Fault root cause localization methods, devices, equipment, storage media and program products
By constructing a cross-layer heterogeneous graph involving faults and combining node features and connection edge features to calculate root cause scores, the problem of low accuracy in fault root cause localization in communication networks is solved, and efficient and accurate fault root cause tracing is achieved.
Patent Information
- Application Number
- CN202611425952.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-14
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, the accuracy of root cause localization of communication network faults is low because the hierarchical aggregation mode severs the inherent dependency of fault propagation across levels, leading to misjudgment of the root cause of the fault.
Construct a cross-layer heterogeneous graph involving faults, including service nodes, device nodes, and link nodes. Calculate root cause scores through node features and connection edge features, quantify the probability value of each node triggering alarm events when it is the root cause of a fault, and realize cross-layer root cause tracing.
It improves the accuracy and efficiency of fault root cause localization, accurately delineates the scope of fault impact, avoids indiscriminate processing of all network data, and enhances the pertinence and efficiency of fault response.
Smart Images

Figure CN123123115A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a method, apparatus, device, storage medium and program product for locating the root cause of a fault. Background Technology
[0002] As communication networks continue to expand, they will exhibit characteristics such as a wide variety of equipment types, deep bearer layers, complex service mapping relationships, and long fault propagation chains. When a single resource entity in a communication network fails, the fault signal will spread rapidly along service channels and physical connections, generating numerous resource entity alarms.
[0003] In related technologies, a communication network topology database can be pre-built, and a hierarchical model can be used to collect various resource entities corresponding to alarms. Fault determination is then based on manually set root cause localization rules. However, the hierarchical collection model only completes the correspondence between alarms and resource entities within a single level, severing the inherent dependency relationship of fault propagation across levels. This can easily lead to the misjudgment of derived alarms as the root cause of the fault, resulting in low accuracy in root cause localization. Summary of the Invention
[0004] This application provides a method, apparatus, device, storage medium, and program product for locating the root cause of a fault, which can solve the problem of low efficiency in locating the root cause of a fault.
[0005] In a first aspect, embodiments of this application provide a fault root cause localization method, the fault root cause localization method comprising: Acquire alarm events of the communication network within the first time window. An alarm event is an event in which at least two alarm objects issue alarms. Based on at least two alarm objects, extract service operation and maintenance information associated with at least two alarm objects from the operation and maintenance information of the communication network; Based on service operation and maintenance information and alarm events, a fault-related cross-layer heterogeneous graph is constructed. The first layer network graph in the fault-related cross-layer heterogeneous graph includes service nodes representing communication services, the second layer network graph includes device nodes representing transmission equipment, the third layer network graph includes link nodes representing physical links, and at least one layer network graph in the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to alarm events. The connection edge between any two nodes in the fault-related cross-layer heterogeneous graph represents the association relationship between any two nodes based on the communication network operation logic and different dimensions. Based on the node characteristics and edge characteristics of the nodes and connecting edges in the cross-layer heterogeneous graph involving faults, the root cause score corresponding to each node in the cross-layer heterogeneous graph involving faults is determined. The root cause score is used to represent the probability value that when the resource entity to which the corresponding node belongs is the root cause of the fault, it will propagate along the resource association relationship and trigger an alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
[0006] Secondly, embodiments of this application provide a fault root cause location device, which includes: The acquisition module is used to acquire alarm events of the communication network within the first time window. An alarm event is an event in which at least two alarm objects issue alarms. The extraction module is used to extract service operation and maintenance information associated with at least two alarm objects from the operation and maintenance information of the communication network. The construction module is used to construct a fault-related cross-layer heterogeneous graph based on service operation and maintenance information and alarm events. The first layer network graph of the fault-related cross-layer heterogeneous graph includes service nodes representing communication services, the second layer network graph includes device nodes representing transmission devices, the third layer network graph includes link nodes representing physical links, and at least one layer network graph of the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to alarm events. The connection edge between any two nodes in the fault-related cross-layer heterogeneous graph represents the association relationship between any two nodes based on the communication network operation logic and different dimensions. The determination module is used to determine the root cause score of each node in the cross-layer heterogeneous graph involving faults based on the node characteristics and edge characteristics of the connecting edges. The root cause score is used to represent the probability value that when the resource entity to which the corresponding node belongs is the root cause of the fault, it will propagate along the resource association relationship and trigger an alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
[0007] Thirdly, embodiments of this application provide a computer device, the computer device including: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements a fault root cause localization method as described in any of the first aspects.
[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement a fault root cause localization method as described in any of the first aspects.
[0009] Fifthly, embodiments of this application provide a computer program product, which includes a computer program or instructions that, when executed by a processor, implement a fault root cause localization method as described in any of the first aspects.
[0010] The fault root cause localization method, apparatus, device, storage medium, and program product of this application, by using alarm events of multiple alarm objects within a first time window as trigger anchor points for fault analysis, can accurately delineate the initial scope of the fault impact, avoiding indiscriminate processing of all network data, and improving the targeting and processing efficiency of fault response from the data source. Using alarm objects as indexes, associated service operation and maintenance information is extracted from operation and maintenance information, and a fault-related cross-layer heterogeneous graph is constructed by combining alarm events, including three types of resource nodes: service nodes, device nodes, and link nodes, as well as alarm event nodes, with connecting edges carrying different dimensions of correlation. Thus, the three-layer network structure of the fault-related cross-layer heterogeneous graph completely restores the hierarchical carrying and dependency relationships between physical links, transmission equipment, and upper-layer communication services in the communication network, matching the real law of fault propagation from bottom to top, realizing cross-layer root cause tracing, and providing a structured carrier that conforms to the fault propagation mechanism for root cause reasoning by incorporating alarm event nodes into a unified graph structure framework. The root cause score of each node is calculated based on the node and edge characteristics of the fault-related cross-layer heterogeneous graph. In this way, the attribute status information of the node itself and the association strength information carried by the connecting edge can be integrated at the same time to quantitatively assess the probability that each resource entity is the root cause of the fault and will cause alarms to be transmitted along the resource association relationship, which effectively improves the accuracy of root cause location. Attached Figure Description
[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A flowchart illustrating a fault root cause localization method provided in some embodiments of this application is shown. Figure 2 The diagram illustrates a specific implementation of step 130 provided in some embodiments of this application; Figure 3 A schematic diagram of a fault-related cross-layer heterogeneous graph in a fault root cause localization method provided in some embodiments of this application is shown; Figure 4 A flowchart illustrating a specific implementation of step 1304 provided in some embodiments of this application is shown; Figure 5 A flowchart illustrating a specific implementation of step 1303 provided in some embodiments of this application is shown; Figure 6 A flowchart illustrating a specific implementation of step 150 provided in some embodiments of this application is shown; Figure 7 A flowchart illustrating a specific implementation of step 160 provided in some embodiments of this application is shown; Figure 8 The present application provides a schematic diagram of the structure of a fault root cause localization device according to some embodiments; Figure 9 A schematic diagram of the structure of a computer device provided in some embodiments of this application is shown. Detailed Implementation
[0013] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.
[0014] It should be noted that, in this document, relational terms such as "first" and "second" are used merely 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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.
[0015] It should be noted that the acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.
[0016] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0017] To address the problems in the aforementioned related technologies, embodiments of this application provide a fault root cause localization method, apparatus, device, storage medium, and program product. The following description, in conjunction with the appendix... Figure 1 To be continued Figure 7The fault root cause localization method provided in this application will be described in detail through specific embodiments and application scenarios.
[0018] Figure 1 A flowchart illustrating some embodiments of the fault root cause localization method provided in this application is shown. Figure 1 As shown, the fault root cause localization method may include steps 110 to 140.
[0019] Step 110: Obtain alarm events of the communication network within the first time window. An alarm event is an event in which at least two alarm objects issue alarms. Step 120: Based on the at least two alarm objects, extract service operation and maintenance information associated with the at least two alarm objects from the operation and maintenance information of the communication network. Step 130: Based on the service operation and maintenance information and alarm events, construct a fault-related cross-layer heterogeneous graph. The first layer network graph in the fault-related cross-layer heterogeneous graph includes service nodes representing communication services; the second layer network graph includes device nodes representing transmission equipment; and the third layer network graph includes links representing physical links. The path node, at least one layer of the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to alarm events. The connection edge between any two nodes in the fault-related cross-layer heterogeneous graph represents the association relationship between any two nodes based on the communication network operation logic of different dimensions. Step 140, based on the node characteristics of the nodes and the edge characteristics of the connection edges in the fault-related cross-layer heterogeneous graph, determine the root cause score corresponding to each node in the fault-related cross-layer heterogeneous graph. The root cause score is used to represent the probability value of the corresponding node's resource entity as the root cause of the fault propagating along the resource association relationship and triggering an alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
[0020] Therefore, by using alarm events from multiple alarm objects within the first time window as trigger anchors for fault analysis, the initial scope of the fault impact can be accurately defined, avoiding indiscriminate processing of all network data and improving the targeting and efficiency of fault response from the data source. Using alarm objects as indexes, related service operation and maintenance information is extracted from operation and maintenance information, and a fault-related cross-layer heterogeneous graph is constructed by combining alarm events with three types of resource nodes: service nodes, device nodes, and link nodes, as well as alarm event nodes. Different dimensions of relationships are carried by connecting edges. Thus, the three-layer network structure of the fault-related cross-layer heterogeneous graph fully restores the hierarchical carrying and dependency relationships between physical links, transmission equipment, and upper-layer communication services in the communication network. It can match the real law of fault propagation from bottom to top, realize cross-layer root cause tracing, and provide a structured carrier that conforms to the fault propagation mechanism for root cause reasoning by incorporating alarm event nodes into a unified graph structure framework. The root cause score of each node is calculated based on the node and edge characteristics of the fault-related cross-layer heterogeneous graph. In this way, the attribute status information of the node itself and the association strength information carried by the connecting edge can be integrated at the same time to quantitatively assess the probability that each resource entity is the root cause of the fault and will cause alarms to be transmitted along the resource association relationship, which effectively improves the accuracy of root cause location.
[0021] The steps described above are explained in detail below.
[0022] First, regarding step 110, the communication network involved in this embodiment refers to a system composed of various communication services, transmission equipment, and physical links, used to realize information transmission and exchange. The communication network may include, but is not limited to: transmission networks, fiber optic networks, wireless networks, and data center networks. The first time window refers to the time window used for fault root cause localization of alarm events, such as 5 minutes or 10 minutes.
[0023] In this application's embodiments, an alarm event refers to a series of abnormal information detected and recorded by the monitoring system when one or more resource entities in a communication network experience an anomaly. An alarm event may include multiple alarm entries, each corresponding to a specific alarm object. The alarm object refers to the specific resource entity in the communication network that issues the alarm, which may include, but is not limited to, communication services, transmission equipment, and physical links.
[0024] In this step, the monitoring system in the communication network can continuously collect operational status information of network devices and services. When an anomaly is detected, the monitoring system generates an alarm entry and stores it in chronological order.
[0025] Secondly, regarding step 120, the service operation and maintenance information in this embodiment refers to comprehensive data used to describe the service operation status, resource configuration, topology, and performance indicators in the communication network. Service operation and maintenance information includes resource topology data, network configuration data, performance indicator data, and reference alarm event root cause localization cases.
[0026] In this step, after acquiring an alarm event, all alarm objects involved in the alarm event are identified from it, such as transmission device identifiers, physical link identifiers, and service names. The root cause localization system can access a centralized operations and maintenance information platform, which stores resource topology data, network configuration data, performance indicator data, and reference alarm event root cause localization cases. Service operations and maintenance information directly or indirectly related to at least two alarm objects is queried from the operations and maintenance information platform based on the alarm objects; for example, which services a certain alarm device carries, or which devices a certain alarm service depends on.
[0027] Furthermore, regarding step 130, the fault-related cross-layer heterogeneous graph in this embodiment refers to a multi-layer graph structure used to uniformly represent resource entities at different levels in a communication network, such as communication services, transmission equipment, physical links, and their interrelationships, and incorporates alarm event information. The graph structure of this fault-related cross-layer heterogeneous graph can reflect the path and impact of fault propagation between different levels. Service nodes refer to nodes representing various communication services provided in the communication network in the fault-related cross-layer heterogeneous graph. Device nodes refer to nodes representing various transmission devices in the communication network, such as routers, switches, servers, and optical transmission equipment. Link nodes refer to nodes representing physical links in the communication network, such as fiber optic links, Ethernet links, and virtual links. Alarm event nodes refer to nodes representing specific alarm events in the fault-related cross-layer heterogeneous graph, used to associate alarm information with the network topology. The relationships represented by connecting edges can be bearer relationships, dependency relationships, topological relationships, etc.
[0028] In some embodiments of this application, the connection edges include a first type of intra-layer connection edge representing the dependency call relationship between communication services, a second type of intra-layer connection edge representing the interconnection relationship between transmission devices, a third type of intra-layer connection edge representing the topological relationship between physical links, a first type of cross-layer connection edge representing the bearer dependency relationship between services and devices, a second type of cross-layer connection edge representing the bearer dependency relationship between devices and links, and an alarm association edge representing the triggering association relationship between alarm object anomalies and alarm events.
[0029] In this embodiment, the first type of intra-layer connection edge represents the dependency call relationship between communication services. For example, an upper-layer service may need to call multiple lower-layer services to complete its function, or there may be data flow or functional collaboration between multiple services. The second type of intra-layer connection edge represents the interconnection relationship between transmission devices. These relationships can be physical direct connections, such as fiber optic or cable connections, or logical interconnections, such as connections implemented through virtual LANs or tunneling technologies. The third type of intra-layer connection edge represents the topological relationship between physical links. For example, multiple physical links may be bundled into a logical link, or multiple fiber optic links may physically share a pipe or route. The first type of cross-layer connection edge represents the bearer dependency relationship between a service and a device, indicating that the operation of a communication service depends on a specific transmission device. For example, a service running on a virtual machine depends on the physical server it resides on, or a containerized service depends on the host machine on which it is deployed. The bearer dependency relationship between a service and a device can be obtained through service deployment information. The second type of cross-layer connection edge represents the bearer dependency relationship between a device and a link, indicating that a transmission device transmits data through a specific physical link. For example, a router connects to a fiber optic link through one of its interfaces. The bearer dependency between devices and links can be determined through network device interface configurations, port-to-link mappings, or network traffic routing paths. Alarm association edges represent the triggering association between alarm object anomalies and alarm events. When a resource entity experiences an anomaly, a corresponding alarm event is triggered. Alarm association edges connect alarm event nodes to the nodes mapped to the alarm object that caused the alarm event, such as service nodes, device nodes, or link nodes.
[0030] Therefore, by refining the connection edges in the fault-related cross-layer heterogeneous graph into three types—a first type of intra-layer connection edge representing the dependency and invocation relationship between communication services, a second type of intra-layer connection edge representing the interconnection relationship between transmission devices, a third type of intra-layer connection edge representing the topological relationship between physical links, a first type of cross-layer connection edge representing the bearer dependency relationship between services and devices, a second type of cross-layer connection edge representing the bearer dependency relationship between devices and links, and an alarm association edge representing the triggering relationship between alarm object anomalies and alarm events—the fault-related cross-layer heterogeneous graph can more accurately reflect the real dependencies, bearer relationships, and triggering relationships between resource entities at different levels, such as communication services, transmission devices, and physical links, in the communication network. For example, when a physical link node fails, the second type of cross-layer connection edge can accurately identify the affected device node, the first type of cross-layer connection edge can further identify the affected service node, and finally, the alarm association edge can locate the alarm event node generated by these affected services or devices. This hierarchical and categorized association modeling makes the fault propagation path clearer and more accurate, enabling more effective tracking of the fault propagation trajectory, thereby significantly improving the accuracy and efficiency of fault root cause localization.
[0031] In some embodiments of this application, service operation and maintenance information includes resource topology data, network configuration data, performance indicator data, and reference alarm event root cause localization cases. An alarm event includes at least two alarm entries, with each alarm entry corresponding to an alarm object; such as... Figure 2 As shown, step 130 above may specifically include steps 1301 to 1304.
[0032] Step 1301: Extract service nodes, device nodes, and link nodes from the resource topology data, and construct, based on the resource entity association relationships in the resource topology data, first-type intra-layer connection edges between service nodes and service nodes, second-type intra-layer connection edges between device nodes, third-type intra-layer connection edges between link nodes, first-type cross-layer connection edges between service nodes and device nodes, and second-type cross-layer connection edges between device nodes and link nodes.
[0033] Resource topology data refers to the data set of physical and logical connections between various resource entities in a communication network. Resource topology data can be pre-configured network device connection diagrams, service deployment architecture diagrams, and other similar data.
[0034] In this step, the resource topology data includes at least resource entities such as network elements, ports, boards, optical cables, optical paths, wave channels, service tunnels, and service circuits, as well as their relationships.
[0035] Step 1302: Based on at least two alarm entries, determine at least two alarm event nodes, and construct an alarm association edge between each alarm event node and the node mapped to the corresponding alarm object.
[0036] An alarm entry refers to an abnormal notification issued by a specific resource entity, i.e., an alarm object, at a specific point in time in the communication network. Each alarm entry may include, but is not limited to: alarm name, alarm level, alarm content, alarm timestamp, alarm object identifier, alarm repetition count, and clearing status.
[0037] In this step, an independent alarm event node can be created for each alarm entry. Then, based on the alarm object identifier recorded in the alarm entry, an alarm association edge is added between the alarm event node and the corresponding service node, device node, or link node in the fault-related cross-layer heterogeneous graph.
[0038] Step 1303: Based on network configuration data, performance index data, reference alarm event root cause localization cases, and alarm events, determine the node characteristics of each node in the fault-affected cross-layer heterogeneous graph.
[0039] Network configuration data includes configuration parameters, service policies, routing information, and security settings for various devices in the communication network, which can be obtained from the network management system or configuration management database. Performance metrics data refers to real-time or historical performance data generated by various resource entities in the communication network during operation, such as CPU utilization, memory utilization, bandwidth utilization, packet loss rate, latency, and error rate. Reference alarm event root cause localization cases refer to the set of mapping relationships between known alarm events and actual root causes in historical fault events.
[0040] In this step, device type, software version, interface configuration, and service importance can be extracted from network configuration data as resource attribute features; bit error rate, received optical level, port utilization, and clock deviation can be extracted from performance indicator data as operational performance features; alarm events can be characterized by the number of alarms, alarm level, alarm duration, and whether the alarms were compressed before being identified as alarm events; and reference alarm event root cause localization cases can be used to extract the total number of faults that occurred on each node in the past preset time period, the historical proportion of each node as the root cause of the fault, the statistical data of each node being the root cause along with other specific nodes, and the confidence level of each node as the root cause fault node in different fault scenarios as reference fault statistical features. These features are then vectorized and concatenated to form the node features.
[0041] Step 1304: Based on resource topology data, network configuration data, alarm events, and reference alarm event root cause localization cases, determine the edge weights of each connection edge in the fault-affected cross-layer heterogeneous graph to obtain the fault-affected cross-layer heterogeneous graph.
[0042] In this step, initial weights can be assigned to different types of edges based on resource topology data and network configuration data. For example, physical connections have higher weights than logical connections, and the bearing relationships of critical services have even higher weights. Simultaneously, the edge weights are dynamically adjusted based on the occurrence of alarm events and reference cases. For instance, if historical cases show that a link failure frequently causes an alarm on a certain device, the edge weight between that link and the device can be increased.
[0043] The following example illustrates the process of constructing a fault-related cross-layer heterogeneous graph.
[0044] A communication network experiences the following alarm events within a certain time window: A fiber optic cable E in the physical layer experiences a disconnection fault; transmission device D in the transport layer generates a device alarm; and communication services A and B in the service layer generate service interruption alarms. The process of constructing a fault-related cross-layer heterogeneous graph based on these alarm events is as follows: First, such as Figure 3 As shown, the alarm objects are identified as communication service A, communication service B, transmission device D, and fiber optic cable E. Using the alarm objects as indexes, service operation and maintenance information associated with these alarm objects is extracted from the operation and maintenance information of the communication network. The resource topology data of the service operation and maintenance information includes the relationships between communication services A and B in the service layer, transmission devices C and D in the transmission layer, and physical links E and F in the physical layer. Based on this resource topology data, the second type of intra-layer connection edge between transmission device C and transmission device D, the third type of intra-layer connection edge between physical link E and physical link F, the first type of cross-layer connection edge between communication service A and transmission device C, the first type of cross-layer connection edge between communication service B and transmission device D, the second type of cross-layer connection edge between transmission device C and physical link E, and the second type of cross-layer connection edge between transmission device D and physical link F are constructed. Thus, a three-layer heterogeneous topology foundation covering the service layer, transmission layer, and physical layer is formed, completely restoring the bottom-up carrying logic of physical links, transmission devices, and upper-layer communication services.
[0045] For each alarm entry, an alarm event node is created. When the physical layer fiber optic cable E experiences a disconnection fault, a corresponding alarm event node 1 is created. When the transmission device D in the transmission layer generates a device alarm, a corresponding alarm event node 2 is created. When the service layer communication service A generates a service interruption alarm, a corresponding alarm event node 3 is created. When the communication service B generates a service interruption alarm, a corresponding alarm event node 4 is created. Subsequently, alarm association edges are constructed between fiber optic cable E and alarm event node 1, between transmission device D and alarm event node 2, between communication service A and alarm event node 3, and between communication service B and alarm event node 4.
[0046] Through the above process, a fault-related cross-layer heterogeneous graph is constructed, which includes three layers of resource nodes, alarm event nodes, as well as intra-layer connection edges, cross-layer bearing edges, and alarm association edges.
[0047] In some embodiments of this application, the connecting edge further includes a time-adjacent directed edge, which is used to characterize the temporal sequence of alarm events. The alarm entry includes the timestamp of the corresponding alarm object issuing the alarm. Before step 1304, the fault root cause localization method further includes: for two alarm event nodes that have a resource topology association path or whose alarm time difference is within a preset time window, determining the order of alarm occurrence based on the timestamp in the corresponding alarm entry, and constructing a time-adjacent directed edge based on the order of alarm occurrence; the time-adjacent directed edge points from the alarm event node with the earlier alarm occurrence time to the alarm event node with the later alarm occurrence time.
[0048] Among them, the temporal adjacency directed edge is used to introduce the time dimension of alarm events into the fault-related cross-layer heterogeneous graph, so as to clearly represent the temporal order relationship between different alarm events. The temporal adjacency directed edge points from the alarm event node that occurred earlier to the alarm event node that occurred later. The temporal adjacency directed edge can intuitively reflect the temporal sequence of fault propagation in the graph structure.
[0049] In this step, all alarm event node pairs are traversed. For each alarm event node pair, it is first checked whether there is a resource topology association path between the alarm objects corresponding to the alarm event node pair, such as whether they share communication services, transmission equipment, or physical links. If a resource topology association path exists, the timestamp difference is further calculated. If the timestamp difference is less than a preset time window, such as 5 seconds, 10 seconds, or 1 minute, the time stamp determines which alarm occurred first and which occurred later, and a time-adjacent directed edge is constructed from the alarm event node that occurred earlier to the alarm event node that occurred later.
[0050] Therefore, when constructing a fault-related cross-layer heterogeneous graph, temporally adjacent directed edges are introduced to characterize the temporal sequence of alarm events. This allows the fault-related cross-layer heterogeneous graph to more comprehensively and accurately reflect the propagation path and causal chain of faults in the communication network. By utilizing the timestamps in alarm entries, combined with resource topology association paths and preset time windows, the temporal dependencies between alarm events can be identified and modeled, effectively distinguishing the cause and effect of the fault. This significantly improves the accuracy and reliability of fault root cause localization, avoids misjudging fault symptoms as root causes, and thus improves the efficiency of fault root cause determination.
[0051] In some embodiments of this application, such as Figure 4 As shown, step 1304 may specifically include steps 13041 to 13043.
[0052] Step 13041: Based on the resource topology data and network configuration data, determine the static topology weight and service impact weight of each connection edge. The static topology weight is used to characterize the inherent connection tightness and dependency strength of the resource entities mapped by the nodes connected by the connection edge at the communication network architecture and service deployment level. The service impact weight is used to characterize the value level and scope of impact of the communication services affected when the resource entities mapped by the nodes connected by the connection edge fail.
[0053] Static topology weights reflect the inherent physical and logical dependencies between resource entities in a communication network, covering multi-dimensional structural relationships such as intra-layer topology connections, cross-layer bearer dependencies, and service call relationships. Static topology weights can quantify the strength of these relationships by parsing resource topology data, identifying the bearer relationships between physical links and transmission equipment, the support relationships between transmission equipment and upper-layer services, and the interconnection relationships between resource entities within the same layer. For example, the static topology weight of a cross-layer bearer edge between a primary transmission link and core transmission equipment is higher than that of a connection edge corresponding to a redundant backup link; the static topology weight of an intra-layer connection edge between service nodes with direct service call relationships is higher than that of indirectly related service nodes.
[0054] Service impact weight is used to measure the degree of impact of resource failures associated with a connection edge on service continuity. It can be determined comprehensively by combining dimensions such as the importance of the service carried, customer level, leased line level, and service interruption scope. For example, the service impact weight of the connection edge corresponding to the transmission link carrying core financial leased lines and high-level customer services is set to a higher value. For example, the service impact weight can also be quantified by assessing the scope of service interruption and the scale of affected users that may be caused by the failure propagating along the edge.
[0055] Step 13042: Determine the alarm association weight of each connection edge based on the alarm event and the reference alarm event root cause localization case; the alarm association weight is used to characterize the confidence level of the causal relationship of fault propagation between resource entities mapped by the nodes connected by the connection edge.
[0056] Among them, the alarm association weight is dynamically generated based on historical fault data and current alarm characteristics, reflecting the co-occurrence pattern and temporal correlation characteristics of alarm events among different resource entities. It can be calculated by statistically analyzing historical alarm sequences and reference root cause localization cases to mine typical fault propagation paths and alarm association patterns.
[0057] In this step, the alarm association weight can be determined by comprehensively considering the co-occurrence frequency of alarm items within the same time window, the consistency of their timing, the matching degree of alarm levels, and the common cause probability of both appearing on the same fault propagation chain in the root cause localization case. For example, if the root cause localization case statistics show that, within a preset time window after a transmission device port alarm occurs, upper-layer service timeout alarms occur frequently, and both belong to the same fault propagation path multiple times, then the alarm association weight of the corresponding cross-layer connection edge will be assigned a higher value.
[0058] Step 13043: Determine the edge weights of each connection edge in the fault cross-layer heterogeneous graph based on the static topology weights, service impact weights, and alarm association weights of each connection edge.
[0059] In this step, for each connecting edge, min-max normalization is applied. For each type of weight, a linear mapping is performed based on the extreme values of the weights of the same type of edges across the entire graph to obtain the normalized static topology weight, the normalized service impact weight, and the normalized alarm association weight. Then, according to the preset fusion coefficients corresponding to each type of weight, the normalized static topology weight, the normalized service impact weight, and the normalized alarm association weight are weighted and fused to obtain the edge weight of each connecting edge.
[0060] Therefore, by integrating the weights of three dimensions—static structural attributes, service value attributes, and dynamic alarm attributes—the physical and logical boundaries of fault propagation are anchored based on inherent topological relationships. The service value dimension is used to match the priority requirements of operation and maintenance support. At the same time, historical alarm patterns are used to strengthen the causal relationship of fault propagation. This allows for a more comprehensive and accurate quantification of the weights of each connecting edge in the fault-related cross-layer heterogeneous graph. As a result, the constructed heterogeneous graph can more realistically reflect the complex dependencies between resource entities in the communication network, the importance of services, and the empirical patterns of fault propagation, thereby significantly improving the accuracy of root cause scoring.
[0061] In some embodiments of this application, such as Figure 5 As shown, step 1303 may specifically include steps 13031 to 13035.
[0062] Step 13031: Extract the resource attribute features of each node from the network configuration data. The resource attribute features are used to characterize the inherent resource attributes of the corresponding node.
[0063] Resource attribute characteristics may include, but are not limited to: device type, software version, interface configuration, and service importance.
[0064] Step 13032: Extract the operating performance characteristics of each node from the performance index data. The performance characteristics are used to characterize the real-time operating status and the degree of abnormal fluctuation of the corresponding node.
[0065] Among these, the operating performance characteristics may include, but are not limited to: bit error rate, received optical level, port utilization, and clock deviation.
[0066] Step 13033: Extract alarm event features of each node from the alarm events. The alarm event features are used to characterize the alarm event statistics of the corresponding node.
[0067] The alarm event characteristics may include, but are not limited to: alarm count, alarm level, alarm duration, and whether the alarm was compressed before feature compression.
[0068] Step 13034: Extract reference fault statistical features of each node from the reference alarm event root cause localization case. The reference fault statistical features are used to characterize the historical fault occurrence frequency and historical probability of the corresponding node as the root cause of the fault.
[0069] Among them, the reference fault statistics features may include, but are not limited to: the total number of times a node has failed in the past preset time period, the historical proportion of a node as the root cause of a fault, the statistical data of a node and other specific nodes acting as the root cause, and the confidence level of a node as the root cause fault node in different fault scenarios.
[0070] Step 13035: The resource attribute characteristics, performance characteristics, alarm characteristics and reference fault statistical characteristics of each node are spliced together to obtain the node characteristics of each node in the fault-related cross-layer heterogeneous graph.
[0071] In one example, the various node features of each node can be connected in a preset order to obtain the node features of each node.
[0072] In another example, the node features of each node can be combined using feature fusion algorithms, such as weighted average, principal component analysis, or feature fusion layers in deep learning, to obtain the node features of each node.
[0073] Therefore, by integrating multi-source heterogeneous information such as node static attributes, dynamic performance, real-time alarms, and historical experience into node features, each node has a more comprehensive and refined representation, enhancing the richness of node features in the fault-related cross-layer heterogeneous graph. This enables more effective identification of the true source of the fault during subsequent fault root cause localization, avoiding misjudgments caused by insufficient or incomplete information, and thus significantly improving the accuracy and efficiency of fault root cause localization.
[0074] Then, regarding step 140, the node features in this embodiment refer to the information carried by each node in the fault-related cross-layer heterogeneous graph, used to describe the specific state and characteristics of the corresponding node. Examples include resource attribute features, operational performance features, alarm event features, and reference fault statistical features. Edge features refer to the attribute information carried by each connecting edge in the fault-related cross-layer heterogeneous graph, used to describe the properties and strength of the connecting edge, such as edge type and edge weight.
[0075] In this embodiment, the root cause score refers to a numerical value calculated for each node by analyzing the cross-layer heterogeneous graph involving the fault. This value quantifies the probability that the resource entity mapped to that node will act as the root cause of the fault, propagating along resource relationships and triggering an alarm event. The higher the root cause score, the greater the likelihood that the node is the root cause of the fault.
[0076] In some embodiments of this application, the edge characteristics of the connecting edges include edge type and edge weight; based on this, step 140 above may specifically include: determining the root cause score corresponding to each node in the fault-related cross-layer heterogeneous graph based on the node characteristics of the nodes and the edge characteristics of the connecting edges through the root cause reasoning model.
[0077] In this step, the root cause reasoning model comprehensively utilizes graph structure information and node features to quantify the probability of each node being a root cause of a fault through learning and reasoning. The root cause reasoning model can be implemented based on various machine learning or deep learning frameworks, such as graph neural networks and causal graph models. By simulating the propagation process of a fault in a communication network, the root cause reasoning model can reverse-engineer the most likely initial fault point.
[0078] Therefore, by introducing a root cause inference model and utilizing the edge type and weight of connecting edges, it is possible to deeply fuse and learn multi-source heterogeneous information in fault-related cross-layer heterogeneous graphs. The root cause inference model can identify the role of different types of relationships in fault propagation and quantify their strength, thereby more accurately simulating the propagation path and causal chain of faults between different levels such as communication services, transmission equipment, and physical links. This allows for accurate identification of the most likely root cause of a fault event, significantly improving the accuracy and efficiency of fault root cause localization.
[0079] In some embodiments of this application, the alarm event includes at least two alarm entries. Based on this, before performing step 140 above, such as Figure 6 As shown, the above-mentioned fault root cause localization method may further include steps 1501 to 1504.
[0080] Step 1501: Generate time embedding features corresponding to each alarm event node based on the timestamps corresponding to each alarm entry in at least two alarm entries and the time difference between alarm entries.
[0081] Among them, the temporal embedding feature is a vectorized representation of alarm timing information in a high-dimensional feature space. It is used to characterize the occurrence time of alarm events, the time interval between alarms, and the temporal sequence relationship, providing feature basis for determining the temporal causal relationship of fault propagation.
[0082] In this step, the timestamps of alarm entries and the time differences between alarm entries can be encoded using a time encoding module. The time encoding module can be implemented in any of the following ways: using a periodic function encoding method composed of sine and cosine, or using an encoding method with a learnable embedding layer.
[0083] Step 1502: Map the time embedding features corresponding to each alarm event node to the target node mapped to the corresponding alarm object to obtain the time features of each target node.
[0084] In this context, the target node represents a resource entity in the communication network, such as a service node, device node, or link node. The target node is the node representation of the alarm object in the fault-related cross-layer heterogeneous graph.
[0085] In this step, the temporal embedding features of alarm event nodes can be mapped to the corresponding target nodes by direct copying and aggregation, which means associating the time information of the alarm occurrence with a specific resource entity. If multiple alarm events are mapped to the same target node, this can be achieved by averaging, summing, or maximizing the temporal embedding features of multiple alarm events, so that the target node also has a temporal dimension feature representation.
[0086] Step 1503: Concatenate the time embedding feature corresponding to each alarm event node with the original node feature of the corresponding alarm event node to obtain the updated node feature of each alarm event node; and concatenate the time feature of each target node with the original node feature of the corresponding target node to obtain the updated node feature of each target node.
[0087] In this step, the concatenation operation connects the time-embedded features with the original node features to form a longer feature that contains more information. For alarm event nodes, their original node features may include alarm type, alarm level, and other features; for target nodes, their original node features may include resource attribute features, operational performance features, and other features.
[0088] Step 1504: Use the updated node features of each alarm event node and the updated node features of each target node as the node features of the input graph neural network model.
[0089] In this step, the updated node features of the alarm event node and the updated node features of the target node are used as input to the graph neural network model. This allows the graph neural network model to make full use of these updated node features, including time information, when performing feature learning and information propagation. As a result, the graph neural network model can better capture the propagation patterns of faults in time and topology, thereby improving the accuracy of root cause reasoning.
[0090] Therefore, by converting the timestamps and time differences of alarm events into temporal embedding features and incorporating them into the node features of alarm event nodes and target nodes, the node features in the cross-layer heterogeneous graph involving faults become more comprehensive. As a result, the root cause reasoning model, when analyzing fault propagation paths, can more accurately capture the dynamic propagation process of faults, distinguish between causal relationships and accompanying relationships, and thus significantly improve the accuracy of fault root cause localization. Especially when facing complex, multi-level communication network faults, it can more effectively identify fault root causes that have strong causal relationships due to temporal sequence, avoiding misjudgments.
[0091] In some embodiments of this application, the root cause reasoning model includes a graph convolutional coding module and a root cause scoring module; based on this, such as Figure 7 As shown, step 140 above may include steps 1401 and 1402.
[0092] Step 1401: Input the node features of the nodes in the fault-related cross-layer heterogeneous graph and the edge features of the connecting edges into the graph convolutional coding module to obtain the fault propagation embedding features of each node output by the graph convolutional coding module. The graph convolutional coding module is used to divide the neighboring nodes of each node according to the edge type to obtain the neighboring node subsets corresponding to different relationship types, and call the edge-level attention calculation module to calculate the edge-level attention coefficients of each edge based on the features of the neighboring node pairs and the edge types of the corresponding edges. The edge-level attention coefficients are fused with the edge weights and used as the aggregation weights to perform weighted aggregation on the neighboring node features in each neighboring node subset, and output the fault propagation embedding features.
[0093] The graph convolutional coding module is a component of the root cause reasoning model used for feature learning and graph structure encoding of fault-related cross-layer heterogeneous graphs. Specifically, the graph convolutional coding module aggregates local neighborhood information of nodes into the node's own feature representation through graph convolution operations, thereby generating embedded features that reflect the fault propagation path and scope of influence. The graph convolutional coding module can employ various graph convolutional network structures, such as graph attention networks, graph convolutional networks, or heterogeneous graph neural networks.
[0094] The edge-level attention computation module dynamically evaluates the contribution of different neighboring nodes to the feature aggregation of the current node during graph convolutional encoding. By learning an attention function, the module assigns an attention coefficient to each edge based on the features of neighboring node pairs and the type of connecting edges. This allows the model to distinguish the importance of different types of edges in fault propagation, thus capturing the path and intensity of fault propagation more precisely. The edge-level attention computation module can be implemented based on a multi-head attention mechanism.
[0095] In this embodiment, the fault propagation embedding feature is the output of the graph convolutional coding module, which is a low-dimensional vector representation of each node after graph convolution. The fault propagation embedding feature encodes the node's own attribute information, as well as the node's local structural information and fault propagation context information in the fault-related cross-layer heterogeneous graph. The fault propagation embedding feature can capture the potential patterns and impacts of fault propagation from one node to another, providing rich semantic information for subsequent root cause scoring.
[0096] In this step, the graph convolutional coding module can be implemented using the following formula (1): (1) in, Let v be the hidden representation of node v at level l. Let v be the set of neighbors of node v under relation r. These are the learnable parameters for the correspondence. σ represents the edge attention weights, and σ is the non-linear activation function.
[0097] Step 1402: Input the fault propagation embedding feature of each node into the root cause scoring module to obtain the root cause score of each node output by the root cause scoring module; wherein, the root cause scoring module is used to calculate the node attention weight of each node based on the fault propagation embedding feature of each node, and output the root cause score of each node based on the fault propagation embedding feature, service impact feature and reference fault statistical feature of each node.
[0098] The root cause scoring module calculates the root cause score for each node based on the fault propagation embedding features output by the graph convolutional coding module, combined with other relevant features. The root cause scoring module may include one or more fully connected layers or attention mechanisms to map the learned embedding features to the final root cause probability value. Specifically, the root cause scoring module can be implemented using classifiers such as logistic regression, support vector machines, or neural networks.
[0099] In this embodiment, the node attention weight is a weight value assigned to each node by the root cause scoring module when calculating the root cause score. The node attention weight reflects the likelihood or importance of the node as the root cause of the fault in the current fault scenario. The node attention weight can be implemented using the following formulas (2) to (3): (2) in, Let be the initial attention score for node v. The fault propagation embedding feature of node v output by the graph convolutional coding module. It is a learnable linear transformation matrix. is a learnable bias, and 'a' is a learnable attention vector.
[0100] (3) in, Let V be the node-level attention weights for node v. The initial attention score for all nodes.
[0101] Service impact features characterize the potential impact on communication services when a resource entity mapped to a node fails. Service impact features may include, but are not limited to: service interruption duration, number of affected users, and service level agreement (SLA) violation risk. Service impact features can be extracted from service operation and maintenance information, such as service topology, traffic volume data, or SLA configuration. Reference fault statistics feature characterizes statistical information about the resource entity mapped to a node as a root cause of historical faults. Reference fault statistics features may include, but are not limited to: the frequency of historical faults occurring at the node, the proportion of faults as root causes, and the average repair time. Reference fault statistics features can be extracted and statistically analyzed from a historical alarm event root cause localization case library.
[0102] (4) in, For the service impact characteristics and reference fault statistics of node v, Let V be the node-level attention weights for node v. The fault propagation embedding feature of node v output by the graph convolutional coding module. , The learnable weights and biases for the root cause scoring module.
[0103] Therefore, by introducing a graph convolutional coding module and a root cause scoring module, the root cause inference model can learn the relationships between different types of nodes and edges in the cross-layer heterogeneous graph involving faults through graph convolution operations, and generate fault propagation embedding features with rich semantic information. The introduction of the edge-level attention calculation module further enhances the model's ability to perceive the importance of different relationship types, making the modeling of fault propagation paths more accurate. The root cause scoring module can comprehensively utilize fault propagation embedding features, service impact features, and reference fault statistical features to perform multi-dimensional fusion inference, thereby outputting a more accurate root cause score. This significantly improves the accuracy of fault root cause localization.
[0104] In some embodiments of this application, after obtaining the root cause score for each node in the cross-layer heterogeneous graph involving the fault, the nodes are sorted in descending order according to the root cause scores, and Top-K candidate root cause objects are output according to a preset parameter K. Here, K can be 3, meaning the top three nodes with the highest root cause confidence are output as the final recommended set of candidate root causes. It is understood that the value of K can be adaptively configured according to the alarm scale and network layer complexity of the operation and maintenance scenario. The sorted candidate results simultaneously carry the root cause score, node characteristics, and associated alarm list for each node.
[0105] In some embodiments of this application, for each output candidate root cause node, corresponding fault propagation chain explanation information is generated synchronously. The fault propagation chain starts from the candidate root cause node, traces back the affected nodes and associated alarms along the topological connection relationship and alarm propagation direction, and restores the propagation path of the fault from the root cause node to the underlying resources and upper-layer services layer by layer.
[0106] In this embodiment, the generated fault propagation chain explanation information may include, but is not limited to: node identifiers and edge types on the propagation path, alarm occurrence sequence of nodes at each level, and attention contribution ratio of each propagation edge. This fault propagation chain explanation information can assist operations and maintenance personnel in quickly locating the scope of fault impact, verifying the rationality of the root cause, and effectively shortening fault handling time.
[0107] In some embodiments of this application, the fault root cause localization method proposed in this application also includes a step of training the root cause inference model, specifically as follows: First, a supervised training dataset is constructed: training samples are constructed based on historical fault work orders, confirmed root cause network element records, and alarm playback samples. Each training sample corresponds to a real fault event, and the sample input is the fault-related cross-layer heterogeneous graph constructed by the fault event; the sample labels are configured according to the actual situation of the fault root cause: if the fault event has multiple collaborative root cause network elements, a multi-label training method is adopted, and all graph nodes corresponding to the real root cause network elements are marked with positive labels; if only the main root cause of the fault needs to be output, a single-label training method is adopted, and only the graph nodes corresponding to the main root cause network element are marked with positive labels, while the remaining nodes are marked with negative labels. To address the problem of sparse root cause samples and imbalanced positive and negative sample categories in communication network operation and maintenance scenarios, a weighted cross-entropy loss can be used to assign higher loss weights to the few root cause positive samples, thereby alleviating the training bias caused by the class imbalance. Training forward propagation phase: The fault-related cross-layer heterogeneous graph corresponding to the training samples is input into the root cause reasoning model. The feature aggregation and encoding process of step 1401 is executed through the graph convolutional coding module, and the fault propagation embedding features of each node in the graph are output. Then, the graph is input into the root cause scoring module to execute step 1402, and the root cause prediction score corresponding to each node v in the graph is calculated. , The predicted probability that node v is the root cause node of the failure; y v Let y be the actual label corresponding to node v. v =1 indicates that the node is the true root cause of the failure. v =0 indicates that the node is not the root cause of the failure.
[0108] The total loss function L used for model training can be achieved by the following formula (5): (5) Where V is the set of all nodes in the fault-related cross-layer heterogeneous graph; the first term in the formula is the cross-entropy loss, which, in a weighted scenario, can be increased by adding sample weight coefficients to measure the classification bias between the model's predicted root cause score and the true label; the second term is the L2 regularization term, where... This represents all trainable model parameters of the root cause reasoning module v, including the graph convolutional coding module and the root cause scoring module. This is a regularization hyperparameter, which is used to constrain the magnitude of model parameters and reduce the risk of overfitting.
[0109] Here, in a multi-label training scenario, it is allowed that the set V contains multiple real labels y across multiple nodes. v Simultaneously equal to 1, the loss function independently calculates the classification loss for all nodes in the graph, adapting to the business requirements of multi-cooperative root cause identification; in single-label training scenarios, only one node's yv The value is 1, and the values of the other nodes are y. v Setting the value to 0 is used to learn the unique principal cause of the output fault. After calculating the total loss L, backpropagation is performed using the gradient descent algorithm to iteratively update all trainable parameters of the model. Including but not limited to: the feature transformation matrix W corresponding to each relation type r The self-connection transformation matrix W0; the weight vector and linear transformation parameters corresponding to attention calculation; and the node attention calculation parameters W. a 1. Output layer weights and corresponding bias terms; repeat the forward propagation, loss calculation, and backward update process until the total loss converges or the preset number of training iterations is reached to complete the training of the root cause reasoning model.
[0110] Taking a fault scenario of a government and enterprise leased line network as an example, the root cause reasoning model that has been trained is used for reasoning verification. Alarm events include, within a certain time window, the interruption of the underlying optical cable of the transmission network, which simultaneously triggers multiple levels of alarms such as optical transmission network equipment port loss, cross-time slot abnormality, and leased line communication service interruption.
[0111] Based on this, a cross-layer heterogeneous fault graph is constructed using service operation and maintenance information and alarm events associated with alarm objects. This graph integrates optical cables, optical transport network equipment ports, cross-connections, leased line communication services, and alarm nodes. The graph is then input into a trained root cause inference model. The graph convolutional coding module of the root cause inference model automatically learns the hierarchical propagation pattern of underlying physical interruptions leading to mid-layer channel anomalies and upper-layer service alarms, thus completing the cross-layer aggregation of fault features. Finally, the root cause scoring module of the root cause inference model outputs the root cause score for each node, outputting the optical cable object and its associated ports as the highest-scoring candidate root cause. Simultaneously, it generates and pushes a complete propagation chain explanation from optical cable interruption to communication service interruption to maintenance personnel.
[0112] The model's performance was quantitatively evaluated through historical work order playback and offline validation. The results show that the root cause inference model can significantly reduce the range of candidate root causes, and the average root cause location latency is reduced from minutes to seconds compared to manual investigation, resulting in a significant improvement in fault handling response speed. Compared to simple rule filtering schemes or single-layer alarm association schemes, the root cause identification accuracy is significantly improved; moreover, on the completed validation dataset, the root cause location accuracy can be improved by more than 30%.
[0113] In some embodiments of this application, after performing step 140 above, as Figure 7 As shown, the above-mentioned fault root cause localization method may further include steps 1601 to 1605.
[0114] Step 1601: If a new alarm entry is detected in the communication network within the second time window, the local area with topological connection relationship is extracted from the cross-layer heterogeneous graph of the fault, based on the node mapped to the alarm object corresponding to the new alarm entry, to obtain a local subgraph. The second time window is the subsequent monitoring time window after the first time window for the corresponding alarm event.
[0115] The second time window is a continuous monitoring period that follows the first time window, used to capture new alarm entries that appear after the initial analysis.
[0116] In this step, the alarm database is periodically queried at preset time intervals, such as every 5 minutes, to check for new alarm records and identify new alarm entries. Each new alarm entry points to a specific alarm object, which corresponds to a node in the fault-related cross-layer heterogeneous graph. Centered on the node corresponding to this alarm object, a breadth-first search or depth-first search algorithm is used to traverse and extract all nodes and edges with topological connections to it within a preset hop count, such as 2 or 3 hops, thus forming a local subgraph.
[0117] Step 1602: Add a new alarm event node corresponding to the new alarm entry in the local subgraph, and construct an alarm association edge between the new alarm event node and the node mapped to the corresponding alarm object.
[0118] In this step, a unique alarm event node is created for each new alarm entry and added to the extracted local subgraph. Here, if multiple new alarm entries are issued by the same alarm object within a very short period of time, the multiple alarm entries can be aggregated into one alarm event node, and the quantity or severity can be associated in the node characteristics.
[0119] Step 1603: Construct a time-adjacent directed edge between the new alarm event node and the existing alarm event node whose alarm time difference is within a preset time window. The time-adjacent directed edge points from the alarm event node with the earlier alarm occurrence time to the alarm event node with the later alarm occurrence time.
[0120] In this step, once a new alarm event node is added to the local subgraph, the corresponding service node, device node, or link node is found in the local subgraph based on the alarm object recorded in the alarm entry. An alarm association edge is added between the new alarm event node and the corresponding node, and a specific edge weight and edge type are assigned to this alarm association edge to reflect the triggering strength or causal relationship between the alarm event and the alarm object. Furthermore, after adding a new alarm event node, all existing alarm event nodes in the local subgraph are traversed, and the time difference between the existing alarm event nodes and the new alarm event node is calculated. If the time difference is within a preset time window, such as 30 seconds or 1 minute, and the new alarm event occurred later than an existing alarm event, a time-adjacent directed edge is constructed from the existing alarm event node to the new alarm event node.
[0121] Step 1604: Update the node features of the affected nodes and the edge features of the corresponding connecting edges in the local subgraph to obtain the updated local subgraph.
[0122] In this step, the node characteristics of new alarm event nodes and their associated alarm object nodes are updated according to the new alarm entries, and the newly constructed alarm association edges and time-adjacent directed edges are assigned initial edge characteristics. Furthermore, in addition to new nodes and edges, the characteristics of other existing nodes related to the new alarm event, such as the service nodes hosted by the new alarm object, may also change due to the emergence of the new alarm; for example, the alarm event statistical characteristics may increase, and corresponding updates are also required.
[0123] Step 1605: Based on the node features of the nodes and the edge features of the connecting edges in the updated local subgraph, determine the root cause score corresponding to each node in the updated local subgraph.
[0124] In this step, the updated local subgraph is input into a pre-trained root cause reasoning model, which calculates root cause scores only for nodes in the local subgraph.
[0125] Therefore, when a new alarm entry is detected within the second time window, global graph reconstruction and root cause scoring are no longer performed. Instead, the local subgraph is extracted based on the new alarm object for incremental updates and local root cause scoring, which greatly reduces computational resource consumption and processing time, and significantly improves the timeliness and accuracy of fault location.
[0126] Based on the fault root cause localization method provided in the above embodiments, this application also provides specific implementation methods of the fault root cause localization device. Please refer to the following embodiments.
[0127] See Figure 8 The fault root cause location device 80 provided in this application embodiment includes: The acquisition module 801 is used to acquire alarm events of the communication network within the first time window. An alarm event is an event in which at least two alarm objects issue alarms. Extraction module 802 is used to extract service operation and maintenance information associated with at least two alarm objects from the operation and maintenance information of the communication network based on at least two alarm objects. Module 803 is used to construct a fault-related cross-layer heterogeneous graph based on service operation and maintenance information and alarm events. The first layer network graph in the fault-related cross-layer heterogeneous graph includes service nodes representing communication services, the second layer network graph includes device nodes representing transmission equipment, the third layer network graph includes link nodes representing physical links, and at least one layer network graph in the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to alarm events. The connection edge between any two nodes in the fault-related cross-layer heterogeneous graph represents the association relationship between any two nodes based on the communication network operation logic and different dimensions. The determination module 804 is used to determine the root cause score corresponding to each node in the cross-layer heterogeneous graph involving faults based on the node characteristics and edge characteristics of the connecting edges. The root cause score is used to represent the probability value that when the resource entity to which the corresponding node belongs is the root cause of the fault, it will propagate along the resource association relationship and trigger an alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
[0128] Therefore, by using the alarm events of multiple alarm objects within the first time window obtained by the acquisition module 802 as the trigger anchor points for fault analysis, the initial scope of the fault impact can be accurately delineated, avoiding indiscriminate processing of all network data and improving the targeting and efficiency of fault response from the data source. The extraction module 802 extracts related service operation and maintenance information from the operation and maintenance information using alarm objects as indexes. The construction module 803, combined with alarm events, constructs a fault-related cross-layer heterogeneous graph including three types of resource nodes: service nodes, device nodes, and link nodes, as well as alarm event nodes, with connecting edges carrying different dimensions of relationships. Thus, the three-layer network structure of the fault-related cross-layer heterogeneous graph fully restores the hierarchical carrying and dependency relationships between physical links, transmission equipment, and upper-layer communication services in the communication network. This matches the real law of fault propagation from bottom to top, enabling cross-layer root cause tracing. Furthermore, by incorporating alarm event nodes into a unified graph structure framework, it provides a structured carrier that conforms to the fault propagation mechanism for root cause reasoning. The root cause score of each node is calculated by module 804 based on the node and edge features of the fault-affected cross-layer heterogeneous graph. In this way, the attribute state information of the node itself and the association strength information carried by the connecting edges can be integrated simultaneously to quantitatively assess the probability that each resource entity is the root cause of the fault and will cause an alarm by propagating along the resource association relationship, thus effectively improving the accuracy of root cause location.
[0129] In some embodiments of this application, the connection edges include a first type of intra-layer connection edge representing the dependency call relationship between communication services, a second type of intra-layer connection edge representing the interconnection relationship between transmission devices, a third type of intra-layer connection edge representing the topological relationship between physical links, a first type of cross-layer connection edge representing the bearer dependency relationship between services and devices, a second type of cross-layer connection edge representing the bearer dependency relationship between devices and links, and an alarm association edge representing the triggering association relationship between alarm object anomalies and alarm events.
[0130] In some embodiments of this application, the construction module 803 in this application embodiment can be specifically used to extract service nodes, device nodes, and link nodes from the resource topology data when the service operation and maintenance information includes resource topology data, network configuration data, performance index data, and reference alarm event root cause localization cases, and the alarm event includes at least two alarm entries, with each alarm entry corresponding to an alarm object. Based on the resource entity association relationship in the resource topology data, it can construct a first type of intra-layer connection edge between service nodes and service nodes, a second type of intra-layer connection edge between device nodes, a third type of intra-layer connection edge between link nodes, a first type of cross-layer connection edge between service nodes and device nodes, and a second type of cross-layer connection edge between device nodes and link nodes. Based on at least two alarm entries, determine at least two alarm event nodes, and construct an alarm association edge between each alarm event node and the node mapped to the corresponding alarm object; Based on network configuration data, performance index data, reference alarm event root cause localization cases, and alarm events, determine the node characteristics of each node in the fault-affected cross-layer heterogeneous graph; Based on resource topology data, network configuration data, alarm events, and root cause localization cases of reference alarm events, the edge weights of each connecting edge in the fault-affected cross-layer heterogeneous graph are determined, thus obtaining the fault-affected cross-layer heterogeneous graph.
[0131] In some embodiments of this application, the fault root cause localization device 800 in this application embodiment may further include an edge construction module.
[0132] The edge construction module is used to determine the edge weights of each connection edge in the fault-related cross-layer heterogeneous graph based on resource topology data, network configuration data, alarm events, and reference alarm event root cause localization cases, before executing the step of determining the edge weights of each connection edge in the fault-related cross-layer heterogeneous graph based on resource topology association path or alarm time difference within a preset time window for two alarm event nodes with resource topology association path or alarm time difference within a preset time window, and constructing a time-adjacent directed edge based on the timestamp in the corresponding alarm entry; the time-adjacent directed edge points from the alarm event node with earlier alarm occurrence time to the alarm event node with later alarm occurrence time.
[0133] In some embodiments of this application, the construction module 803 in this application embodiment can be specifically used to determine the static topology weight and service impact weight of each connection edge based on resource topology data and network configuration data; the static topology weight is used to characterize the inherent connection tightness and dependency strength of the resource entities mapped by the nodes connected by the connection edge at the communication network architecture and service deployment level, and the service impact weight is used to characterize the value level and impact range of the communication service affected when the resource entities mapped by the nodes connected by the connection edge fail. Based on alarm events and reference alarm event root cause localization cases, determine the alarm association weight of each connection edge; the alarm association weight is used to characterize the confidence level of the causal association of fault propagation between resource entities mapped by the nodes connected by the connection edge. The edge weights of each connection edge in the fault cross-layer heterogeneous graph are determined based on the static topology weights, service impact weights, and alarm association weights of each connection edge.
[0134] In some embodiments of this application, the construction module in the embodiments of this application can be specifically used to extract the resource attribute features of each node from the network configuration data. The resource attribute features are used to characterize the inherent resource attributes of the corresponding node. The performance characteristics of each node are extracted from the performance index data. These performance characteristics are used to characterize the real-time operating status and the degree of abnormal fluctuation of the corresponding node's indicators. The alarm event features of each node are extracted from the alarm events. These alarm event features are used to characterize the alarm event statistics of the corresponding node. Reference fault statistical features of each node are extracted from the reference alarm event root cause localization cases. The reference fault statistical features are used to characterize the historical fault occurrence frequency and historical probability of the corresponding node as the root cause of the fault. By concatenating the resource attribute characteristics, performance characteristics, alarm characteristics, and reference fault statistical characteristics of each node, the node characteristics of each node in the fault-related cross-layer heterogeneous graph are obtained.
[0135] In some embodiments of this application, the determining module in these embodiments can be specifically used to determine the root cause score corresponding to each node in the fault-affected cross-layer heterogeneous graph based on the node characteristics of the nodes and the edge characteristics of the connecting edges, when the edge characteristics of the connecting edges include edge type and edge weight.
[0136] In some embodiments of this application, the fault root cause localization device 800 in this application embodiment may further include a time embedding module.
[0137] The time embedding module is used to generate time embedding features corresponding to each alarm event node based on the timestamps corresponding to each alarm entry in the at least two alarm entries and the time difference between the alarm entries when the alarm event includes at least two alarm entries and the alarm entries include the timestamps of the corresponding alarm objects. The time embedding features corresponding to each alarm event node are mapped to the target node mapped to the corresponding alarm object to obtain the time features of each target node; The time-embedded features corresponding to each alarm event node are concatenated with the original node features of the corresponding alarm event node to obtain the updated node features of each alarm event node; and the time features of each target node are concatenated with the original node features of the corresponding target node to obtain the updated node features of each target node. The updated node features of each alarm event node and the updated node features of each target node are used as the node features of the input graph neural network model.
[0138] In some embodiments of this application, the determination module 804 in this application embodiment can be specifically used to, when the root cause reasoning model includes a graph convolutional coding module and a root cause scoring module, input the node features of the nodes in the fault-related cross-layer heterogeneous graph and the edge features of the connecting edges into the graph convolutional coding module to obtain the fault propagation embedding features of each node output by the graph convolutional coding module; wherein, the graph convolutional coding module is used to divide the neighboring nodes of each node according to the edge type to obtain the neighboring node subsets corresponding to different relationship types, and call the edge-level attention calculation module to calculate the edge-level attention coefficient of each edge based on the features of the neighboring node pairs and the edge type of the corresponding edges, and to perform weighted aggregation on the neighboring node features in each neighboring node subset after fusing the edge-level attention coefficients with the edge weights, and output the fault propagation embedding features; The fault propagation embedding features of each node are input into the root cause scoring module to obtain the root cause score for each node. The root cause scoring module is used to calculate the node attention weight of each node based on the fault propagation embedding features of each node, and output the root cause score for each node based on the fault propagation embedding features, service impact features and reference fault statistics features of each node.
[0139] In some embodiments of this application, the fault root cause localization device 800 in the embodiments of this application may further include an incremental update module.
[0140] The incremental update module, after determining the root cause score for each node in the fault-affected cross-layer heterogeneous graph based on the node characteristics and edge characteristics of the connecting edges, performs the following steps: If a new alarm entry is detected in the communication network within the second time window, the local area with topological connection relationship is extracted from the fault-related cross-layer heterogeneous graph based on the node mapped to the alarm object corresponding to the new alarm entry, and a local subgraph is obtained. The second time window is the subsequent monitoring time window after the first time window and corresponding to the alarm event. Add a new alarm event node corresponding to the new alarm entry in the local subgraph, and construct an alarm association edge between the new alarm event node and the node mapped to the corresponding alarm object; Between a new alarm event node and an existing alarm event node whose alarm time difference is within a preset time window, a time-adjacent directed edge is constructed, with the time-adjacent directed edge pointing from the alarm event node whose alarm occurred earlier to the alarm event node whose alarm occurred later. Update the node features of the affected nodes and the edge features of the corresponding connecting edges within the local subgraph to obtain the updated local subgraph; Based on the node characteristics of the nodes and the edge characteristics of the connecting edges in the updated local subgraph, determine the root cause score corresponding to each node in the updated local subgraph.
[0141] The various modules of the fault root cause localization device 80 provided in this application embodiment can achieve Figures 1 to 7 The functions of each step in the provided fault root cause localization method and the corresponding technical effects are described briefly and will not be elaborated here.
[0142] Figure 9 The illustration shows a schematic diagram of the hardware structure of a computer device provided in some embodiments of this application.
[0143] The computer device 90 may include a processor 901 and a memory 902 storing computer program instructions.
[0144] Specifically, the processor 901 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.
[0145] Memory 902 may include mass storage for data or instructions. For example, and not limitingly, memory 902 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 902 may include removable or non-removable (or fixed) media. Where appropriate, memory 902 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 902 is non-volatile solid-state memory.
[0146] In a particular embodiment, memory 902 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory 902 includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the fault root cause localization method in the above embodiments according to this application.
[0147] The processor 901 reads and executes computer program instructions stored in the memory 902 to implement any of the fault root cause localization methods in the above embodiments.
[0148] In one example, computer device 90 may further include a communication interface 903 and a bus 910. Wherein, as... Figure 9 As shown, the processor 901, memory 902, and communication interface 903 are connected through bus 910 and complete communication with each other.
[0149] The communication interface 903 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.
[0150] Bus 910 includes hardware, software, or both, that couples components of computer device 90 together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count Bus (LPC), a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VESA Local Bus, VLB), or other suitable buses, or a combination of two or more of these. Where appropriate, bus 910 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.
[0151] The computer device 90 can execute the fault root cause localization method in the embodiments of this application, thereby achieving a combination of Figures 1 to 8 The described method and apparatus for locating the root cause of a fault.
[0152] Furthermore, in conjunction with the fault root cause localization methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the fault root cause localization methods in the above embodiments. Examples of computer-readable storage media include non-transitory computer-readable storage media, such as portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, etc.
[0153] Furthermore, in conjunction with the fault root cause localization methods in the above embodiments, this application embodiment can provide a computer program product for implementation. This program product is stored in a storage medium and may specifically include a computer program or instructions. When executed by a processor, the computer program or instructions implement any of the fault root cause localization methods in the above embodiments. This program product is executed by at least one processor to implement the various processes of the fault root cause localization method embodiments described above, and can achieve the same technical effects. To avoid repetition, further details are omitted here.
[0154] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.
[0155] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this disclosure are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable read-only memory (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.
[0156] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.
[0157] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in 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, a special-purpose computer, or other programmable root cause location apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable root cause location apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.
[0158] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.
Claims
1. A method for locating the root cause of a fault, characterized in that, include: Acquire alarm events of the communication network within a first time window, wherein the alarm event is an event in which at least two alarm objects issue alarms; Based on the at least two alarm objects, extract service operation and maintenance information associated with the at least two alarm objects from the operation and maintenance information of the communication network; Based on the service operation and maintenance information and the alarm events, a fault-related cross-layer heterogeneous graph is constructed; wherein, the first layer network graph of the fault-related cross-layer heterogeneous graph includes service nodes representing communication services, the second layer network graph of the fault-related cross-layer heterogeneous graph includes device nodes representing transmission equipment, the third layer network graph of the fault-related cross-layer heterogeneous graph includes link nodes representing physical links, and at least one layer network graph of the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to the alarm events, and the connection edge between any two nodes in the fault-related cross-layer heterogeneous graph represents the association relationship between the two nodes based on the communication network operation logic and different dimensions. Based on the node characteristics and edge characteristics of the nodes and connecting edges in the fault-related cross-layer heterogeneous graph, the root cause score corresponding to each node in the fault-related cross-layer heterogeneous graph is determined. The root cause score is used to represent the probability value that when the resource entity to which the corresponding node belongs is the root cause of the fault, the fault propagates along the resource association relationship and triggers the alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
2. The method according to claim 1, characterized in that, The connection edges include: a first type of intra-layer connection edge representing the dependency call relationship between communication services; a second type of intra-layer connection edge representing the interconnection relationship between transmission devices; a third type of intra-layer connection edge representing the topological relationship between physical links; a first type of cross-layer connection edge representing the bearer dependency relationship between services and devices; a second type of cross-layer connection edge representing the bearer dependency relationship between devices and links; and an alarm association edge representing the triggering association relationship between alarm object anomalies and alarm events.
3. The method according to claim 2, characterized in that, The service operation and maintenance information includes resource topology data, network configuration data, performance index data, and reference alarm event root cause localization cases. The alarm event includes at least two alarm entries, and one alarm entry corresponds to one alarm object. The step of constructing a fault-related cross-layer heterogeneous graph based on the service operation and maintenance information and the alarm events includes: Service nodes, device nodes, and link nodes are extracted from the resource topology data. Based on the resource entity association relationships in the resource topology data, the following are constructed: a first type of intra-layer connection edge between service nodes and service nodes, a second type of intra-layer connection edge between device nodes, a third type of intra-layer connection edge between link nodes, a first type of cross-layer connection edge between service nodes and device nodes, and a second type of cross-layer connection edge between device nodes and link nodes. Based on the at least two alarm entries, at least two alarm event nodes are determined, and an alarm association edge is constructed between each alarm event node and the node mapped to the corresponding alarm object. Based on the network configuration data, the performance index data, the reference alarm event root cause localization case, and the alarm event, determine the node characteristics of each node in the fault-related cross-layer heterogeneous graph; Based on the resource topology data, the network configuration data, the alarm events, and the reference alarm event root cause localization cases, the edge weights of each connection edge in the fault-related cross-layer heterogeneous graph are determined, thus obtaining the fault-related cross-layer heterogeneous graph.
4. The method according to claim 3, characterized in that, The connecting edge also includes a time-adjacent directed edge, which is used to characterize the temporal sequence of alarm events. The alarm entry includes the timestamp of the alarm issued by the corresponding alarm object. Before determining the edge weights of each connection edge in the fault-affected cross-layer heterogeneous graph based on the resource topology data, the network configuration data, the alarm events, and the reference alarm event root cause localization cases, and obtaining the fault-affected cross-layer heterogeneous graph, the method further includes: For two alarm event nodes that have a resource topology association path or whose alarm time difference is within a preset time window, the order of alarm occurrence is determined according to the timestamp in the corresponding alarm entry, and a time-adjacent directed edge is constructed according to the order of alarm occurrence; the time-adjacent directed edge points from the alarm event node with the earlier alarm occurrence time to the alarm event node with the later alarm occurrence time.
5. The method according to claim 3, characterized in that, The step of determining the edge weights of each connection edge in the fault-affected cross-layer heterogeneous graph based on the resource topology data, the network configuration data, the alarm events, and the reference alarm event root cause localization cases includes: Based on the resource topology data and the network configuration data, determine the static topology weight and service impact weight of each connection edge; the static topology weight is used to characterize the inherent connection tightness and dependency strength of the resource entities mapped by the nodes connected by the connection edge at the communication network architecture and service deployment level; the service impact weight is used to characterize the value level and scope of impact of the communication services affected when the resource entities mapped by the nodes connected by the connection edge fail. Based on the alarm events and the root cause localization cases of the reference alarm events, the alarm association weight of each connection edge is determined; the alarm association weight is used to characterize the confidence level of the causal association of fault propagation between resource entities mapped by the nodes connected by the connection edge. The edge weights of each connection edge in the fault cross-layer heterogeneous graph are determined based on the static topology weights, service impact weights, and alarm association weights of each connection edge.
6. The method according to claim 3, characterized in that, The step of determining the node characteristics of each node in the fault-affected cross-layer heterogeneous graph based on the network configuration data, the performance index data, the reference alarm event root cause localization case, and the alarm event includes: Resource attribute features of each node are extracted from the network configuration data, and the resource attribute features are used to characterize the inherent resource attributes of the corresponding node. The operational performance characteristics of each node are extracted from the performance index data. These performance characteristics are used to characterize the real-time operational status and the degree of abnormal fluctuations in the corresponding nodes. The alarm event features of each node are extracted from the alarm events, and the alarm event features are used to characterize the alarm event statistics of the corresponding node; Reference fault statistical features of each node are extracted from the reference alarm event root cause localization cases. The reference fault statistical features are used to characterize the historical fault occurrence frequency and historical probability of the corresponding node as the root cause of the fault. The node characteristics of each node are obtained by concatenating the resource attribute characteristics, performance characteristics, alarm characteristics, and reference fault statistical characteristics of each node in the fault-related cross-layer heterogeneous graph.
7. The method according to claim 1, characterized in that, The edge features of the connecting edges include edge type and edge weight; determining the root cause score corresponding to each node in the fault-affected cross-layer heterogeneous graph based on the node features of the nodes and the edge features of the connecting edges includes: Using the root cause reasoning model, the root cause score corresponding to each node in the fault-related cross-layer heterogeneous graph is determined based on the node characteristics and edge characteristics of the connecting edges.
8. The method according to claim 7, characterized in that, The alarm event includes at least two alarm entries, each alarm entry including a timestamp indicating that the alarm was issued by the corresponding alarm object; the method further includes: Based on the timestamps corresponding to each alarm entry in at least two alarm entries and the time difference between alarm entries, generate the time embedding features corresponding to each alarm event node; The time embedding features corresponding to each alarm event node are mapped to the target node mapped to the corresponding alarm object to obtain the time features of each target node; The time-embedded features corresponding to each alarm event node are concatenated with the original node features of the corresponding alarm event node to obtain the updated node features of each alarm event node; and the time features of each target node are concatenated with the original node features of the corresponding target node to obtain the updated node features of each target node. The updated node features of each alarm event node and the updated node features of each target node are used as the node features input to the root cause reasoning model.
9. The method according to claim 7, characterized in that, The root cause reasoning model includes a graph convolutional coding module and a root cause scoring module; The step of determining the root cause score for each node in the fault-related cross-layer heterogeneous graph using the root cause reasoning model, based on the node characteristics and edge characteristics of the connected edges, includes: The node features and edge features of the nodes in the fault-related cross-layer heterogeneous graph are input into the graph convolutional coding module to obtain the fault propagation embedding feature of each node output by the graph convolutional coding module. The graph convolutional coding module is used to divide the neighboring nodes of each node according to the edge type to obtain the neighboring node subsets corresponding to different relationship types, and call the edge-level attention calculation module to calculate the edge-level attention coefficient of each edge based on the features of the neighboring node pairs and the edge type of the corresponding edge. The edge-level attention coefficients are fused with the edge weights and used as the aggregation weight to perform weighted aggregation on the neighboring node features in each neighboring node subset, and output the fault propagation embedding feature. The fault propagation embedding feature of each node is input into the root cause scoring module to obtain the root cause score of each node output by the root cause scoring module; wherein, the root cause scoring module is used to calculate the node attention weight of each node based on the fault propagation embedding feature of each node, and output the root cause score of each node based on the fault propagation embedding feature, service impact feature and reference fault statistical feature of each node.
10. The method according to claim 1, characterized in that, After determining the root cause score corresponding to each node in the fault-affected cross-layer heterogeneous graph based on the node characteristics and edge characteristics of the connecting edges, the method further includes: If a new alarm entry is detected in the communication network within the second time window, the local area with topological connection relationship is extracted from the fault-related cross-layer heterogeneous graph based on the node mapped to the alarm object corresponding to the new alarm entry, and a local subgraph is obtained. The second time window is the subsequent monitoring time window after the first time window, corresponding to the alarm event. A new alarm event node corresponding to the new alarm entry is added to the local subgraph, and an alarm association edge is constructed between the new alarm event node and the node mapped to the corresponding alarm object; Between the new alarm event node and the existing alarm event node whose alarm time difference is within a preset time window, a time-adjacent directed edge is constructed, wherein the time-adjacent directed edge points from the alarm event node whose alarm occurrence time is earlier to the alarm event node whose alarm occurrence time is later. Update the node features of the affected nodes and the edge features of the corresponding connecting edges in the local subgraph to obtain the updated local subgraph; Based on the node features of the nodes and the edge features of the connecting edges in the updated local subgraph, the root cause score corresponding to each node in the updated local subgraph is determined.
11. A fault root cause location device, characterized in that, include: The acquisition module is used to acquire alarm events of the communication network within a first time window, wherein the alarm event is an event in which at least two alarm objects issue alarms; The extraction module is used to extract service operation and maintenance information associated with the at least two alarm objects from the operation and maintenance information of the communication network. A construction module is used to construct a fault-related cross-layer heterogeneous graph based on the service operation and maintenance information and the alarm events. The first layer network graph in the fault-related cross-layer heterogeneous graph includes service nodes representing communication services; the second layer network graph includes device nodes representing transmission equipment; the third layer network graph includes link nodes representing physical links; at least one layer network graph in the fault-related cross-layer heterogeneous graph includes alarm event nodes corresponding to the alarm events; and the connection edges between any two nodes in the fault-related cross-layer heterogeneous graph represent different-dimensional associations between the two nodes based on the communication network operation logic. The determination module is used to determine the root cause score corresponding to each node in the fault-related cross-layer heterogeneous graph based on the node characteristics and edge characteristics of the connecting edges of the nodes in the fault-related cross-layer heterogeneous graph. The root cause score is used to represent the probability value that when the resource entity to which the corresponding node belongs is the root cause of the fault, it will propagate along the resource association relationship and trigger the alarm event. The root cause score is used to determine the fault root cause location result of the alarm event.
12. A computer device, characterized in that, The computer device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the fault root cause localization method as described in any one of claims 1-10.
13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the fault root cause localization method as described in any one of claims 1-10.
14. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the fault root cause localization method as described in any one of claims 1-10.
