAN settings are parsed from device configuration information, physical connections between devices are identified from network topology information, abnormal events are extracted from performance indicator data, and fault patterns are mined from historical fault records. Next, semantic extraction techniques (such as Natural Language Processing, NLP) or rule matching are used to map entities to nodes in the knowledge graph, and edges are generated based on the association information in the multidimensional data. For example, the system analyzes the configuration parameters of the BRAS and ONU to generate device dependency edges; through causal analysis of historical fault records, it generates causal relationship edges between abnormal optical power and line drops. The knowledge graph is stored in a graph database (such as Neo4j), supporting efficient querying and expansion.

[0081] Step 103: Perform graph reasoning based on the knowledge graph to obtain the graph reasoning result.

[0082] Graph reasoning analyzes the relationships between nodes in a knowledge graph using graph algorithms, specifically including device dependencies, network configuration information, logical dependencies, and causal relationships. The reasoning process first uses graph traversal or shortest path algorithms to mine abnormal patterns in the knowledge graph, such as identifying the association path between ONU optical power degradation and BRAS authentication failure. Based on the characteristics of the target fault, the system extracts relevant nodes and edges from the knowledge graph to generate abnormal patterns describing the potential causes of the fault. Next, the system combines logical dependencies and causal relationships to analyze the scope of the fault's impact; for example, logical dependencies determine the set of devices affected by inconsistent VLAN configurations, and causal relationships trace cascading faults caused by fiber optic cable breaks. During the reasoning process, the system also utilizes pattern matching from historical fault records to enhance the accuracy of the reasoning, such as comparing the similarity between the current fault and historical optical power anomalies. Graph reasoning results include the fault cause, associated paths, and scope of impact. To improve reasoning efficiency, the system employs a distributed computing framework (such as Apache Spark) to process large-scale knowledge graph data in parallel and uses Redis caching technology to store intermediate calculation results, supporting high-concurrency queries. The reasoning results are output in a structured format, containing detailed descriptions of the causes of the faults and related path information, providing an accurate basis for the generation of subsequent diagnostic information.

[0083] Step 104: Generate diagnostic information for the target fault based on the graph reasoning results.

[0084] Diagnostic information includes fault type (e.g., fiber breakage, authentication failure, configuration error), fault location (e.g., specific ONU device or BRAS port), and root cause (e.g., optical power below -27dBm or inconsistent VLAN configuration). The generation process first analyzes the fault causes and associated paths in the graph inference results. For example, it extracts fiber breakage as the primary cause and the propagation path from ONU to OLT as clues to the fault location. The system verifies the reliability of the inference results by combining performance metrics data from multi-dimensional data, such as confirming the severity of the fiber fault by comparing optical power thresholds. Next, the system integrates anomaly detection results generated by a machine learning model with the graph inference results, using a weighted fusion algorithm to combine the information from both, ensuring the comprehensiveness and accuracy of the diagnostic information. For example, if the graph inference results point to fiber breakage, while the anomaly detection results show high packet loss rate and abnormal optical power, the system comprehensively determines the fault type as fiber breakage and locates it to the specific ONU port. After the diagnostic information is generated, the system records the fault details through structured output and supports visualization, facilitating quick understanding by maintenance personnel.

[0085] Compared to threshold judgment based on single-dimensional data, the comprehensive collection of multi-dimensional data in this invention can fully capture the network's operational status, significantly improving the detection accuracy and root cause localization accuracy of complex faults (such as multi-point faults or configuration conflicts). Secondly, the knowledge graph, through the structured representation of nodes and edges, systematically analyzes the physical and logical relationships between devices, overcoming the deficiency of lacking relationship analysis, effectively identifying fault propagation paths and impact ranges, thereby shortening fault recovery time and improving user experience. Furthermore, graph reasoning based on the knowledge graph enables automated diagnosis, replacing rule bases that rely on manual configuration, adapting to frequent network configuration changes and emerging fault modes, and reducing maintenance costs.

[0086] In an exemplary embodiment of the present invention, one implementation method for obtaining graph reasoning results based on knowledge graphs is as follows: analyze device dependencies and network configuration information through graph algorithms to generate anomaly patterns associated with the target fault; determine the affected scope of the target fault based on logical dependencies and causal relationships; and generate graph reasoning results based on the anomaly patterns and the affected scope, wherein the graph reasoning results include fault causes and associated paths.

[0087] First, the system uses graph algorithms to analyze device dependencies and network configuration information between nodes in the knowledge graph, extracting abnormal patterns of target faults, such as identifying the correlation between BRAS authentication failure and ONU optical power degradation. This process involves traversing the edges of the knowledge graph to uncover hidden fault-related relationships, generating structured abnormal patterns, and describing the potential trigger points of the fault. Second, based on the logical dependencies and causal relationships in the knowledge graph, the system analyzes the affected scope of the target fault. For example, logical dependencies can be used to determine the set of devices affected by configuration inconsistencies, or causal relationships can be used to trace the user scope affected by cascading faults caused by fiber optic breaks. The system combines historical fault records for pattern matching to enhance the accuracy of the affected scope. Finally, the system integrates the abnormal patterns and affected scope to generate graph inference results, including the fault cause and associated paths, outputting them in JSON format to support the generation of subsequent diagnostic information.

[0088] This implementation utilizes the structured semantics of knowledge graphs and employs multi-dimensional relational reasoning to accurately locate the root cause of a fault and its propagation path, overcoming the limitations of traditional methods in identifying multiple faults or configuration conflicts.

[0089] In an exemplary embodiment of the present invention, one implementation method for generating diagnostic information for a target fault based on the graph inference results is as follows: using a machine learning model to extract features and detect anomalies in multidimensional data to generate anomaly detection results; fusing the anomaly detection results with the graph inference results to generate diagnostic information, which includes fault type, fault location, and fault root cause.

[0090] First, the system utilizes a machine learning model to extract features and detect anomalies in multidimensional data, generating anomaly detection results. The machine learning model employs a random forest algorithm to extract key features (such as network latency, packet loss rate, and optical power values), and trains the anomaly detection model based on historical fault data. It analyzes performance metrics in real time to generate anomaly detection results, including fault occurrence probability and fault type labels. Second, the system fuses the anomaly detection results with graph inference results to generate diagnostic information. Graph inference results include fault causes and associated paths. The fusion process uses a weighted fusion algorithm, combining the probability values ​​of the anomaly detection results with the semantic information of the graph inference results to generate diagnostic information, clearly identifying the fault type, fault location, and root cause. The system verifies performance metrics data to ensure the reliability of the fusion results and outputs diagnostic information in JSON format, supporting visualization. To improve efficiency, the system uses a distributed database to store intermediate fusion data and records the fusion process through a log system to ensure traceability.

[0091] This implementation utilizes comprehensive analysis of multidimensional data and semantic reasoning of knowledge graphs to accurately locate the root cause and scope of impact of complex faults. The integration of machine learning and graph reasoning overcomes the limitations of traditional methods in adapting to dynamic network environments, reduces the false positive rate and maintenance costs, and solves the problems of inaccurate root cause location and low efficiency in the background technology.

[0092] In one exemplary embodiment of the present invention, one implementation method for feature extraction and anomaly detection of multidimensional data using a machine learning model is as follows: key features are extracted from multidimensional data using a random forest algorithm, including bandwidth utilization, network latency, packet loss rate, and optical power value; an anomaly detection model is trained based on the key features and historical fault records; the anomaly detection model is used to analyze performance index data in real time to generate anomaly detection results, including fault occurrence probability and fault type label.

[0093] First, the random forest algorithm constructs multiple decision trees, evaluates feature importance, and filters high-dimensional features strongly correlated with faults from massive multidimensional data, generating a structured key feature set. The system accelerates the feature extraction process through a parallel computing framework, ensuring efficiency in processing large-scale data. Second, the system trains an anomaly detection model based on key features and historical fault records. Historical fault records include the type, time, and related performance indicators of past faults. The system uses supervised learning methods, taking key features as input and fault type as label, to construct the anomaly detection model. The training process optimizes model parameters through cross-validation, reducing the risk of overfitting and ensuring the model's generalization ability. Finally, the system uses the trained anomaly detection model to analyze performance indicator data in real time and generate anomaly detection results. The anomaly detection results are output in JSON format and stored in a distributed database, supporting subsequent fusion with graph inference results.

[0094] This implementation utilizes key features of multidimensional data and machine learning models to accurately identify complex fault modes and reduces the risk of misjudgment through probability quantification. The combination of real-time analysis and distributed computing overcomes the limitations of low efficiency in traditional methods and solves the problem of inaccurate root cause localization in the background technology.

[0095] In one exemplary embodiment of the present invention, after generating diagnostic information for the target fault based on the graph reasoning results, one implementation method is as follows: generating a repair strategy for the target fault based on the diagnostic information, the repair strategy including configuration adjustment, device restart, or traffic redirection; automatically executing the repair strategy and verifying the repair effect through performance index data; and updating the parameters of the knowledge graph and machine learning model based on the repair effect.

[0096] First, the repair strategy includes configuration adjustments (such as correcting VLAN parameters of the broadband access server) and device restarts (such as restarting optical network units) or traffic redirection (such as switching to redundant links). Based on network topology information and device dependencies in the knowledge graph, the system dynamically generates repair paths, prioritizing backup nodes or links to ensure repair efficiency. Second, the system automatically executes the repair strategy, implementing operations through predefined scripts or dynamically orchestrated execution flows, such as sending configuration commands via Simple Network Management Protocol (SNMP) or triggering device restarts via remote interfaces. During execution, the system monitors performance metrics in real time to verify the repair effect, such as checking whether optical power has recovered to above -27dBm or whether packet loss rate has decreased to the normal range. Finally, the system updates the parameters of the knowledge graph and machine learning model based on the repair effect. If the repair is successful, the system updates the device dependencies, logical dependencies, and network configuration information in the knowledge graph; if the repair fails, the system records the failure mode and supplements historical failure records.

[0097] This implementation utilizes the precise location of diagnostic information to generate dynamic repair paths, automatically executes configuration adjustments or traffic redirection, and shortens recovery time; the closed-loop update mechanism optimizes the knowledge graph and model through real-time feedback, overcoming the limitations of traditional methods in adapting to network changes and solving the problems of low efficiency and high maintenance costs in the background technology.

[0098] In an exemplary embodiment of the present invention, one implementation of automatically executing the repair strategy is as follows: dynamically generating a repair path based on network topology information, physical connection relationships, and device dependencies; and performing predefined basic repair operations or composite repair operations according to the repair path. The basic repair operations include disk cleanup or service restart, and the composite repair operations include performing optical power calibration after switching to a backup port.

[0099] First, the system uses graph algorithms to analyze the knowledge graph and generate repair paths, prioritizing redundant links or backup nodes to ensure efficient repair and minimize service interruptions. Path generation considers real-time performance metrics to ensure path availability. Second, the system executes predefined basic repair operations or dynamically orchestrated composite repair operations according to the repair paths. Basic repair operations include disk cleanup (e.g., clearing BRAS cache files) or service restart (e.g., restarting ONU service processes), implemented by sending commands via Simple Network Management Protocol (SMMP). Composite repair operations involve multi-step coordination; for example, performing optical power calibration after switching to a backup port. Specifically, this involves switching the ONU's backup port via SNMP and then calibrating the optical signal strength to above -27dBm using an optical power monitoring module. The execution process is automated through scripts, and the system monitors performance metrics in real time to verify the operation's effectiveness and stores execution logs in a distributed database, supporting fault traceability.

[0100] This implementation utilizes network topology information and dependencies from a knowledge graph to generate the optimal repair path, automates precise operations, and shortens service recovery time. The layered repair strategy (basic and composite) adapts to diverse fault scenarios, overcomes the limitations of traditional static rules, and solves the problems of low efficiency and high maintenance costs in the background technology.

[0101] In an exemplary embodiment of the present invention, the machine learning model includes a fault prediction model, which is generated by supervised training based on historical fault records and equipment dependencies, logical dependencies, and network configuration information; the fault diagnosis process is optimized using a reinforcement learning algorithm, and the parameters of the initial fault prediction model are adjusted according to the optimization results of the fault diagnosis process; the parameters of the initial fault prediction model are fine-tuned periodically based on newly collected multidimensional data to obtain a trained fault prediction model.

[0102] First, the system employs a classifier (such as a random forest or gradient boosting tree) with historical fault records, equipment dependencies, logical dependencies, and network configuration information as input, and fault type as the label. A supervised learning algorithm is used to train the model, generating an initial fault prediction model. The training process optimizes parameters through cross-validation to reduce the risk of overfitting. Second, the system utilizes a reinforcement learning framework to optimize the fault diagnosis process and adjust the parameters of the initial fault prediction model. Reinforcement learning simulates diagnostic scenarios, using diagnostic efficiency and accuracy as reward functions to explore the optimal diagnostic path. Based on the optimization results (e.g., a 50% reduction in diagnostic time), the model parameters are updated, and the model weights are fine-tuned using a policy gradient method. Finally, the system periodically fine-tunes the parameters of the initial fault prediction model based on newly collected multidimensional data to obtain a fully trained fault prediction model.

[0103] This implementation uses the semantic relationships of historical fault records and knowledge graphs to train the model and accurately predict complex faults; reinforcement learning and incremental fine-tuning overcome the limitations of high maintenance costs in traditional methods and solve the problem of poor adaptability to emerging fault modes in the background technology.

[0104] In an exemplary embodiment of the present invention, one way to obtain multidimensional data of a network device is as follows: fine-grained detection is performed on the network device through an atomic diagnostic module, which includes an optical power monitoring module, a broadband account status detection module, a routing and bridging mode verification module, a VLAN configuration check module, and a LAN port status monitoring module; multidimensional data is formed based on the detection results of the atomic diagnostic module and historical fault records.

[0105] First, the optical power monitoring module detects optical signal strength in real time; the broadband account status detection module verifies user authentication status; the routing and bridging mode verification module checks the device's operating mode; the VLAN configuration check module verifies configuration consistency; and the LAN port status monitoring module monitors the connection status of local area network ports. These modules periodically collect data through Simple Network Management Protocol (SMMP) or dedicated interfaces and output the detection results in a structured format, storing them in a distributed database. Second, the system integrates the detection results from the atomic diagnostic module and historical fault records to form multidimensional data. The detection results include device configuration information, network topology information, and performance indicator data; historical fault records provide fault types and related indicators. The system integrates these two data sources through data cleaning and format conversion to ensure the integrity and consistency of the multidimensional data.

[0106] This implementation utilizes multi-module collaborative detection to capture the dynamic characteristics of device configuration, network topology, and performance indicators, and combines historical fault records to enhance fault mode analysis capabilities, accurately supporting the diagnosis of complex faults.

[0107] Based on the above description of the network fault diagnosis method embodiments, a broadband fault diagnosis and handling method (hereinafter referred to as the diagnosis and handling method) is introduced below. This method utilizes intelligent and automated technologies to address the increasingly complex fault types and higher service availability requirements in modern home broadband networks. The core of this method lies in the integration of knowledge graph and machine learning technologies. It can not only accurately identify and locate faults but also automatically formulate and implement repair plans, thereby significantly reducing fault handling time, enhancing network stability and self-management capabilities, and ultimately achieving efficient network optimization and maintenance.

[0108] The implementation objectives of this diagnostic and treatment method include the following aspects:

[0109] Automated Fault Diagnosis: Leveraging the intelligent detection capabilities and dynamic diagnostic process orchestration of the atomic diagnostic module, rapid location and recovery from home broadband faults are achieved. Based on knowledge graphs and machine learning models, the method automatically analyzes the root cause of faults and adaptively generates diagnostic processes according to the fault type. This covers real-time monitoring and analysis of network devices, line status, and user terminals, completely eliminating the need for manual intervention. The atomic diagnostic module performs detailed testing on each network component, combining the global topology view of the knowledge graph to accurately identify problem nodes, reduce human error, improve operational efficiency, and ensure the continuous and stable operation of home broadband.

[0110] Precise Fault Location: Utilizing a knowledge graph to comprehensively map network topology and device relationships, precise fault location and rapid definition of the impact range are achieved. The atomic diagnostic module analyzes device health status in real time, dynamically adjusts the diagnostic process, and quickly identifies problem nodes and their upstream and downstream impacts. The intelligent diagnostic process significantly improves the speed and accuracy of fault location, effectively reduces the scope and duration of service interruptions, and enhances the overall efficiency and stability of network operations and maintenance.

[0111] Intelligent fault self-healing: Based on diagnostic results, repair strategies are automatically generated and executed in real time to quickly restore network services and shorten downtime. Through adaptive learning and feedback mechanisms, the repair strategies are continuously optimized, enhancing the system's adaptability and robustness to complex environments.

[0112] Continuous system optimization: Through real-time feedback and historical data analysis, the fault diagnosis and repair model is continuously improved to enhance processing efficiency and accuracy. Reinforcement learning mechanisms support the system's self-evolution during long-term operation, gradually improving fault prediction accuracy and repair efficiency, ensuring the reliability and high availability of home broadband services.

[0113] The innovative highlights of this method are reflected in the following aspects:

[0114] Knowledge graph-driven network status visualization: Utilizing knowledge graph technology, it provides comprehensive visual management of network status, encompassing multi-dimensional information such as physical topology, network traffic, configuration dependencies, and historical fault records, forming a dynamically updated network status view. By intuitively displaying device connections, data flow, and configuration changes, operations and maintenance personnel can quickly locate critical nodes and potential risks, enabling preventative maintenance or rapid fault response. Compared to traditional monitoring tools, the three-dimensional presentation of knowledge graphs significantly improves the speed and accuracy of fault location in complex networks.

[0115] Machine learning-assisted root cause analysis: By leveraging machine learning techniques, key indicators are automatically extracted from massive amounts of network data to build anomaly detection models. These models flexibly adapt to different fault types, improving detection accuracy and efficiency. Dynamic root cause analysis combined with historical data reduces the risk of misjudgment. A continuous learning mechanism optimizes the model, and time series forecasting enables early warning of potential faults, enhancing the intelligence and efficiency of operation and maintenance management.

[0116] Real-time self-healing and feedback optimization: A real-time self-healing mechanism is introduced to perform repair operations immediately upon detecting anomalies, record the effects, and adjust strategies through machine learning. Compared to a passive response model, real-time self-healing ensures rapid recovery, while feedback optimization enables continuous strategy improvement, reduces human intervention, minimizes the scope and duration of fault impact, and enhances network reliability.

[0117] Highly scalable and fault-tolerant design: Employing a distributed architecture, the system supports on-demand scaling of processing capacity, leverages Redis caching technology to ensure high-concurrency response, and its fault-tolerant design supports automatic switching to backup paths in case of node failure. Compared to traditional management methods, the distributed architecture guarantees high performance, and the fault-tolerant design enhances robustness, adapting to current and future complex network requirements.

[0118] Reference Figure 2 The diagram illustrates a flowchart of a broadband fault diagnosis and handling method based on knowledge graphs and machine learning, according to an embodiment of the present invention. The execution process of this broadband fault diagnosis and handling method based on knowledge graphs and machine learning can be as follows:

[0119] When a user reports a problem with their broadband connection, the system first receives the user's fault report request and prompts the user to enter their corresponding broadband account information as the entry point for fault location.

[0120] The system performs initial troubleshooting to determine if automatic troubleshooting can be achieved using existing rules and knowledge. The decision-making process is as follows: If the fault can be resolved automatically, the fault elimination and root cause are identified, and the fault recovery and verification process is executed, ending the process. If the fault cannot be directly resolved, the system proceeds to the fault diagnosis process.

[0121] The system starts the fault diagnosis module and loads relevant rule content from the fault diagnosis rule base, including but not limited to: diagnosis process, diagnosis items, diagnosis instructions, diagnosis indicators, diagnosis conclusions, and handling suggestions.

[0122] The system uses atomic diagnostic capabilities to collect diagnostic-related data from various network status monitoring systems as diagnostic inputs, including but not limited to: ONU receiving OLT optical power, broadband account status, ITMS network status, AAA online status, broadband LAN port status, broadband VLAN information comparison, and PON port status.

[0123] Further reasoning and identification are performed by calling the fault database (a knowledge graph implemented using Neo4j). Specifically, this includes: using the network state information and the association between nodes in the knowledge graph to locate faults; combining the graph structure and historical cases to identify faults in the graph graph (graph structure pattern recognition); and further performing machine learning reasoning on the corresponding identification results.

[0124] By combining fault maps with existing historical fault data, the following algorithms are used to train and infer the model to improve identification accuracy: K-means clustering, random forest, and reinforcement learning. The trained model assists in identifying fault characteristic patterns, paths, and suggested handling measures.

[0125] Based on the diagnostic results, the system automatically generates optimized fault repair strategies and marks the risk level and provides necessary manual intervention prompts.

[0126] The system performs fault repair operations according to the automatically generated strategy, and verifies that the fault has been resolved through a verification process. If user interaction is involved, relevant guidance will be provided.

[0127] Once the fault is successfully recovered and verified, the entire fault diagnosis and handling process is complete.

[0128] Reference Figure 3 This diagram illustrates a diagnostic orchestration flowchart of a broadband fault diagnosis and handling method based on knowledge graphs and machine learning, according to an embodiment of the present invention. The entire scenario orchestration process involves a series of diagnostics, decisions, and capability invocations. Each stage is specifically supported by the underlying capability orchestration.

[0129] Before making core decisions, the system will perform preliminary diagnostics in sequence.

[0130] Diagnostic item A is initiated, which depends on the results of diagnostic indicator 1 and diagnostic indicator 2.

[0131] Capability orchestration (capability 1, capability 2, capability 3) is supported sequentially through interface A, interface B, and interface C. These interfaces support diagnostic indicator 1 and diagnostic indicator 2.

[0132] Next, diagnostic item B is performed. Diagnostic item B depends on the results of diagnostic indicator 2 and diagnostic indicator 3.

[0133] Then there is diagnostic item C, which depends on the results of diagnostic indicators 3 and 4.

[0134] The next step is diagnostic item D, which depends on the results of diagnostic indicators 4 and 5.

[0135] Capability orchestration (capability 4, capability 5, capability 6) is supported by interfaces D, E, and F, which support diagnostic indicators 3, 4, and 5.

[0136] After the initial diagnosis is completed, the process enters stage E, which calls diagnostic item E. The execution of diagnostic item E depends on the results of diagnostic indicator 6.

[0137] Capability orchestration (capabilities 7, 8, and 9) is supported through interfaces G, H, and I, which support diagnostic metric 6. Diagnostic metric 6 and diagnostic metric 7 receive the results of capabilities 6 and 7 through microcomponent D.

[0138] Based on the decision outcome of stage E, the process will be diverted to either F or G.

[0139] If the process enters F, it will directly call the interface N provided by orchestration capability 11.

[0140] If the process enters G, diagnostic item F will be initiated, which depends on the results of diagnostic indicator 7.

[0141] Simultaneously, microcomponent F will also be invoked, which depends on the results of diagnostic indicators 9 and 10. Diagnostic item G depends on the result of diagnostic indicator 8.

[0142] Capability orchestration (capability 10, capability 11, capability 12) is supported by interfaces J, K, and L, which support diagnostic metric 8.

[0143] Following decision-making and subsequent diagnosis, the process enters stage H.

[0144] At this stage, the system will directly call the interfaces M, N, and O provided by the capability orchestration (capability 10, capability 11, capability 12) to complete the final operation or output (not shown in the figure).

[0145] The entire scene arrangement process is complete.

[0146] This invention also proposes an intelligent fault diagnosis method for home broadband, achieving intelligent and automated fault handling through the synergistic effect of knowledge graphs and machine learning technologies. The core of this method lies in structuring multi-dimensional network data into a knowledge graph, encompassing key information such as the model, brand, and firmware version of routers, switches, and terminal devices. It integrates static data (such as IP addresses, VLAN settings, port configurations, MAC addresses, and Quality of Service (QoS) policies) with dynamic data (such as bandwidth utilization, network latency, packet loss rate, jitter, TCP connection count, and error packet rate) to comprehensively reflect network performance status. Simultaneously, historical fault records are systematically stored, including fault occurrence time, impact range, and repair measures, providing a data foundation for subsequent analysis. By intelligently processing this multi-dimensional data using the random forest algorithm, this method achieves real-time and accurate fault diagnosis, quickly locates root causes, and provides scientific support for network performance optimization.

[0147] This method relies on dynamically updated network topology to ensure that the knowledge graph reflects the status and interconnection of each network element in real time. When a fault occurs, the system quickly extracts relevant device information, performs in-depth analysis in conjunction with performance indicators, and leverages the anomaly detection capabilities of machine learning models to improve fault location and repair efficiency. Furthermore, the method features adaptive optimization capabilities, adjusting diagnostic strategies through real-time feedback to adapt to complex and dynamic network environments, thereby shortening fault recovery time and enhancing the stability and reliability of home broadband.

[0148] The core modules of this method consist of:

[0149] 1. Knowledge Graph Construction Module

[0150] Data collection and preprocessing: Collect heterogeneous data from multiple sources, including network devices, configuration files, logs, and traffic monitoring systems. Then, through cleaning, standardization, and feature extraction, transform the raw data into structured information.

[0151] Knowledge graph generation and maintenance: A knowledge graph is constructed using the graph database Neo4j. Network devices, links, configuration parameters, and historical fault records are used as nodes. These nodes are connected through semantic relationships to form a dynamic and scalable semantic network, which is updated in real time to reflect the latest network status.

[0152] Relational Reasoning and Graph Enhancement: Automatically uncover hidden relationships through reasoning rules and graph algorithms to enrich the content of knowledge graphs.

[0153] 2. Machine Learning Model Training and Application Module

[0154] Data annotation and model training: Based on historical fault data and knowledge graph node relationships, supervised training is used to build anomaly detection, fault classification and fault prediction models. Random forest algorithm is selected to identify fault characteristics, classify types and predict potential faults.

[0155] Model Deployment and Inference Engine: The trained model is deployed to the inference engine, which combines knowledge graphs to process network operation data in real time, enabling automatic fault detection and localization, including predicting fault probability, analyzing causes, and generating optimization and repair solutions.

[0156] Adaptive model updates: The model is fine-tuned periodically based on newly acquired data to ensure it adapts to changes in the network environment and maintains diagnostic accuracy and robustness.

[0157] 3. Intelligent Fault Diagnosis and Handling Module

[0158] Fault detection: By monitoring performance metrics such as traffic, latency, and packet loss rate in real time, machine learning models identify abnormal states, trigger alarms, and provide information on fault type, affected devices, location, and severity.

[0159] Intelligent fault location: Based on the topology and device relationships of the knowledge graph, it accurately locates fault nodes and links, analyzes dependencies to narrow down the scope of impact, and helps to determine the root cause.

[0160] Repair suggestion generation: Based on knowledge graph reasoning and machine learning output, repair suggestions are automatically generated, including configuration adjustments, device restarts, and traffic rerouting. For specific faults, predefined policies (such as triggering disk cleanup exceeding the 90% threshold, or triggering service restart after 3 consecutive timeouts) and composite policies (such as optical power calibration after switching to a backup port) are supported, and repair paths are adjusted in conjunction with real-time topology (such as prioritizing redundant links).

[0161] Workflow orchestration: Through atomic diagnostic capabilities and dynamic orchestration, we ensure efficient collaboration across all stages from detection to repair.

[0162] 4. Adaptive Optimization and Evolutionary Module

[0163] Feedback learning mechanism: After fault repair, collect user feedback and performance change data, and use reinforcement learning to optimize the diagnostic model and repair strategy to improve the level of intelligent diagnosis.

[0164] Dynamic strategy adjustment: Based on real-time analysis and historical data, optimize resource allocation and process configuration, such as prioritizing the handling of critical node failures during traffic surges to ensure service continuity.

[0165] Specific Implementation Example: Scenarios Where Broadband Internet Access is Unavailable

[0166] Scenario: A user complains to customer service that their home broadband cannot access the internet, and the maintenance personnel enter the number to initiate a diagnostic.

[0167] Step 1: Definition of Atomic Diagnostic Capability

[0168] The interface encapsulates ONU optical power, broadband account status, routing bridging mode, cloud broadband status, AAA online status, ITMS gateway status, LAN port status, network cable port error, downlink optical attenuation, speed audit, VLAN information comparison, PON port status, etc., and standardizes input and output parameters to form basic capabilities.

[0169] Step 2: Diagnostic Process Arrangement

[0170] By using a visual interface with atomic capabilities as nodes, a diagnostic process for internet connectivity issues can be configured and script files generated.

[0171] Step 3: Fault Diagnosis

[0172] The script is triggered to detect a decrease in optical power or authentication failure, and the problem node is located by combining the knowledge graph.

[0173] Step 4: Troubleshooting Guide

[0174] Based on the results, a repair script (such as checking fiber optic cables or adjusting VLANs) is generated, automatically executed, and the effect is verified, providing operation guidance.

[0175] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0176] Reference Figure 4 The diagram illustrates a structural block diagram of a network fault diagnosis system according to an embodiment of the present invention. Specifically, the network fault diagnosis system may include the following modules.

[0177] The multidimensional data acquisition module 41 is used to acquire multidimensional data of network devices, including device configuration information, network topology information, performance index data and historical fault records.

[0178] The knowledge graph construction module 42 is used to construct a knowledge graph based on the multidimensional data. The knowledge graph includes nodes representing network devices, device links, network events, and device configuration items, as well as edges representing physical connection relationships, logical dependencies, device dependencies, network configuration information, and causal relationships between the nodes.

[0179] The knowledge graph reasoning module 43 is used to perform graph reasoning based on the knowledge graph to obtain graph reasoning results.

[0180] The diagnostic information generation module 44 is used to generate diagnostic information for the target fault based on the graph reasoning results.

[0181] In an exemplary embodiment of the present invention, the knowledge graph reasoning module 43 includes:

[0182] An anomaly pattern generation module is used to analyze the device dependencies and network configuration information using graph algorithms to generate anomaly patterns associated with the target fault.

[0183] An affected scope determination module is used to determine the affected scope of the target fault based on the logical dependency and the causal relationship;

[0184] The graph inference result generation module generates the graph inference result based on the abnormal pattern and the affected range. The graph inference result includes the cause of the fault and the associated path.

[0185] In an exemplary embodiment of the present invention, the diagnostic information generation module 44 includes:

[0186] An anomaly detection result generation module is used to perform feature extraction and anomaly detection on the multidimensional data using a machine learning model, and generate anomaly detection results.

[0187] The result fusion module is used to fuse the anomaly detection results with the graph inference results to generate the diagnostic information, which includes the fault type, fault location, and fault root cause.

[0188] In an exemplary embodiment of the present invention, the anomaly detection result generation module includes:

[0189] The key feature extraction module is used to extract key features from the multidimensional data using the random forest algorithm. The key features include bandwidth utilization, network latency, packet loss rate, and optical power value.

[0190] An anomaly detection model training module is used to train an anomaly detection model based on the key features and the historical fault records.

[0191] The performance index data analysis module is used to analyze the performance index data in real time using the anomaly detection model and generate the anomaly detection results, which include the probability of failure and the fault type label.

[0192] In one exemplary embodiment of the present invention, the system further includes:

[0193] The repair strategy generation module is used to generate a repair strategy for the target fault based on the diagnostic information generated by the diagnostic information generation module 44 according to the graph inference result. The repair strategy includes configuration adjustment, device restart or traffic redirection.

[0194] The repair strategy execution module is used to automatically execute the repair strategy and verify the repair effect through the performance indicator data.

[0195] The knowledge graph parameter update module is used to update the parameters of the knowledge graph and the machine learning model based on the repair effect.

[0196] In an exemplary embodiment of the present invention, the repair strategy execution module includes:

[0197] The repair path generation module is used to dynamically generate repair paths based on the network topology information, the physical connection relationships, and the device dependency relationships.

[0198] The repair operation execution module is used to perform predefined basic repair operations or composite repair operations according to the repair path. The basic repair operations include disk cleanup or service restart, and the composite repair operations include performing optical power calibration after switching to a backup port.

[0199] In one exemplary embodiment of the present invention, the machine learning model includes a fault prediction model, and the system further includes:

[0200] The initial prediction model generation module is used to perform supervised training based on the historical fault records, the device dependencies, the logical dependencies, and the network configuration information to generate an initial fault prediction model.

[0201] The initial model parameter adjustment module is used to optimize the fault diagnosis process using reinforcement learning algorithms and adjust the parameters of the initial fault prediction model based on the optimization results of the fault diagnosis process.

[0202] The initial model parameter fine-tuning module is used to periodically fine-tune the parameters of the initial fault prediction model based on newly collected multidimensional data to obtain the trained fault prediction model.

[0203] In an exemplary embodiment of the present invention, the multidimensional data acquisition module 41 includes:

[0204] The fine-grained detection module is used to perform fine-grained detection on the network device through the atomic diagnostic module. The atomic diagnostic module includes an optical power monitoring module, a broadband account status detection module, a routing and bridging mode verification module, a VLAN configuration check module, and a LAN port status monitoring module.

[0205] A multidimensional data forming module is used to form the multidimensional data based on the detection results of the atomic diagnostic module and the historical fault records.

[0206] As the system implementation is basically similar to the method implementation, it is described in a relatively simple way. For relevant details, please refer to the description of the method implementation.

[0207] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0208] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0209] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0210] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0211] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0212] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0213] Finally, 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 terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0214] The present invention has provided a detailed description of a network fault diagnosis method and a network fault diagnosis system. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A network fault diagnosis method, characterized in that, The method includes: Acquire multidimensional data of network devices, including device configuration information, network topology information, performance index data, and historical fault records; A knowledge graph is constructed based on the multidimensional data. The knowledge graph includes nodes representing network devices, device links, network events, and device configuration items, as well as edges representing physical connection relationships, logical dependencies, device dependencies, network configuration information, and causal relationships between the nodes. Graph reasoning is performed based on the knowledge graph to obtain graph reasoning results; Diagnostic information for the target fault is generated based on the inference results of the graph.

2. The method according to claim 1, characterized in that, The graph reasoning based on the knowledge graph to obtain the graph reasoning result includes: The device dependencies and network configuration information are analyzed using graph algorithms to generate anomaly patterns associated with the target fault. Based on the logical dependencies and causal relationships, determine the scope of the target fault. Based on the abnormal pattern and the affected range, the graph inference result is generated, which includes the cause of the failure and the associated path.

3. The method according to claim 1 or 2, characterized in that, The step of generating diagnostic information for the target fault based on the inference results of the graph includes: The machine learning model is used to extract features and detect anomalies in the multidimensional data, generating anomaly detection results. The anomaly detection results are fused with the graph inference results to generate the diagnostic information, which includes the fault type, fault location, and fault root cause.

4. The method according to claim 3, characterized in that, The process of using a machine learning model to extract features and detect anomalies in the multidimensional data includes: Key features were extracted from the multidimensional data using the random forest algorithm. These key features included bandwidth utilization, network latency, packet loss rate, and optical power. An anomaly detection model is trained based on the key features and the historical fault records; The anomaly detection model is used to analyze the performance index data in real time and generate the anomaly detection results, which include the probability of failure and the fault type label.

5. The method according to claim 3, characterized in that, After generating diagnostic information for the target fault based on the inference results of the graph, the method further includes: Based on the diagnostic information, a repair strategy is generated for the target fault, and the repair strategy includes configuration adjustment, device restart, or traffic redirection; The repair strategy is executed automatically, and the repair effect is verified using the performance metrics data. The parameters of the knowledge graph and the machine learning model are updated based on the repair results.

6. The method according to claim 5, characterized in that, The automatic execution of the repair strategy includes: Based on the network topology information, the physical connection relationships, and the device dependencies, a repair path is dynamically generated. Perform predefined basic repair operations or composite repair operations according to the repair path. The basic repair operations include disk cleanup or service restart, and the composite repair operations include performing optical power calibration after switching to a backup port.

7. The method according to claim 3, characterized in that, The machine learning model includes a fault prediction model, and the method further includes: Based on the historical fault records, the device dependencies, the logical dependencies, and the network configuration information, supervised training is performed to generate an initial fault prediction model. The fault diagnosis process is optimized using reinforcement learning algorithms, and the parameters of the initial fault prediction model are adjusted based on the optimization results of the fault diagnosis process. The parameters of the initial fault prediction model are periodically fine-tuned based on newly collected multidimensional data to obtain the trained fault prediction model.

8. The method according to claim 1, characterized in that, The acquisition of multidimensional data from network devices includes: The network device is subjected to fine-grained testing through an atomic diagnostic module, which includes an optical power monitoring module, a broadband account status detection module, a routing and bridging mode verification module, a VLAN configuration check module, and a LAN port status monitoring module. The multidimensional data is formed based on the detection results of the atomic diagnostic module and the historical fault records.

9. A network fault diagnosis system, characterized in that, The system includes: The multidimensional data acquisition module is used to acquire multidimensional data of network devices, including device configuration information, network topology information, performance index data, and historical fault records. The knowledge graph construction module is used to construct a knowledge graph based on the multidimensional data. The knowledge graph includes nodes representing network devices, device links, network events, and device configuration items, as well as edges representing physical connection relationships, logical dependencies, device dependencies, network configuration information, and causal relationships between the nodes. The knowledge graph reasoning module is used to perform graph reasoning based on the knowledge graph to obtain graph reasoning results. The diagnostic information generation module is used to generate diagnostic information for the target fault based on the inference results of the graph.

10. An electronic device, characterized in that, include: One or more processors; and One or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, cause the electronic device to perform the network fault diagnosis method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The stored computer program causes the processor to execute the network fault diagnosis method as described in any one of claims 1 to 8.

Images