Method, device and equipment for managing dynamic knowledge graph

Abstract

Embodiments of the present application provide a kind of dynamic knowledge graph's management and control method, device and equipment.It relates to operation and maintenance management technical field.The method comprises: in response to the trigger of target operation and maintenance event, obtain multiple candidate operation and maintenance events within a preset time length;In multiple candidate operation and maintenance events, find the cause event that exists causal association with target operation and maintenance event, generate the target triple corresponding to target operation and maintenance event;Based on target triple and the operation and maintenance triple that already exists in dynamic knowledge graph and the causal event chain corresponding to operation and maintenance triple, matching is carried out, and matching result is obtained, dynamic knowledge graph stores the plurality of causal event chains constructed according to time dimension, each causal event chain corresponds to multiple operation and maintenance events that exist causal association in time sequence and the operation and maintenance triple corresponding to operation and maintenance event;According to matching result, the update of corresponding dynamic knowledge graph is executed, and the updated dynamic knowledge graph is obtained.The present application is used to reach the effect of improving the efficiency and accuracy of operation and maintenance diagnosis.

Description

Technical Field

[0001] This application relates to the field of operation and maintenance management technology, and in particular to a method, device and equipment for controlling dynamic knowledge graphs. Background Technology

[0002] The Baseboard Management Controller (BMC) is an embedded System-on-Chip (SOC) system that includes a processor, DRAM, Flash memory, and numerous sensor interfaces. In the operation and maintenance management of large-scale server clusters, the BMC serves as a core component for hardware monitoring and fault diagnosis. It needs to process multi-source operational information in real time, including sensor data, log events, and configuration changes, to remotely manage, monitor, and diagnose server faults.

[0003] In related technologies, scattered texts such as server operation and maintenance manuals and fault logs are transformed into static knowledge graphs. BMC enables operation and maintenance queries by performing full-graph searches or re-indexing within the static knowledge graph. However, this approach cannot respond promptly to dynamic needs arising from changes in server hardware configurations and new failure modes, leading to delays or even misdiagnoses in operation and maintenance diagnostics. Summary of the Invention

[0004] This application provides a method, apparatus, and equipment for managing dynamic knowledge graphs, which can improve the efficiency and accuracy of operation and maintenance diagnosis.

[0005] In a first aspect, embodiments of this application provide a method for managing dynamic knowledge graphs, including:

[0006] In response to the triggering of a target maintenance event, obtain multiple candidate maintenance events within a preset time period;

[0007] Among multiple candidate operation and maintenance events, find the antecedent events that are causally related to the target operation and maintenance event, and generate the target triplet corresponding to the target operation and maintenance event;

[0008] Matching is performed between the target triplet and the existing operation and maintenance triplets and the corresponding causal event chains in the dynamic knowledge graph to obtain the matching results. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time sequence and the operation and maintenance triplets corresponding to the operation and maintenance events.

[0009] Based on the matching results, the corresponding dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph.

[0010] In one possible implementation, the matching results include: no match, conflict, and no conflict. Matching is performed based on the target triplet against existing operation and maintenance triplets and their corresponding causal event chains in the dynamic knowledge graph to obtain the matching results, including:

[0011] If the knowledge graph does not have an operation and maintenance triple that matches the target triple, the matching result is no match;

[0012] If the dynamic knowledge graph has an operation and maintenance triple that matches the target triple, and the causal event chain to which the target triple and the operation and maintenance triple belong conflict, the matching result is determined to be conflicting.

[0013] If the dynamic knowledge graph has an existing operation and maintenance triple that matches the target triple, and there is no conflict in the causal event chain to which the target triple belongs and the existing operation and maintenance triple belongs, the matching result is determined to be no conflict.

[0014] In one possible implementation, if the matching result indicates a conflict, the corresponding dynamic knowledge graph is updated based on the matching result to obtain the updated dynamic knowledge graph, including:

[0015] Determine whether the target triplet and the operation and maintenance triplet belong to the same causal event chain;

[0016] If they belong to the same causal event chain, calculate the semantic similarity between the target triplet and the other operation and maintenance triplets in the causal event chain. If the semantic similarity is greater than the preset similarity threshold, replace the operation and maintenance triplet with the target triplet; otherwise, discard the target triplet.

[0017] If they do not belong to the same causal event chain, calculate the consistency score of the target triplet and the consistency score of the operation and maintenance triplet based on the causal event chain to which the target triplet belongs and the causal event chain to which the operation and maintenance triplet belongs. Keep the triplet corresponding to the one with the higher consistency score in the dynamic knowledge graph.

[0018] In one possible implementation, based on the causal event chain to which the target triple belongs and the causal event chain to which the operation and maintenance triple belongs, the consistency score of the target triple and the consistency score of the operation and maintenance triple are calculated, including:

[0019] The target triplet and the operation and maintenance triplet are respectively used as the triplet to be tested;

[0020] The consistency score of the triplet to be tested is obtained by calculating the consistency between the triplet to be tested and the other triplets in the causal event chain to which the triplet to be tested belongs.

[0021] The consistency score of the triplet under test is obtained by weighting and multiplying the consistency scores based on the chain quality weights corresponding to the triplet under test. The chain quality weights are obtained by weighting the event credibility and chain length of the event chain to which the triplet under test belongs.

[0022] In one possible implementation, if the matching result indicates no conflict, the corresponding dynamic knowledge graph is updated based on the matching result to obtain the updated dynamic knowledge graph, including:

[0023] Establish a branch node in the event chain to which the operation and maintenance triple belongs, add the target triple to the branch node, and obtain the updated dynamic knowledge graph; the branch node has the same upstream connection relationship as the existing operation and maintenance triple.

[0024] In one possible implementation, if the matching result is a mismatch, the corresponding dynamic knowledge graph is updated based on the matching result to obtain the updated dynamic knowledge graph, including:

[0025] The target triples are trained to correspond to small-scale incremental models, and the vector representations of the event chains associated with the target triples are updated to integrate the target triples into the knowledge graph.

[0026] In one possible implementation, the management method for dynamic knowledge graphs further includes:

[0027] In response to receiving a retrieval request from the querying end, the retrieval request is parsed to obtain the query keywords or query vector;

[0028] The target subgraph is obtained by searching for subgraphs related to query keywords or query vectors in the dynamic knowledge graph through semantic indexing.

[0029] The target subgraph is converted into a question-and-answer context in the form of a relational path, and the question-and-answer context is sent to the query end.

[0030] In one possible implementation, the management method for dynamic knowledge graphs further includes:

[0031] In the process of responding to the target operation and maintenance event, an exclusive write lock is acquired on the partition involved in the target triplet through read-write locks, and a shared read lock is opened on the other partitions;

[0032] When updating the dynamic knowledge graph, the version number of the dynamic knowledge graph is updated synchronously.

[0033] In response to a received retrieval request, a shared read lock is allocated using a read-write lock to enable concurrent retrieval operations on the dynamic knowledge graph.

[0034] Secondly, embodiments of this application provide a BMC operation and maintenance method based on a dynamic knowledge graph, including:

[0035] In response to the hardware alarm signal detected by BMC, the target operation and maintenance event corresponding to the hardware alarm signal in the dynamic knowledge graph is triggered, and the updated dynamic knowledge graph is obtained. The dynamic knowledge graph is managed based on the first aspect and / or the dynamic knowledge graph management method of various possible implementations of the first aspect.

[0036] The target causal event chain associated with the hardware alarm signal was retrieved from the updated dynamic knowledge graph.

[0037] Based on the target causal event chain, output the corresponding repair strategy for the hardware alarm signal.

[0038] Thirdly, embodiments of this application provide a control device for dynamic knowledge graphs, comprising:

[0039] The acquisition module is used to acquire multiple candidate operation and maintenance events within a preset time period in response to the triggering of the target operation and maintenance event;

[0040] The processing module is used to find antecedent events that are causally related to the target operation and maintenance event among multiple candidate operation and maintenance events, and generate target triples corresponding to the target operation and maintenance event. Based on the target triples, it performs matching with existing operation and maintenance triples and corresponding causal event chains in the dynamic knowledge graph to obtain matching results. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time and the operation and maintenance triples corresponding to the operation and maintenance events. According to the matching results, the corresponding dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph.

[0041] In one possible implementation, the matching results include: no match, conflict, and no conflict; the processing module is specifically used for:

[0042] If the knowledge graph does not have an operation and maintenance triple that matches the target triple, the matching result is no match;

[0043] If the dynamic knowledge graph has an operation and maintenance triple that matches the target triple, and the causal event chain to which the target triple and the operation and maintenance triple belong conflict, the matching result is determined to be conflicting.

[0044] If the dynamic knowledge graph has an existing operation and maintenance triple that matches the target triple, and there is no conflict in the causal event chain to which the target triple belongs and the existing operation and maintenance triple belongs, the matching result is determined to be no conflict.

[0045] In one possible implementation, if the matching result indicates a conflict, the processing module is specifically used for:

[0046] Determine whether the target triplet and the operation and maintenance triplet belong to the same causal event chain;

[0047] If they belong to the same causal event chain, calculate the semantic similarity between the target triplet and the other operation and maintenance triplets in the causal event chain. If the semantic similarity is greater than the preset similarity threshold, replace the operation and maintenance triplet with the target triplet; otherwise, discard the target triplet.

[0048] If they do not belong to the same causal event chain, calculate the consistency score of the target triplet and the consistency score of the operation and maintenance triplet based on the causal event chain to which the target triplet belongs and the causal event chain to which the operation and maintenance triplet belongs. Keep the triplet corresponding to the one with the higher consistency score in the dynamic knowledge graph.

[0049] In one possible implementation, the processing module is further configured to:

[0050] The target triplet and the operation and maintenance triplet are respectively used as the triplet to be tested;

[0051] The consistency score of the triplet to be tested is obtained by calculating the consistency between the triplet to be tested and the other triplets in the causal event chain to which the triplet to be tested belongs.

[0052] The consistency score of the triplet under test is obtained by weighting and multiplying the consistency scores based on the chain quality weights corresponding to the triplet under test. The chain quality weights are obtained by weighting the event credibility and chain length of the event chain to which the triplet under test belongs.

[0053] In one possible implementation, if the matching result is that there is no conflict, the processing module is specifically used for:

[0054] Establish a branch node in the event chain to which the operation and maintenance triple belongs, add the target triple to the branch node, and obtain the updated dynamic knowledge graph; the branch node has the same upstream connection relationship as the existing operation and maintenance triple.

[0055] In one possible implementation, if the matching result is a non-match, the processing module is specifically used for:

[0056] The target triples are trained to correspond to small-scale incremental models, and the vector representations of the event chains associated with the target triples are updated to integrate the target triples into the knowledge graph.

[0057] In one possible implementation, the processing module is further configured to:

[0058] In response to receiving a retrieval request from the querying end, the retrieval request is parsed to obtain the query keywords or query vector;

[0059] The target subgraph is obtained by searching for subgraphs related to query keywords or query vectors in the dynamic knowledge graph through semantic indexing.

[0060] The target subgraph is converted into a question-and-answer context in the form of a relational path, and the question-and-answer context is sent to the query end.

[0061] In one possible implementation, the processing module is further configured to:

[0062] In the process of responding to the target operation and maintenance event, an exclusive write lock is acquired on the partition involved in the target triplet through read-write locks, and a shared read lock is opened on the other partitions;

[0063] When updating the dynamic knowledge graph, the version number of the dynamic knowledge graph is updated synchronously.

[0064] In response to a received retrieval request, a shared read lock is allocated using a read-write lock to enable concurrent retrieval operations on the dynamic knowledge graph.

[0065] Fourthly, embodiments of this application provide a BMC operation and maintenance device based on a dynamic knowledge graph, comprising:

[0066] The acquisition module is used to respond to the hardware alarm signal detected by the BMC, trigger the target operation and maintenance event corresponding to the hardware alarm signal in the dynamic knowledge graph, and acquire the updated dynamic knowledge graph. The dynamic knowledge graph is managed by the dynamic knowledge graph management method of the first aspect and / or various possible implementations of the first aspect.

[0067] The processing module is used to retrieve the target causal event chain associated with the hardware alarm signal from the updated dynamic knowledge graph; and output the repair strategy corresponding to the hardware alarm signal based on the target causal event chain.

[0068] Fifthly, embodiments of this application provide a management and control device for dynamic knowledge graphs, including: a memory and a processor;

[0069] The memory stores instructions that the computer executes;

[0070] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above, and / or, the second aspect and / or various possible implementations of the second aspect as described above.

[0071] In a sixth aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect, and / or the second aspect and / or various possible implementations of the second aspect.

[0072] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect, and / or the second aspect and / or various possible implementations of the second aspect.

[0073] The dynamic knowledge graph management method, apparatus, and device provided in this application receive and respond to target operation and maintenance events, search for antecedent events with causal relationships to the target operation and maintenance event from multiple candidate operation and maintenance events within a preset time period, and generate target triples corresponding to the target operation and maintenance event. Then, the target triples are matched with existing operation and maintenance triples in the dynamic knowledge graph, and the corresponding dynamic knowledge graph is updated based on the matching results. This enables the dynamic knowledge graph to evolve in a timely manner and be maintained in a targeted manner around the temporal causal relationship of operation and maintenance events. By updating the dynamic knowledge graph, the timeliness of multi-source operation and maintenance knowledge, the accuracy of association processing, and the stability and interpretability for fault diagnosis and operation and maintenance retrieval are improved. Attached Figure Description

[0074] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0075] Figure 1 A schematic diagram illustrating a scenario for the management and control method of dynamic knowledge graphs provided in this application embodiment;

[0076] Figure 2 A flowchart illustrating a method for managing dynamic knowledge graphs according to an embodiment of this application;

[0077] Figure 3 A flowchart illustrating a method for managing dynamic knowledge graphs according to an embodiment of this application;

[0078] Figure 4 A schematic diagram of the structure of the control device for the dynamic knowledge graph provided in the embodiments of this application;

[0079] Figure 5 A schematic diagram of the structure of the control device for the dynamic knowledge graph provided in the embodiments of this application;

[0080] Figure 6 A schematic diagram of the structure of the control device for the dynamic knowledge graph provided in this application embodiment.

[0081] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0082] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0083] Server BMC (Building Management Center) operations and maintenance fall under the field of intelligent fault diagnosis and knowledge management technology for embedded management environments. Typically deployed on the BMC side of servers or cluster nodes, it receives sensor alarms, system logs, asset configuration and status change information to provide operations and maintenance personnel with fault location, root cause analysis, and remediation suggestions. In this operational scenario, operational information is characterized by diverse sources, frequent changes, and complex event correlations, requiring rapid identification of operational information to clarify the causal relationships between operational events.

[0084] In related technologies, static knowledge bases, traditional knowledge graphs, or question-and-answer methods based on unstructured text are often used to process operational data. Historical alarms, equipment manuals, and log rules are typically pre-organized and stored uniformly, and knowledge maintenance is completed through keyword matching, fixed rule retrieval, or periodic offline updates. During operation, correlation judgments are made based on alarm content, timestamps, or data source weights, and fault causes or maintenance suggestions are output accordingly. In actual BMC operation and maintenance environments, the above methods are difficult to reflect hardware topology changes, the emergence of new fault modes, and knowledge changes after upgrades in a timely manner, leading to a disconnect between knowledge content and on-site conditions. In particular, when multiple sensor signals, log records, and configuration changes occur simultaneously, relying solely on time sequence or source weights for judgment often fails to accurately distinguish between the true antecedent events and accompanying events, easily retaining outdated knowledge or ignoring key causal relationships. Meanwhile, due to the lack of ability to locate causal chains and dynamically adjust the knowledge graph around the target operational event, the retrieval of relevant knowledge often results in an overly broad scope and slow location, ultimately leading to insufficient stability of diagnostic results and unclear reasoning paths. This makes it difficult to continuously provide low-latency and interpretable operational support on the resource-constrained BMC side. In summary, the inability to respond promptly to dynamic needs arising from server hardware configuration changes and new failure modes leads to delays in operational diagnosis and even misdiagnosis.

[0085] This application provides a method for managing a dynamic knowledge graph. After a target maintenance event is triggered, multiple candidate maintenance events within a preset time period are acquired. Antecedent events causally related to the target maintenance event are identified, and corresponding target triples are generated. Then, the target triples are matched with existing maintenance triples in the dynamic knowledge graph, and updates are performed based on the matching results. This allows the causal event chain constructed along the time dimension to dynamically adjust as maintenance events change. This approach enables the dynamic knowledge graph to evolve and be maintained in a timely manner around the temporal causal relationships of maintenance events. By updating the dynamic knowledge graph, the timeliness of multi-source maintenance knowledge, the accuracy of association processing, and the stability and interpretability for fault diagnosis and maintenance retrieval are improved.

[0086] In terms of system deployment, the dynamic knowledge graph management method provided in this application embodiment can be combined with the event-driven and daemon process architecture in the BMC embedded Linux environment. The data acquisition, knowledge graph storage and related processing modules work together to complete event acquisition, causal chain organization and graph update, so that operation and maintenance knowledge can be closer to the actual time-series evolution process, providing a more accurate foundation for subsequent fault diagnosis and knowledge retrieval.

[0087] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0088] Figure 1 This is a schematic diagram illustrating a scenario for the dynamic knowledge graph management method provided in this application embodiment, such as... Figure 1 As shown, the specific application scenario of this application embodiment includes user 11, server 12, and processing center 13, wherein:

[0089] A communication connection exists between server 12 and processing center 13. Through this connection, processing center 13 can obtain real-time sensor alarms, system logs, asset configurations, and status change information from server 12, thus receiving operational events. Subsequently, processing center 13 updates the dynamic knowledge graph stored within itself based on the newly obtained operational events.

[0090] Simultaneously, the processing center 13 can also receive retrieval requests initiated by user 11 through the query terminal, and respond to the retrieval requests by retrieving content corresponding to the retrieval request within the dynamic knowledge graph to obtain the question-and-answer context corresponding to the retrieval request. Afterwards, the processing center 13 can send the question-and-answer context to the query terminal. Optionally, the query terminal can be at least one of the human-computer interaction interfaces provided by the server 12, or provided by the processing center 13, or other methods enabling human-computer interaction.

[0091] Figure 2 A flowchart illustrating a method for managing dynamic knowledge graphs according to an embodiment of this application is shown below. Figure 2 As shown, the method includes:

[0092] S201. In response to the triggering of the target operation and maintenance event, obtain multiple candidate operation and maintenance events within a preset time period.

[0093] The target operation and maintenance event serves as a triggering input event, used to initiate processes such as candidate event backtracking, causal identification, and graph updating surrounding the target operation and maintenance event. The target operation and maintenance event can be an operation and maintenance event that triggers updates to the dynamic knowledge graph.

[0094] The preset duration is used to limit the causal backtracking window. The preset duration is a time period set backward from the trigger time of the target maintenance event. In practical applications, the preset duration can be customized according to actual needs.

[0095] Optionally, when an operation and maintenance event is received, the operation and maintenance event is marked as the target operation and maintenance event.

[0096] Optionally, trigger conditions can be set. When an operation and maintenance event that meets the trigger conditions is received, the operation and maintenance event is marked as the target operation and maintenance event. The trigger conditions can be restrictions on the operation and maintenance event type, device, timestamp, etc., used to select the target operation and maintenance event from multiple operation and maintenance events.

[0097] Specifically, received operation and maintenance (O&M) events are processed uniformly. Upon triggering a target O&M event, its trigger time is read. Then, a preset time period is retrieved backward from the trigger time to obtain a set of all O&M events within that preset time period. Subsequently, events related to the target O&M event are extracted from this set, forming multiple candidate O&M events. These candidate O&M events constitute the set of events to be analyzed, providing a candidate range for causal backtracking of the target O&M event. Furthermore, to ensure temporal consistency in subsequent causal searches, the multiple candidate O&M events can be sorted chronologically before output.

[0098] Optionally, when obtaining multiple candidate operation and maintenance events, event cleaning processing can also be performed on the multiple candidate operation and maintenance events to remove messages in the multiple candidate operation and maintenance events that are unrelated to the target operation and maintenance event, and only retain events that have a characterizing effect on state changes.

[0099] Based on the above processing, the generated candidate event set not only retains the temporal context before the occurrence of the target operation and maintenance event, but also narrows the retrieval scope for subsequent causal identification, so that the subsequent updates of the dynamic knowledge graph are based on relevant events.

[0100] S202. Among multiple candidate operation and maintenance events, find the antecedent events that are causally related to the target operation and maintenance event, and generate the target triplet corresponding to the target operation and maintenance event.

[0101] Antecedent events are upstream events that have a causal relationship with the target operation and maintenance event. Antecedent events are candidate operation and maintenance events that can serve as causes of the target operation and maintenance event. Target triples are structured knowledge units used to represent the target operation and maintenance event and its antecedent relationships; they are the basic objects for subsequent matching with existing operation and maintenance triples in the dynamic knowledge graph.

[0102] Operation and maintenance triples are used to describe the operation and maintenance knowledge records corresponding to existing operation and maintenance events in the dynamic knowledge graph.

[0103] Specifically, a relationship analysis approach is used to sequentially analyze the relationship between each candidate operation and maintenance event and the target operation and maintenance event among multiple candidate operation and maintenance events. This relationship analysis can combine chronological order, event content relationship, and causal relationship rules. Chronological order is used to determine whether a candidate event occurred before the target operation and maintenance event; event content relationship is used to determine whether there is a correlation between the two events; and causal relationship rules are used to determine whether a candidate event meets the corresponding causal conditions.

[0104] By comparing candidate maintenance events with target maintenance events based on event sequence, object association, and preset causal relationship rules, antecedent events are identified from multiple candidate events. When multiple antecedent events exist among the candidate maintenance events, one or more of these antecedent events can be selected as the final antecedent event. Those skilled in the art should understand that there are various ways to select antecedent events from multiple candidate events.

[0105] Then, the antecedent events, target maintenance events, and the causal relationships between them are combined to form a structured representation that meets the requirements for writing dynamic knowledge graphs, so as to obtain the target triplet corresponding to the target maintenance event.

[0106] In one possible implementation, the target triple can take the form of "antecedent event entity - causal relationship - target operation and maintenance event entity". If multiple antecedent events are identified as constituting the upstream cause of the target operation and maintenance event, multiple target triples can be generated separately.

[0107] Based on the above analysis, by locating the causal chain around the target operation and maintenance event, the retained antecedent events no longer rely solely on simple chronological order, but are filtered by combining correlation and causal rules, so that the generated target triple can more accurately reflect the causal relationship between operation and maintenance events.

[0108] S203. Based on the target triplet and the existing operation and maintenance triplets and the corresponding causal event chains in the dynamic knowledge graph, a matching is performed to obtain the matching result. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time sequence and the operation and maintenance triplets corresponding to the operation and maintenance events.

[0109] A dynamic knowledge graph is a knowledge graph carrier used to store and update operation and maintenance knowledge triples and causal event chains. A dynamic knowledge graph stores existing operation and maintenance triples, event entity nodes, relation edges, and multiple causal event chains. Each causal event chain organizes multiple operation and maintenance events according to the time dimension to reflect the causal sequence of the same fault evolution process and to provide historical context for the current target triple.

[0110] Operation and maintenance triples, as existing knowledge records in the graph, are the matching objects for the target triples and the objects for subsequent updates.

[0111] Optionally, the matching process can be collaboratively handled by the graph access module and the comparison module of the dynamic knowledge graph. Specifically, the graph access module retrieves a set of candidate operation and maintenance triples from the dynamic knowledge graph based on the relevant attributes in the target triple. Then, the comparison module performs a matching comparison on the candidate operation and maintenance triple set and outputs the matching results.

[0112] Specifically, based on the antecedent events and target operation and maintenance events in the target triple, the corresponding existing records are searched in the dynamic knowledge graph, and compared with the relevant causal event chains to determine the matching between the target triple and the operation and maintenance triple.

[0113] A match is considered complete when an operations and maintenance triplet corresponding to the target triplet exists in the dynamic knowledge graph. A match is considered incomplete when no operations and maintenance triplet corresponding to the target triplet exists in the dynamic knowledge graph.

[0114] Optionally, if the entities in the target triple and the operation and maintenance triple are identical, the target triple and the operation and maintenance triple are considered to match. If the entities in the target triple and the operation and maintenance triple are inconsistent, the target triple and the operation and maintenance triple are considered to be mismatched.

[0115] For example, if the target triple is "entity A-relation 1-entity B" and the operation triple is "entity A-relation 2-entity B", then the target triple matches the operation triple.

[0116] For example, if the target triple is "entity A-relationship 1-entity B" and the operation triple is "entity A-relationship 1-entity B", then the target triple matches the operation triple.

[0117] Optionally, if the target triple and the operation triple have at least one consistent entity, the target triple and the operation triple are considered to be a match.

[0118] For example, if the target triple is "entity A-relation 1-entity B" and the operation triple is "entity A-relation 2-entity C", then the target triple matches the operation triple.

[0119] Furthermore, to ensure that the dynamic knowledge graph can reflect state evolution, the correlation between the target triple and the operation triple can be combined during comparison to distinguish between antecedent events with real correlation paths and accompanying events that are only close in time.

[0120] Based on the above analysis, it can be seen that by organizing causal event chains in the time dimension and matching the target triple with the existing operation and maintenance triple in context, the scope of knowledge retrieval can be narrowed and the pertinence of causal judgment can be improved, so that the update of dynamic knowledge graph is based on a clearer reasoning path.

[0121] S204. Based on the matching results, perform the corresponding dynamic knowledge graph update to obtain the updated dynamic knowledge graph.

[0122] The update process is used to implement the impact of the current target triple on the dynamic knowledge graph into the storage of the dynamic knowledge graph, forming an updated dynamic knowledge graph that reflects the latest operational status and causal relationships. The corresponding dynamic knowledge graph update includes at least one of the following update methods: writing a new target triple, replacing an existing operational triple with the target triple, deleting conflicting operational triples, or keeping existing operational triples unchanged.

[0123] Specifically, the corresponding knowledge graph update strategy is invoked based on the matching results. If the target triple matches the operations triple, the existing record can be updated or left unchanged. If the target triple does not match the operations triple, the target triple is written into the dynamic knowledge graph.

[0124] Based on the matching status between the current target triples and the operation and maintenance triples, the relevant records in the graph are adjusted. For cases of mismatch, after writing the target triple, the update module can attach the target triple to the corresponding causal event chain based on the chronological order of the antecedent events and the target operation and maintenance events, or create a new causal event chain. During the writing process, relevant association information can be updated synchronously to ensure that subsequent searches can utilize the newly added knowledge.

[0125] The dynamic knowledge graph is updated using a transactional write mechanism. Before an update begins, a change set to be committed is built, and after successful verification, it is written to the graph storage. If an exception occurs during the write process, the system can roll back to the state before the update to ensure the consistency of the dynamic knowledge graph.

[0126] Based on the above analysis, it can be seen that by performing targeted updates according to the matching results, the causal event chain constructed according to the time dimension can be dynamically adjusted as the operation and maintenance events change. The updated dynamic knowledge graph can continuously retain causal relationship records consistent with the on-site status and provide a traceable knowledge base for subsequent fault location, cause analysis and handling suggestions.

[0127] This application provides a method for managing dynamic knowledge graphs. By establishing a candidate event set of multiple candidate maintenance events within a preset time period around a target maintenance event, a target triplet corresponding to the target maintenance event is generated. Subsequently, matching and updating are performed within a causal event chain constructed along the time dimension, enabling the dynamic knowledge graph to continuously reflect state evolution and newly emerging fault modes. By directly incorporating the historical context of the target maintenance event into the causal chain judgment process, the scope of searches unrelated to the current target is reduced, and the maintenance triplets in the graph are updated as on-site events evolve, thus providing the diagnostic path with clear temporal and relational basis.

[0128] The specific storage structure of the preset duration, the encoding method of the target triple, the matching result type, and the dynamic knowledge graph in this application can all be adjusted according to the actual operation and maintenance scenario, as long as it can realize causal backtracking around the target operation and maintenance event and complete the dynamic update of the knowledge graph.

[0129] Figure 3 A flowchart illustrating a method for managing dynamic knowledge graphs according to an embodiment of this application is shown below. Figure 3 As shown, in this embodiment... Figure 2 Based on the examples, the management method for dynamic knowledge graphs is described in detail, including:

[0130] In one possible implementation, the matching results include: mismatch, conflict, and no conflict. Step S203 may further include:

[0131] S2031. If the knowledge graph does not have an operation and maintenance triple that matches the target triple, the matching result is no match.

[0132] Perform consistency retrieval of the entities and relations in the target triples with the entities and relations in the existing operation and maintenance triples in the dynamic knowledge graph.

[0133] If no existing operation and maintenance triplet that meets the preset conditions for entity and relation is found, the target triplet is judged as a mismatch.

[0134] The preset conditions can be any one of the following: entity consistency, consistency of both entity and relation, and consistency of at least one entity.

[0135] S2032. If the dynamic knowledge graph has an operation and maintenance triple that matches the target triple, and the causal event chain to which the target triple and the operation and maintenance triple belong conflict, the matching result is determined to be conflict.

[0136] When an existing operation and maintenance triplet that meets at least one entity and relation satisfying the preset conditions is retrieved, the operation and maintenance triplet is determined as the operation and maintenance triplet that matches the target triplet.

[0137] Next, the causal event chain to which the operation and maintenance triplet matching the target triplet belongs is obtained. A conflict determination is then performed between the causal event chain to which the target triplet is to belong and the causal event chain to which the matching operation and maintenance triplet belongs, to determine whether a conflict exists between the target triplet and the causal event chain to which the operation and maintenance triplet belongs. Conflict determination can include at least one of the following: event sequence crossover conflict determination, causal conflict determination, and contradictory conflict determination.

[0138] When it is determined that there is a conflict between the target triple and the operation triple in the causal event chain, the matching result is determined to be a conflict.

[0139] S2033. If the dynamic knowledge graph has an existing operation and maintenance triple that matches the target triple, and there is no conflict in the causal event chain to which the target triple belongs and the existing operation and maintenance triple belongs, the matching result is determined to be that there is no conflict.

[0140] When it is determined that there is no conflict between the causal event chain to which the target triple and the operation triple belong, the matching result is determined to be no conflict.

[0141] First, a matching retrieval is performed based on the target triple to obtain the operation and maintenance triple that matches the target triple. Then, conflict analysis is performed on the causal event chain corresponding to the matched operation and maintenance triple to obtain the matching result, which directly reflects the adaptation status between the target triple and existing knowledge.

[0142] By adopting the above method, the access of target triples can be differentiated and processed, and different matching states can be mapped to different update branches, thereby ensuring the consistency of the causal event chain in the graph and providing more accurate matching basis for subsequent fault location and knowledge retrieval.

[0143] In one possible implementation, if the matching result indicates a conflict, step S204 may further include:

[0144] S2041. Determine whether the target triplet and the operation and maintenance triplet belong to the same causal event chain.

[0145] After determining that a conflict exists in the matching result, the entity and relation corresponding to the target triple are read first, and the operation and maintenance triple that conflicts with the target triple and its corresponding causal event chain are located in the dynamic knowledge graph. Then, it is determined whether the target triple and the operation and maintenance triple that conflicts with the target triple and its corresponding causal event chain belong to the same causal event chain.

[0146] S2042. If they belong to the same causal event chain, calculate the semantic similarity between the target triplet and the other operation and maintenance triplets in the causal event chain. If the semantic similarity is greater than the preset similarity threshold, replace the operation and maintenance triplet with the target triplet; otherwise, discard the target triplet.

[0147] If the target triple and the maintenance triple that conflicts with the target triple belong to the same causal event chain, then the semantic similarity between the target triple and the other maintenance triples in that causal event chain (excluding the conflicting object) is calculated. When the semantic similarity is greater than a preset similarity threshold, the target triple is written into the dynamic knowledge graph and replaces the original maintenance triple. When the semantic similarity is not greater than the preset similarity threshold, the original maintenance triple remains unchanged, and the target triple is discarded.

[0148] Optionally, firstly, semantic encoding is performed on the target triple and the other operation and maintenance triples within the causal event chain, excluding the conflicting object, to obtain the first semantic encoding of the target triple and the second semantic encoding of the other operation and maintenance triples. Based on vector cosine similarity or text matching similarity, the similarity between the first and second semantic encodings is calculated to obtain the semantic similarity.

[0149] S2043. If they do not belong to the same causal event chain, calculate the consistency score of the target triplet and the consistency score of the operation and maintenance triplet based on the causal event chain to which the target triplet belongs and the causal event chain to which the operation and maintenance triplet belongs. Keep the triplet corresponding to the one with the higher consistency score in the dynamic knowledge graph.

[0150] If the target triple and the operation and maintenance triple that conflicts with the target triple do not belong to the same causal event chain, then the consistency evaluation is performed on the causal event chain to which the target triple belongs and the causal event chain to which the operation and maintenance triple belongs, respectively, to calculate the consistency score of the target triple and the consistency score of the operation and maintenance triple. Optionally, semantic encoding can be implemented by a pre-trained language model, rule feature vectors, or a combination of both.

[0151] Optionally, after obtaining the consistency score, a chain quality weight is generated by combining the event credibility and chain length of the causal event chain. The consistency score is then weighted using this chain quality weight, and the weighted score is used as the final consistency score. The chain quality weight can be generated based on the weighted result of event credibility and chain length.

[0152] Next, the triples corresponding to the pair with the higher consistency score are retained in the dynamic knowledge graph. Specifically, if the target triple has a higher consistency score, it is written into the dynamic knowledge graph and replaces the original operations triple. If the operations triple has a higher consistency score, the original operations triple remains unchanged, and the target triple is discarded.

[0153] Through the above methods, intra-chain conflicts are replaced or discarded based on semantic similarity, while inter-chain conflicts are retained based on chain-level consistency scores. This enables the dynamic knowledge graph to perform local corrections and inter-chain adjudication around causal event chains. As a result, the knowledge retained in the graph remains consistent with the causal relationships of current operational events, and the updated knowledge units are more suitable for subsequent retrieval and diagnostic reasoning, while reducing the impact of conflicting knowledge on the graph's stability.

[0154] In one possible implementation, based on the causal event chain to which the target triple belongs and the causal event chain to which the operation and maintenance triple belongs, the consistency score of the target triple and the consistency score of the operation and maintenance triple are calculated, including: treating the target triple and the operation and maintenance triple as triples to be tested in sequence; obtaining the consistency score corresponding to the triple to be tested by calculating the consistency between the triple to be tested and the other triples in the causal event chain to which the triple to be tested belongs; and multiplying the consistency score by weighted summation based on the chain quality weight corresponding to the triple to be tested, wherein the chain quality weight is obtained by weighting the event credibility and chain length of the event chain to which the triple to be tested belongs.

[0155] First, the target triplet and the operation and maintenance triplet are selected as the triplets to be tested in turn.

[0156] Next, the remaining triples (excluding the triples in the test) are extracted from the causal event chain to which the triples in the test belong. Based on the knowledge corresponding to the remaining triples, the consistency score between the remaining triples and the triples in the test is calculated. The knowledge includes any one of the following: entity semantic vectors, relation type vectors, and temporal adjacency constraints.

[0157] Optionally, a similarity score can be obtained by calculating the similarity between the test triple and each of the other triples. Then, the similarity scores are summed and normalized to obtain a consistency score. Furthermore, the similarity scores can be weighted and summed based on causal direction consistency, temporal continuity, and relationship type to obtain a final consistency score.

[0158] Event credibility can be determined by a combination of the event source credibility level, alarm hit confidence, and log evidence completeness. Chain length can be directly taken as the number of triples in the causal event chain. Then, the event credibility and chain length are normalized and weighted to obtain the chain quality weight. When the event credibility of the causal event chain is high and the chain length is long, the chain quality weight increases accordingly. When the event credibility of the causal event chain is low and the chain length is short, the chain quality weight decreases accordingly.

[0159] Multiplying the chain quality weight by the consistency score yields the consistency score of the triple under test. This method allows us to obtain the consistency scores of both the target triple and the operational triple.

[0160] The consistency score is used to determine which of the target triples and the operational triples is more suitable to be retained in the dynamic knowledge graph. The triples with higher consistency scores are considered to be more consistent with the current causal event chain structure. The consistency score not only considers the consistency between a single triple and its adjacency relationships within the chain, but also introduces overall quality constraints on the causal event chain, thereby making the retention results more consistent with the actual causal evolution in the operational field.

[0161] By adopting the above method, the calculation of triple scores reflects both local consistency and overall chain quality, thus improving the stability of judgment during conflict resolution and making the causal relationships retained in the dynamic knowledge graph more consistent with the temporal correlation and the credibility of evidence in actual operation and maintenance scenarios.

[0162] In one possible implementation, if the matching result is that there is no conflict, step S204 may further include:

[0163] S2044. Establish a branch node in the event chain to which the operation and maintenance triple belongs, add the target triple to the branch node, and obtain the updated dynamic knowledge graph; the branch node has an upstream connection relationship consistent with the existing operation and maintenance triple.

[0164] Once it is determined that the target triplet does not conflict with the existing operation and maintenance triplet, the causal event chain in which the existing operation and maintenance triplet belongs is located. Then, a branch node is generated in the storage subgraph corresponding to the causal event chain, and the parent node, sibling nodes, and event source constraints of the branch node are kept consistent with the original triplet.

[0165] Subsequently, the target triple is written into the edge set and node attribute set corresponding to the branch node, so that the event identifier, resource identifier, timestamp and causal direction of the target triple are simultaneously included in the dynamic knowledge graph index.

[0166] Optionally, to facilitate incremental updates, branch nodes can adopt a lightweight adjacency list structure or a pointer-based index structure. In practical applications, branch nodes can also be implemented in other ways, and this application does not limit this.

[0167] Optionally, perform cycle detection on the updated dynamic knowledge graph. If a cycle is detected, modify the causal edges forming the cycle in the branch nodes to undirected edges.

[0168] After the write operation is complete, a loop detection process is performed on the updated dynamic knowledge graph. Loop detection is used to determine whether loops exist in the dynamic knowledge graph.

[0169] Optionally, based on depth-first traversal or topological backtracking, determine whether the newly added edge points back to a visited node in order to perform cycle detection on the updated dynamic knowledge graph.

[0170] If a loop is detected, a type conversion is performed on the causal edges participating in the closure within the branch node, changing the original unidirectional causal edges to undirected edges to maintain the reachability between nodes while preserving the weak correlation information between events. Through this processing, the target triple can be incorporated into the original causal event chain in the form of a branch, and the dynamic knowledge graph maintains structural consistency while preserving its updability.

[0171] In this way, the newly added target triples can be incorporated into the dynamic knowledge graph without disrupting the original causal event chain structure. Simultaneously, when loops occur, the relevant causal edges are adjusted to undirected edges, thus making the graph update results more stable and ensuring consistency between subsequent retrieval and inference paths.

[0172] In one possible implementation, if the matching result is a mismatch, step S204 may further include:

[0173] S2045. Train the small-scale incremental model corresponding to the target triplet, update the vector representation of the event chain associated with the target triplet, and merge the target triplet into the knowledge graph.

[0174] If the matching result is a mismatch, the entity and relation are first extracted from the target triple, and then mapped to a preset embedding space. Subsequently, based on the candidate causal event chain to which the target triple belongs, a small-scale incremental model is invoked to fine-tune the parameters of the local subgraph related to the target triple.

[0175] Specifically, during the fine-tuning process, the small-scale incremental model uses the contextual consistency between the target triplet and its adjacent triplets in the chain as a training signal to update the affected node representations, relation representations, and chain-level aggregation vectors in the causal event chain, so that the updated chain vectors can reflect the semantics and causal position of the newly added triplet.

[0176] After updating the vector representation, the target triple and its associated causal event chain are re-encoded, and the new chain representation is written into the storage structure of the dynamic knowledge graph, so that the target triple forms a new node association or edge association in the dynamic knowledge graph.

[0177] If the causal event chain corresponding to the target triple was originally stored only as a historical vector index, then the target triple and the dynamic knowledge graph can be merged by replacing the old chain vector and supplementing the local adjacency relationship of the target triple.

[0178] If the causal event chain corresponding to the target triple already has a partial structure, the original index and retrieval weights are adjusted synchronously based on the updated chain vector, thereby completing the incremental expansion of the graph content.

[0179] After adopting the above processing method, the dynamic knowledge graph can partially absorb new operation and maintenance knowledge without retraining the entire model. The representation of the causal event chain associated with the target triple will also change synchronously with the newly added causal information, so that the operation and maintenance relationships in the graph are continuously updated, which facilitates subsequent semantic retrieval and fault reasoning based on chain vectors. Since incremental training is only performed on local links, the system can still maintain low computational overhead in the resource-constrained BMC environment, and enable new triples to be incorporated into the existing dynamic knowledge graph more quickly.

[0180] Optionally, a small-scale incremental model corresponding to the target triple is trained, and the vector representation of the event chain associated with the target triple is updated, including: selecting K historical event nodes in the dynamic knowledge graph that have the same resource type as the event nodes in the target triple as an anchor set, where K is an integer greater than 2; using the anchor set as a constraint, and keeping the node vector representation in the anchor set unchanged, only the node vector represented by the target triple is embedded and trained to obtain the vector representation of the target triple; and the vector representation is merged into the dynamic knowledge graph.

[0181] First, historical event nodes with resource types consistent with the event nodes in the target triple are retrieved from the dynamic knowledge graph. K nodes are then selected to form an anchor set based on chronological order, position within the chain, or structural similarity to the target triple. Subsequently, the node vectors corresponding to the anchor set are used as fixed parameters to input training constraints. During training, only the embedding optimization is performed on the node vectors corresponding to the target triple and their associated edge parameters; the vectors within the anchor set remain unchanged. A gradient update method based on negative sampling can be used during training to complete small-scale incremental learning within a limited number of rounds, outputting a vector representation consistent with the target triple. After training, the vector representation is written into the node table or edge table of the dynamic knowledge graph and associated with the original graph index, thus enabling the target triple to be incorporated into the existing causal event chain.

[0182] Anchor point sets provide stable references, ensuring spatial consistency between the new representation of the target triple and the existing historical structure. Small-scale incremental models update parameters only for locally added content, avoiding retraining the entire graph representation. After the vector representation is merged into the dynamic knowledge graph, the new representation can be directly called in subsequent retrieval, matching, and conflict determination to complete the association reasoning.

[0183] By adopting this approach, target triples can be incorporated into the dynamic knowledge graph in a local incremental manner. The graph update range is concentrated and controllable, which reduces the disturbance to the existing causal event chain representation and enables newly added operation and maintenance knowledge to be incorporated into the graph and retrieved more quickly in resource-constrained environments.

[0184] In one possible implementation, the management method for dynamic knowledge graphs further includes: in response to receiving a retrieval request sent by the query end, parsing the retrieval request to obtain query keywords or query vectors; searching for subgraphs related to the query keywords or query vectors in the dynamic knowledge graph through semantic indexing to obtain target subgraphs; converting the target subgraphs into question-and-answer contexts in the form of relational paths, and sending the question-and-answer contexts to the query end.

[0185] The query endpoint can be an operations and maintenance terminal, a console, or an upper-layer application that calls a question-and-answer service. The query keywords carried in the retrieval request directly express the device name, alarm type, log characteristics, or handling action. The query vector is obtained by vectorizing the natural language question using a pre-trained semantic encoding model.

[0186] Semantic indexes can be built on entity vectors, relation vectors, and subgraph vectors of dynamic knowledge graphs, and can map event nodes, attribute nodes, and relation edges in dynamic knowledge graphs to the same vector space, so as to retrieve local subgraphs related to the query content based on vector similarity, entity matching results, or path recall results.

[0187] A target subgraph typically contains causal event chain fragments, causal edges, and attribute descriptions that are highly relevant to the search intent corresponding to the search request. The target subgraph can be further pruned according to node type and relationship direction to maintain a smaller scope.

[0188] When converting the target subgraph into a relational path format, multi-hop relational chains can be extracted according to the connection order from the starting entity to the target entity in the subgraph. Simultaneously, entity names, relation names, and necessary attribute values ​​are organized into a linear text context. For cases with multiple candidate paths, several relational paths can be selected based on path length, semantic score, and causal event chain credibility to jointly constitute the question-answering context. The question-answering context can be directly output as structured text, or concatenated into hints adapted to large-scale model inference and sent to the query end for the query end's question-answering generation module to organize the answer.

[0189] This processing method allows retrieval requests to be first mapped into semantic expressions, then the semantic index locates the relevant subgraphs in the knowledge graph, and further converts the graph structure into readable relational path text. This outputs context suitable for question-answering reasoning while preserving the graph's relational information. Because the question-answering context originates from the real-time updated local structure of the dynamic knowledge graph, the content obtained by the query end remains consistent with the current operational status, and the path expressions clearly reflect the relationships between entities, facilitating subsequent fault analysis, cause tracing, and the generation of handling suggestions.

[0190] In one possible implementation, the management method for dynamic knowledge graphs further includes: during the response to a target maintenance event, applying an exclusive write lock to the partitions involved in the target triplet using a read-write lock, and opening a shared read lock to the remaining partitions; when updating the dynamic knowledge graph, synchronously updating the version number of the dynamic knowledge graph; and during the response to receiving a retrieval request, allocating a shared read lock using a read-write lock to enable concurrent retrieval operations of the dynamic knowledge graph.

[0191] In the process of responding to the target maintenance event, firstly, the partition identifier corresponding to the target triple is resolved, and then an exclusive write lock is requested on the partition corresponding to the target triple, so that the update processing only applies to the affected data range. At the same time, a shared read lock is reserved on partitions not involved in the target triple to maintain the retrieval availability of other triples.

[0192] After writing, replacing, or deleting the target triple, the version number of the dynamic knowledge graph is synchronously incremented or reset, and the new version number is written into the graph metadata so that subsequent searches can identify the latest knowledge state.

[0193] When a retrieval request is received, the required partitions are located based on the query keywords, query vectors, or path constraints in the retrieval request, and shared read locks are assigned to these partitions, thereby allowing multiple retrieval requests to concurrently access the same dynamic knowledge graph without conflicting with update operations.

[0194] This management approach, by limiting the update scope to the relevant partitions of the target triple and synchronously refreshing the version number after the update is committed, combined with shared read lock control during the retrieval phase, enables graph updates and queries to be executed collaboratively in the same runtime environment. Therefore, the dynamic knowledge graph, while maintaining local update consistency, can still support multiple retrieval tasks reading in parallel, thereby improving the real-time performance and access stability of the operational knowledge service.

[0195] This application provides a BMC operation and maintenance method based on a dynamic knowledge graph, including: responding to a hardware alarm signal detected by the BMC, triggering a target operation and maintenance event corresponding to the hardware alarm signal in the dynamic knowledge graph, and obtaining an updated dynamic knowledge graph, wherein the dynamic knowledge graph is managed based on the above embodiment or any possible implementation of the above embodiment's dynamic knowledge graph management method; retrieving a target causal event chain associated with the hardware alarm signal in the updated dynamic knowledge graph; and outputting a repair strategy corresponding to the hardware alarm signal based on the target causal event chain.

[0196] When the BMC detects a hardware alarm signal, it responds to the hardware alarm signal by triggering the corresponding target operation and maintenance event in the knowledge graph, and calls the dynamic knowledge graph management method of the aforementioned embodiments to obtain the updated knowledge graph, so as to ensure that the graph contains the latest causal relationship information before diagnosis.

[0197] Subsequently, the updated knowledge graph retrieves the target causal event chain associated with the hardware alarm signal. This target causal event chain fully reveals the fault propagation path from the root cause of the alarm to the hardware alarm signal. Following this, based on the target causal event chain, corresponding remediation strategies are output, providing operations and maintenance personnel with accurate and interpretable fault handling suggestions. The entire process achieves closed-loop processing from alarm triggering, knowledge updating, causal tracing to strategy output, effectively improving the automation and intelligence level of BMC operations and maintenance.

[0198] Figure 4 This is a schematic diagram of the structure of the control device for dynamic knowledge graphs provided in the embodiments of this application, such as... Figure 4 As shown, the dynamic knowledge graph management device 40 provided in this embodiment includes:

[0199] The acquisition module 401 is used to acquire multiple candidate maintenance events within a preset time period in response to the triggering of the target maintenance event;

[0200] Processing module 402 is used to find antecedent events that are causally related to the target operation and maintenance event among multiple candidate operation and maintenance events, and generate target triples corresponding to the target operation and maintenance event; based on the target triples, it performs matching with existing operation and maintenance triples and corresponding causal event chains in the dynamic knowledge graph to obtain matching results. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time sequence and the operation and maintenance triples corresponding to the operation and maintenance events; according to the matching results, it performs the corresponding dynamic knowledge graph update to obtain the updated dynamic knowledge graph.

[0201] In one possible implementation, the matching results include: no match, conflict exists, and no conflict exists. The processing module 402 is specifically used for:

[0202] If the knowledge graph does not have an operation and maintenance triple that matches the target triple, the matching result is no match;

[0203] If the dynamic knowledge graph has an operation and maintenance triple that matches the target triple, and the causal event chain to which the target triple and the operation and maintenance triple belong conflict, the matching result is determined to be conflicting.

[0204] If the dynamic knowledge graph has an existing operation and maintenance triple that matches the target triple, and there is no conflict in the causal event chain to which the target triple belongs and the existing operation and maintenance triple belongs, the matching result is determined to be no conflict.

[0205] In one possible implementation, if the matching result indicates a conflict, the processing module 402 is specifically used for:

[0206] Determine whether the target triplet and the operation and maintenance triplet belong to the same causal event chain;

[0207] If they belong to the same causal event chain, calculate the semantic similarity between the target triplet and the other operation and maintenance triplets in the causal event chain. If the semantic similarity is greater than the preset similarity threshold, replace the operation and maintenance triplet with the target triplet; otherwise, discard the target triplet.

[0208] If they do not belong to the same causal event chain, calculate the consistency score of the target triplet and the consistency score of the operation and maintenance triplet based on the causal event chain to which the target triplet belongs and the causal event chain to which the operation and maintenance triplet belongs. Keep the triplet corresponding to the one with the higher consistency score in the dynamic knowledge graph.

[0209] In one possible implementation, the processing module 402 is further configured to:

[0210] The target triplet and the operation and maintenance triplet are respectively used as the triplet to be tested;

[0211] The consistency score of the triplet to be tested is obtained by calculating the consistency between the triplet to be tested and the other triplets in the causal event chain to which the triplet to be tested belongs.

[0212] The consistency score of the triplet under test is obtained by weighting and multiplying the consistency scores based on the chain quality weights corresponding to the triplet under test. The chain quality weights are obtained by weighting the event credibility and chain length of the event chain to which the triplet under test belongs.

[0213] In one possible implementation, if the matching result is that there is no conflict, the processing module 402 is specifically used for:

[0214] Establish a branch node in the event chain to which the operation and maintenance triple belongs, add the target triple to the branch node, and obtain the updated dynamic knowledge graph; the branch node has the same upstream connection relationship as the existing operation and maintenance triple.

[0215] In one possible implementation, if the matching result is a mismatch, the processing module 402 is specifically used for:

[0216] The target triples are trained to correspond to small-scale incremental models, and the vector representations of the event chains associated with the target triples are updated to integrate the target triples into the knowledge graph.

[0217] In one possible implementation, the processing module 402 is further configured to:

[0218] In response to receiving a retrieval request from the querying end, the retrieval request is parsed to obtain the query keywords or query vector;

[0219] The target subgraph is obtained by searching for subgraphs related to query keywords or query vectors in the dynamic knowledge graph through semantic indexing.

[0220] The target subgraph is converted into a question-and-answer context in the form of a relational path, and the question-and-answer context is sent to the query end.

[0221] In one possible implementation, the processing module 402 is further configured to:

[0222] In the process of responding to the target operation and maintenance event, an exclusive write lock is acquired on the partition involved in the target triplet through read-write locks, and a shared read lock is opened on the other partitions;

[0223] When updating the dynamic knowledge graph, the version number of the dynamic knowledge graph is updated synchronously.

[0224] In response to a received retrieval request, a shared read lock is allocated using a read-write lock to enable concurrent retrieval operations on the dynamic knowledge graph.

[0225] The dynamic knowledge graph management device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0226] Figure 5 This is a schematic diagram of the structure of the BMC operation and maintenance device based on dynamic knowledge graph provided in the embodiments of this application, as shown below. Figure 5 As shown, the BMC operation and maintenance device 50 based on dynamic knowledge graph provided in this embodiment includes:

[0227] The acquisition module 501 is used to respond to the hardware alarm signal monitored by the BMC, trigger the target operation and maintenance event corresponding to the hardware alarm signal in the dynamic knowledge graph, and acquire the updated dynamic knowledge graph. The dynamic knowledge graph is managed based on the dynamic knowledge graph management method provided in the embodiments of this application.

[0228] The processing module 502 is used to retrieve the target causal event chain associated with the hardware alarm signal from the updated dynamic knowledge graph; and output the repair strategy corresponding to the hardware alarm signal based on the target causal event chain.

[0229] The BMC operation and maintenance device based on dynamic knowledge graph provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0230] Figure 6 This is a schematic diagram of the structure of the control device for the dynamic knowledge graph provided in an embodiment of this application. Figure 6 As shown, the dynamic knowledge graph management device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the dynamic knowledge graph management device 60 further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0231] In a specific implementation, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to perform the above-described method.

[0232] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0233] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0234] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0235] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0236] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0237] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0238] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0239] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0240] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0241] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0242] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0243] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0244] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0245] Finally, it should be noted that other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from its scope.

Claims

1. A method for managing dynamic knowledge graphs, characterized in that, include: In response to the triggering of a target maintenance event, obtain multiple candidate maintenance events within a preset time period; Among multiple candidate operation and maintenance events, find the antecedent events that have a causal relationship with the target operation and maintenance event, and generate the target triplet corresponding to the target operation and maintenance event; Based on the target triplet and the existing operation and maintenance triplets and the causal event chains corresponding to the operation and maintenance triplets in the dynamic knowledge graph, a matching result is obtained. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time sequence and the operation and maintenance triplets corresponding to the operation and maintenance events. Based on the matching results, the corresponding dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph.

2. The method for managing dynamic knowledge graphs according to claim 1, characterized in that, The matching results include: no match, conflict exists, and no conflict exists. The matching is based on the target triplet and existing operation and maintenance triplets and their corresponding causal event chains in the dynamic knowledge graph, yielding matching results including: If the knowledge graph does not have an operation and maintenance triple that matches the target triple, the matching result is a mismatch. If the dynamic knowledge graph contains an operation and maintenance triple that matches the target triple, and the causal event chain to which the target triple belongs conflicts with that of the operation and maintenance triple, then the matching result is determined to be conflicting. If the dynamic knowledge graph has an existing maintenance triple that matches the target triple, and the causal event chain to which the target triple belongs does not conflict with that of the existing maintenance triple, then the matching result is determined to be non-conflicting.

3. The method for managing dynamic knowledge graphs according to claim 2, characterized in that, If the matching result indicates a conflict, the step of updating the corresponding dynamic knowledge graph based on the matching result to obtain the updated dynamic knowledge graph includes: Determine whether the target triple and the operation and maintenance triple belong to the same causal event chain; If they belong to the same causal event chain, calculate the semantic similarity between the target triplet and the other operation and maintenance triplets in the causal event chain. If the semantic similarity is greater than a preset similarity threshold, replace the operation and maintenance triplet with the target triplet; otherwise, discard the target triplet. If they do not belong to the same causal event chain, based on the causal event chain to which the target triple belongs and the causal event chain to which the operation and maintenance triple belongs, calculate the consistency score of the target triple and the consistency score of the operation and maintenance triple, and retain the triple corresponding to the one with the higher consistency score in the dynamic knowledge graph.

4. The method for managing dynamic knowledge graphs according to claim 3, characterized in that, The process of calculating the consistency score of the target triplet and the consistency score of the operation and maintenance triplet based on the causal event chain to which the target triplet belongs and the causal event chain to which the operation and maintenance triplet belongs includes: The target triplet and the operation and maintenance triplet are sequentially used as the triplet to be tested; The consistency score of the triplet to be tested is obtained by calculating the consistency between the triplet to be tested and the other triplets in the causal event chain to which the triplet to be tested belongs. The consistency score is obtained by weighting and multiplying the consistency scores based on the chain quality weights corresponding to the triplet to be tested. The chain quality weights are obtained by weighting the event credibility and chain length of the event chain to which the triplet to be tested belongs.

5. The method for managing dynamic knowledge graphs according to claim 2, characterized in that, If the matching result is that there is no conflict, the step of updating the corresponding dynamic knowledge graph based on the matching result to obtain the updated dynamic knowledge graph includes: A branch node is established in the event chain to which the operation and maintenance triple belongs, and the target triple is added to the branch node to obtain an updated dynamic knowledge graph; the branch node has an upstream connection relationship consistent with the existing operation and maintenance triple.

6. The method for managing dynamic knowledge graphs according to claim 2, characterized in that, If the matching result is a mismatch, the step of updating the corresponding dynamic knowledge graph based on the matching result to obtain the updated dynamic knowledge graph includes: Train a small-scale incremental model corresponding to the target triple and update the vector representation of the event chain associated with the target triple to merge the target triple into the knowledge graph.

7. The method for managing dynamic knowledge graphs according to any one of claims 1 to 6, characterized in that, The management and control methods for the dynamic knowledge graph also include: In response to receiving a retrieval request from the querying end, the retrieval request is parsed to obtain the query keywords or query vector; The target subgraph is obtained by searching the subgraph related to the query keyword or the query vector in the dynamic knowledge graph using semantic indexing. The target subgraph is converted into a question-and-answer context in the form of a relational path, and the question-and-answer context is sent to the query end.

8. The method for managing dynamic knowledge graphs according to any one of claims 1 to 6, characterized in that, The management and control methods for the dynamic knowledge graph also include: During the response to the target operation and maintenance event, an exclusive write lock is acquired on the partition involved in the target triplet using a read-write lock, while a shared read lock is granted to the other partitions. When updating the dynamic knowledge graph, the version number of the dynamic knowledge graph is updated synchronously. In response to receiving a retrieval request, a shared read lock is allocated through the read-write lock to enable concurrent retrieval operations of the dynamic knowledge graph.

9. A BMC operation and maintenance method based on dynamic knowledge graph, characterized in that, include: In response to a hardware alarm signal detected by the BMC, a target operation and maintenance event corresponding to the hardware alarm signal is triggered within the dynamic knowledge graph, and an updated dynamic knowledge graph is obtained. The dynamic knowledge graph is managed based on the dynamic knowledge graph management method according to any one of claims 1 to 8. The target causal event chain associated with the hardware alarm signal is retrieved from the updated dynamic knowledge graph; Based on the target causal event chain, output the repair strategy corresponding to the hardware alarm signal.

10. A control device for a dynamic knowledge graph, characterized in that, include: The acquisition module is used to acquire multiple candidate operation and maintenance events within a preset time period in response to the triggering of the target operation and maintenance event; The processing module is used to find antecedent events that are causally related to the target operation and maintenance event among multiple candidate operation and maintenance events, and generate a target triplet corresponding to the target operation and maintenance event; Based on the target triplet and the existing operation and maintenance triplets and the causal event chains corresponding to the operation and maintenance triplets in the dynamic knowledge graph, a matching result is obtained. The dynamic knowledge graph stores multiple causal event chains constructed according to the time dimension. Each causal event chain corresponds to multiple operation and maintenance events that are causally related in time sequence and the operation and maintenance triplets corresponding to the operation and maintenance events. Based on the matching results, the corresponding dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph.

11. A BMC operation and maintenance device based on dynamic knowledge graph, characterized in that, include: The acquisition module is used to respond to the hardware alarm signal detected by the BMC, trigger the target operation and maintenance event corresponding to the hardware alarm signal in the dynamic knowledge graph, and acquire the updated dynamic knowledge graph, wherein the dynamic knowledge graph is managed based on the dynamic knowledge graph management method according to any one of claims 1 to 8; The processing module is used to retrieve the target causal event chain associated with the hardware alarm signal from the updated dynamic knowledge graph; and output the repair strategy corresponding to the hardware alarm signal based on the target causal event chain.

Images