Fault knowledge base archiving method, system, management device and cloud server

Abstract

The application provides a fault knowledge base archiving method, system, management device and cloud server, and relates to the technical field of knowledge base management. The fault knowledge base archiving method comprises the following steps: collecting a to-be-archived fault case; identifying whether the to-be-archived fault case belongs to an existing case based on the double comparison results of the to-be-archived fault case and each existing case in a preset fault knowledge base in the vector dimension and the semantic dimension; in response to the situation that the double comparison results jointly point to the to-be-archived fault case not belonging to the existing case, storing the to-be-archived fault case in the preset fault knowledge base in a first archiving state, and adding the fault case in the first archiving state to case search matching of the preset fault knowledge base in a limited participation mode; and based on the situation that the fault case in the first archiving state is cited by search matching within a preset observation period, adjusting the participation permission of the fault case in the preset fault knowledge base. The application can improve the search reliability of the fault knowledge base.

Description

Technical Field

[0001] This application relates to the field of knowledge base management technology, and in particular to a method, system, management device and cloud server for archiving a fault knowledge base. Background Technology

[0002] The automatic archiving mechanism of a fault knowledge base typically uses a single vector similarity threshold as the criterion. By converting the fault description text and maintenance instruction text of the fault case to be archived into a case semantic vector using a vector encoding model, the similarity score of this vector with the existing case vectors in the knowledge base is calculated. If the similarity score is lower than a preset threshold, it is determined as a new case and directly added to the database; if the similarity score reaches or exceeds the threshold, it is determined as an existing case and merged into the matching entry. Once a new case is added to the database, it is assigned the exact same retrieval participation weight as existing cases, without undergoing any quality verification process.

[0003] The above mechanism has two drawbacks. First, a single vector dimension cannot distinguish between fault cases that are "similar in appearance but different in content." Two cases may have highly similar fault descriptions in the vector space, but their fault attribution logic and repair operation sequence may differ fundamentally. Relying solely on the vector similarity threshold will lead to new cases being misclassified as existing cases or existing cases being repeatedly archived. Second, there is no quality verification period after new cases are added to the database. If the fault attribution analysis or repair procedure description of a newly archived case has deviations (e.g., omission of key spare parts replacement steps or incorrect specification of test point parameters in the repair guide), these deviations will continuously output low-quality repair suggestions with each subsequent search, continuously polluting the retrieval quality of the knowledge base.

[0004] Therefore, the existing fault knowledge base archiving mechanism lacks confidence level management and observation and verification process for newly archived cases. Low-quality cases directly participate in retrieval matching with equal weight without buffer verification, which is an urgent problem to be solved in the current fault knowledge base management technology field. Summary of the Invention

[0005] The main objective of this invention is to provide a method for archiving a fault knowledge base, which aims to achieve confidence-level verification of newly archived fault cases, allowing new cases to be added to the retrieval matching in a restricted participation manner during the observation and verification period, and then adjusting their participation permissions after quality assessment.

[0006] To achieve the above objectives, the present invention provides a fault knowledge base archiving method, the fault knowledge base archiving method comprising: Collect fault cases to be archived, which include fault description text and maintenance guidance text; Based on the comparison results of the fault cases to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions, it is determined whether the fault cases to be archived belong to existing cases. In response to the dual comparison results both pointing to the fact that the fault case to be archived does not belong to the existing cases, the fault case to be archived is stored in the preset fault knowledge base in the first archiving state, and the fault case in the first archiving state is added to the case retrieval matching of the preset fault knowledge base in a restricted participation manner. During the preset observation period, based on the retrieval and matching of fault cases in the first archived state, the participation permission of the fault case in the preset fault knowledge base is adjusted.

[0007] Optionally, identifying whether the fault case to be archived belongs to an existing case includes: Based on the vector distance distribution characteristics between the case semantic vector of the fault case to be archived and the vector of the initially screened hit case, the first confidence parameter is extracted; Based on the confidence score output by performing a semantic consistency comparison between the fault cases to be archived and existing cases using a large language model, a second confidence score parameter is extracted. A weighted fusion is performed on the first confidence parameter and the second confidence parameter to generate a fused confidence parameter; The fusion confidence parameter is compared with a preset fusion judgment threshold to generate the dual comparison result.

[0008] Optionally, the fault knowledge base archiving method further includes: Establish a verification-ready association between the fault cases to be archived and the existing cases matched by the comparison results of the vector dimension; The case semantic vector of the fault case to be archived is added to the set of case vectors of existing cases matched by the comparison results of the semantic dimension.

[0009] Optionally, the preset observation period is determined as follows: Based on the device model of the target device corresponding to the fault case to be archived, match the device reliability level in the preset device reliability level mapping table, and determine the first observation period based on the device reliability level. The fault type of the fault case to be archived is matched with the fault occurrence frequency level in the preset fault frequency mapping table, and the second observation duration is determined by the fault occurrence frequency level. The fault level of the fault case to be archived is matched with the maintenance urgency level in the preset urgency mapping table, and the third observation period is determined based on the maintenance urgency level. The maximum value among the first observation duration, the second observation duration, and the third observation duration is set as the preset observation period.

[0010] Optionally, adjusting the participation permission of the fault case in the preset fault knowledge base includes: Each time a fault case in the first archived state is retrieved and matched, the matching context information of that match is recorded. The matching context information includes the fault description text that triggered the current retrieval, the matching similarity score between the fault case and the fault description text that triggered the current retrieval, and the sorting position of the fault case in the recommended case sequence of the current retrieval. During the preset observation period, a matching quality score is generated based on the accumulated matching context information. The matching quality score reaching a preset matching quality threshold and the cumulative number of citations of the fault case reaching a preset citation threshold are used as trigger conditions to switch the fault case from the first archive state to the second archive state.

[0011] Optionally, the fault knowledge base archiving method further includes: For at least two fault cases in the second archive state that are jointly cited in the same recommended case set during the same retrieval and matching process in the preset fault knowledge base, a co-citation association edge is established between their case identifiers; The co-citation strength value of the co-citation associated edge is updated based on the cumulative number of times the fault case corresponding to the co-citation associated edge is co-cited in the subsequent retrieval and matching process; Using the case identifiers of each fault case in the second archive state in the preset fault knowledge base as vertices and the common reference association edges as edges, a common reference association topology graph is constructed.

[0012] Optionally, the method for comparing the vector dimensions is specifically as follows: Vector encoding is performed on the fault cases to be archived using at least one of residual network architecture, visual transformer architecture, and convolutional network architecture. The semantic dimension comparison method is as follows: At least one of rule engines, expert systems, and Bayesian networks is used to perform a semantic consistency comparison between the fault cases to be archived and existing cases.

[0013] Furthermore, to achieve the above objectives, the present invention also provides a fault knowledge base archiving system, the fault knowledge base archiving system comprising: The data acquisition module is used to acquire fault cases to be archived, which include fault description text and maintenance guidance text. The comparison module is used to identify whether the fault case to be archived belongs to the existing cases based on the dual comparison results of the fault case to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions. The storage module is configured to, in response to a situation where both comparison results indicate that the fault case to be archived does not belong to an existing case, store the fault case to be archived in the preset fault knowledge base in a first archived state, and the fault case in the first archived state is added to the case retrieval matching of the preset fault knowledge base in a restricted participation manner. An adjustment module is used to adjust the participation permission of a fault case in the preset fault knowledge base based on the retrieval and matching reference of a fault case in the first archived state during a preset observation period.

[0014] In addition, to achieve the above objectives, the present invention also provides a management device, the management device comprising: a memory, a processor, and a fault knowledge base archiving program stored in the memory and executable on the processor, the fault knowledge base archiving program being configured to implement the fault knowledge base archiving method as described above.

[0015] In addition, to achieve the above objectives, the present invention also provides a cloud server, including the management device described above.

[0016] This invention collects fault cases to be archived, each containing a fault description and repair instructions. Based on a dual comparison of the fault cases to be archived with existing cases in a preset fault knowledge base across both vector and semantic dimensions, it identifies whether the fault case to be archived belongs to the existing cases. Finally, in response to the dual comparison results indicating that the fault case to be archived does not belong to the existing cases, it is stored in the preset fault knowledge base in a first archive state. Fault cases in the first archive state are added to the case retrieval matching of the preset fault knowledge base with restricted participation. During a preset observation period, based on the retrieval and matching references of fault cases in the first archive state, the participation permissions of these fault cases in the preset fault knowledge base are adjusted. Thus, by generating dual comparison results through dual comparison of vector and semantic dimensions, and by gradually incorporating new cases into the knowledge base in a restricted first archive state based on these results, it achieves graded management of the confidence level of new case archiving quality. This controls the impact of retrieval anomalies of new cases during the verification period, thereby achieving dual assurance of knowledge base archiving quality and retrieval reliability, and improving the retrieval reliability of the fault knowledge base. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a schematic diagram of a fault knowledge base archiving method according to an embodiment of the present invention; Figure 2 for Figure 1 A schematic diagram of the fault knowledge base archiving method in step S200; Figure 3 A flowchart illustrating the method for establishing the relationships to be verified and supplementing the semantic vectors of case studies; Figure 4 for Figure 1 A schematic diagram of the fault knowledge base archiving method in step S400; Figure 5 This is a schematic diagram of a fault knowledge base archiving method according to another embodiment of the present invention; Figure 6 This is a schematic diagram of a fault knowledge base archiving method according to another embodiment of the present invention; Figure 7 This is a schematic diagram of a fault knowledge base archiving method according to another embodiment of the present invention; Figure 8 This is a structural block diagram of a fault knowledge base archiving system according to an embodiment of the present invention.

[0019] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Well-known modules, units, and their connections, links, communications, or operations are not shown or described in detail. Furthermore, the described features, architectures, or functions can be combined in any way in one or more embodiments. Those skilled in the art should understand that the various embodiments described below are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. It is also readily understood that the modules, units, or processing methods in the various embodiments shown herein and in the accompanying drawings can be combined and designed in various different configurations. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Because existing technologies use a single vector similarity threshold to determine the inclusion of new fault cases into the knowledge base in the fault repair scenario of blockchain servers and their accessories, they cannot distinguish between superficial similarity of vectors and the actual similarity of fault mechanisms, and lack confidence level classification and observation verification mechanisms for new archived cases.

[0022] The main solution of this application embodiment is: to perform a dual comparison of vector dimension and semantic dimension for each fault case to be archived, to identify whether it belongs to an existing case, to store new cases that do not belong to existing cases in the knowledge base in the first archiving state and to participate in subsequent retrieval in a restricted manner, and to adjust its participation permission based on the actual retrieval matching performance of the case within a preset observation period.

[0023] In this embodiment, for ease of description, the management device will be used as the execution subject in the following description.

[0024] This application provides a solution that generates dual comparison results through dual comparison of vector and semantic dimensions, and incorporates new cases into the knowledge base in stages with a restricted first archiving state based on the dual comparison results. This achieves confidence-level management of the archiving quality of new cases, controls the impact of retrieval anomalies of new cases during the verification period, and thus achieves dual protection of the archiving quality and retrieval reliability of the knowledge base. This significantly improves the data asset quality of the blockchain server fault knowledge base and enhances the retrieval reliability of the fault knowledge base.

[0025] To this end, the present invention proposes a fault knowledge base archiving method; it is understood that the cloud server is equipped with a management device for storing and executing the following method. The management device can be implemented using a main controller, such as an MCU (Micro Controller Unit), a DSP (Digital Signal Processor), an FPGA (Field Programmable Gate Array), or a SOC (System On Chip).

[0026] In existing technologies, when two fault cases are highly similar in vector space but have essential differences in fault attribution logic and maintenance operation sequence, relying solely on vector similarity cannot identify such "similar on the surface but different in substance" fault cases. Low-quality cases directly entered into the database will have the same retrieval participation weight as high-quality cases, causing continuous pollution to subsequent maintenance retrieval results.

[0027] Therefore, referring to Figure 1 In one embodiment of the present invention, the fault knowledge base archiving method includes steps S100-S400, wherein: S100. Collect fault cases to be archived. Fault cases to be archived include fault description text and maintenance guidance text. S200. Based on the dual comparison results of the fault case to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions, identify whether the fault case to be archived belongs to the existing cases. S300. In response to the situation where the dual comparison results both point to the fault case to be archived not belonging to the existing cases, the fault case to be archived is stored in the preset fault knowledge base in the first archived state. The fault case in the first archived state is added to the case retrieval and matching of the preset fault knowledge base in a restricted participation manner. S400. During the preset observation period, based on the search and matching of fault cases in the first archived state, adjust the participation permission of the fault case in the preset fault knowledge base.

[0028] Among them, fault description text refers to text data that records descriptive content such as fault phenomena, fault occurrence environment, and fault codes; maintenance guidance text refers to text data that records guidance content such as maintenance steps, operating procedures, and spare parts replacement details for the fault; vector dimension refers to the numerical vector space formed after feature vectorization of each fault case using a vector encoding model, within which the similarity between cases is represented by the distance between vectors; semantic dimension refers to the semantic space formed after semantic consistency analysis of fault description text and maintenance guidance text using a semantic comparison model, within which the substantial matching relationship between cases in fault attribution and maintenance steps is represented by semantic comparison logic. The system; the preset fault knowledge base refers to a set of fault cases that are pre-built and persistently stored. Each existing case in the set contains at least one case vector; the dual comparison result refers to the comprehensive conclusion formed by the fusion processing of the vector dimension comparison output and the semantic dimension comparison output, pointing to whether the case to be archived belongs to the existing cases; the first archiving state refers to the restricted archiving level of the new case during the preset observation period. Cases in this state are controlled in their participation in subsequent searches; the restricted participation method means that the new case does not participate in the case matching process with the same search weight as existing cases during the preset observation period; the preset observation period refers to the time interval from when the new case is stored in the first archiving state until its participation permission is adjusted.

[0029] Secondly, the management device receives fault cases to be archived uploaded from maintenance terminals or fault diagnosis equipment, and sends the fault description text and maintenance guidance text to the vector encoder and semantic comparison model, respectively. The vector dimension comparison process is as follows: the fault description text and maintenance guidance text are vector-encoded to generate a case semantic vector; the case semantic vector is then compared with each existing case vector in a preset fault knowledge base for vector similarity, and the similarity score between each existing case vector and the current case semantic vector is recorded. The semantic dimension comparison process is as follows: the fault description text and maintenance guidance text are input into the semantic comparison model along with the corresponding text of existing cases to obtain a semantic consistency comparison output. Subsequently, the vector dimension comparison results and the semantic dimension comparison results are fused to generate a dual comparison result. Based on this dual comparison result, it is determined whether the fault case to be archived matches a certain existing case at both the vector and semantic levels. In one optional implementation, the management device multiplies the similarity score between the fault case in the first archive state and the query vector by a preset attenuation coefficient of 0.5, and then combines it with the original similarity score of the fault case in the second archive state for descending sorting, or shifts the recommended sorting position of the fault case in the first archive state backward by K positions (e.g., K=5).

[0030] Finally, when the dual comparison results do not point to any existing cases in either dimension, the fault case to be archived is written into the preset fault knowledge base in the first archive state. The participation permission of this case is set to restricted participation mode, that is, in the subsequent retrieval and matching process, it will only be included in the candidate range with reduced weight or restricted conditions. At the end of the preset observation period, the frequency of the case being retrieved and matched during the observation period and the matching quality index are counted. According to the preset permission adjustment strategy, it is decided to promote it from the first archive state to the second archive state or keep it in the restricted state.

[0031] Through the above technical solution, this embodiment applies a dual comparison of vector and semantic dimensions to the fault cases to be archived, and includes new cases in the knowledge base with a first archiving state and restricted participation mode. During the preset observation period, the participation permissions are adjusted based on the retrieval matching and citation performance, so that the superficial similarity of vectors and the substantial similarity of fault mechanisms can be distinguished. After the quality of the new archived cases is controlled during the verification period, they can obtain the same retrieval participation weight as existing cases. This avoids the continuous interference of low-quality cases to subsequent retrieval after they are included in the knowledge base, and achieves quality assurance in the knowledge base archiving process.

[0032] This embodiment introduces a tiered archiving status and a preset observation period verification mechanism for newly archived cases. In existing technologies, under automatic archiving scenarios, once a case is identified as new, it is directly added to the database and given the same retrieval permissions as existing cases. This is achieved by using vector similarity as the sole criterion for new case determination—similarity below a threshold is considered new, while above the threshold is considered existing. A one-time database entry or update operation is then performed based on the determination result. However, this approach has limitations on two levels: First, at the discrimination level, single-dimensional vector similarity cannot handle case pairs that are "similar in vector space location but have different fault mechanisms," leading to false negatives or false positives in new case determination. Second, at the archiving level, the retrieval permissions of newly added cases are no different from those of high-quality existing cases, lacking a verification period. Once a judgment error occurs, incorrectly added cases will continue to affect the quality of subsequent retrieval results. Because the knowledge base archiving management in this field typically adopts the operational convention of "judgment as final, inclusion as participation, and participation as equality," technicians often tend to optimize the accuracy of the judgment algorithm itself when facing new case archiving quality issues. This could involve improving the vector encoding model or adjusting the similarity threshold, rather than introducing a hierarchical status and verification period after judgment to buffer the impact of judgment uncertainty. Compared with existing technologies, this embodiment first includes new cases in the knowledge base with a first archiving status and restricted participation, and then adjusts their participation permissions based on retrieval matching reference data within a preset observation period. This transforms the archiving operation from a "one-time judgment" into a hierarchical progressive process of "judgment-buffering-verification-grading."

[0033] This invention collects fault cases to be archived, each containing a fault description and repair instructions. Based on a dual comparison of the fault cases to be archived with existing cases in a preset fault knowledge base across both vector and semantic dimensions, it identifies whether the fault case to be archived belongs to the existing cases. Finally, in response to the dual comparison results indicating that the fault case to be archived does not belong to the existing cases, it is stored in the preset fault knowledge base in a first archive state. Fault cases in the first archive state are added to the case retrieval matching of the preset fault knowledge base with restricted participation. During a preset observation period, based on the retrieval and matching references of fault cases in the first archive state, the participation permissions of these fault cases in the preset fault knowledge base are adjusted. Thus, by generating dual comparison results through dual comparison of vector and semantic dimensions, and by gradually incorporating new cases into the knowledge base in a restricted first archive state based on these results, it achieves graded management of the confidence level of new case archiving quality. This controls the impact of retrieval anomalies of new cases during the verification period, thereby achieving dual assurance of knowledge base archiving quality and retrieval reliability, and improving the retrieval reliability of the fault knowledge base.

[0034] When vector similarity is in the critical range, relying solely on vector threshold judgment may lead to misclassification of cases with ambiguous confidence boundaries. Furthermore, the semantic comparison output of large language models itself has uncertainty. How to quantify and integrate the uncertainty of the two dimensions into a reliable judgment basis directly affects the accuracy of new case recognition results.

[0035] Therefore, referring to Figure 2 Another embodiment of the present invention provides a fault knowledge base archiving method, based on the above. Figure 1 The illustrated embodiment identifies whether a fault case to be archived belongs to an existing case, including steps S210-S240, wherein: S210. Based on the vector distance distribution features between the case semantic vector of the fault case to be archived and the vector of the initially screened hit case, extract the first confidence parameter. S220. Based on the large language model, perform a semantic consistency comparison between the archived fault cases and existing cases to extract the second confidence parameter. S230. Perform weighted fusion on the first confidence parameter and the second confidence parameter to generate a fused confidence parameter; S240. Compare the fusion confidence parameter with the preset fusion judgment threshold to generate a dual comparison result.

[0036] Among them, the initial screening hit case vector refers to the existing case vector whose similarity score with the case semantic vector of the fault case to be archived meets the initial screening conditions in the vector dimension comparison. The initial screening conditions can be that the existing case vector ranks in the top K in similarity score or the similarity score exceeds the preset initial screening threshold; the vector distance distribution feature refers to the statistical distribution formed by the distance values ​​between the case semantic vector and each initial screening hit case vector, including the distance mean, distance variance, and distance minimum; the first confidence parameter refers to the quantitative index that represents the credibility of the vector dimension comparison result, calculated based on the vector distance distribution feature; the large language model refers to the language generation model that is pre-trained on a large-scale corpus and has the ability to understand semantics and judge consistency; the confidence score output by the semantic consistency comparison refers to The large language model assigns a certainty score when comparing the attribution logic and maintenance procedure sequence of two fault cases to determine their consistency. The second confidence parameter is a quantitative indicator extracted from the confidence score of the large language model, representing the credibility of the semantic dimension comparison results. Weighted fusion refers to the summation of the first and second confidence parameters after assigning them preset first and second weight coefficients, respectively. The fusion confidence parameter is the comprehensive confidence value obtained after weighted fusion. The preset fusion judgment threshold is the confidence threshold used to determine whether a fault case to be archived belongs to an existing case. The dual comparison result is the judgment conclusion generated after comparing the fusion confidence parameter with the preset fusion judgment threshold, indicating whether the fault case to be archived belongs to an existing case or not.

[0037] Secondly, after completing the vector encoding processing of the fault cases to be archived, the management device calculates the similarity between the case semantic vector and each existing case vector in the preset fault knowledge base. It then selects a set of preliminary screening hit case vectors from the existing case vectors whose similarity scores meet the initial screening criteria. The device calculates the vector distance between the case semantic vector and each preliminary screening hit case vector, and statistically analyzes the mean, variance, and minimum value of this distance to form a vector distance distribution feature. Based on the vector distance distribution feature and the preset confidence mapping function, a first confidence parameter is generated. For example, when the minimum distance is lower than the preset lower limit and the distance variance is lower than the preset variance threshold, the first confidence parameter takes a higher value, indicating that the vector dimension comparison result is more reliable; otherwise, a lower value is taken.

[0038] Subsequently, the fault description text and maintenance instruction text of the fault cases to be archived, as well as the fault description text and maintenance instruction text of the existing cases corresponding to the initial screening hit case vectors, are input into the big language model. Through the preset semantic consistency comparison prompt word template, the big language model is instructed to perform semantic analysis on the consistency of the two cases in fault attribution logic and maintenance operation step sequence. The big language model outputs the consistency judgment conclusion and the corresponding confidence score. The confidence score is extracted from the output of the big language model as the second confidence parameter. When the confidence score output by the big language model is in the form of a probability distribution, the confidence value corresponding to the maximum value is taken as the second confidence parameter.

[0039] Finally, the first confidence parameter is multiplied by a preset first weight coefficient, and the second confidence parameter is multiplied by a preset second weight coefficient. The sum of the two is used to generate a fusion confidence parameter. The fusion confidence parameter is compared with a preset fusion judgment threshold. When the fusion confidence parameter is lower than the preset fusion judgment threshold, a double comparison result indicating that the fault case to be archived does not belong to the existing cases is generated. When the fusion confidence parameter reaches or exceeds the preset fusion judgment threshold, a double comparison result indicating that the fault case to be archived belongs to the existing cases is generated.

[0040] Through the above technical solution, this embodiment quantifies the distance distribution features of the vector dimension into a first confidence parameter, quantifies the semantic consistency judgment confidence of the large language model into a second confidence parameter, and performs weighted fusion on the two to generate a fused confidence parameter, which is then compared with the judgment threshold. This allows the statistical confidence of the vector dimension and the semantic confidence of the semantic dimension to be jointly incorporated into the judgment process, thereby avoiding the one-sidedness of single-dimensional judgment and the uncertainty of critical interval judgment, and realizing the quantifiable and fused confidence of the dual comparison results.

[0041] This embodiment quantifies the comparison outputs of the vector dimension and semantic dimension into independent confidence parameters and then performs weighted fusion. Existing technologies, in scenarios combining vector comparison with semantic analysis from a large language model, typically employ a cascaded decision-making approach. This involves initial screening using vector similarity followed by a review using the semantic model, or vice versa. The outputs of the two dimensions are used as sequential filtering nodes, with the output of the previous node directly determining whether the next node is triggered. However, this approach has limitations on two levels: First, when the vector comparison results are in a critical range, the decision of the vector dimension in the cascaded structure has already been made, and the semantic dimension can only review cases judged as "pending confirmation" by the vector dimension, unable to compensate for misjudgments already made by the vector dimension. Second, the comparison confidence of the vector dimension and the semantic confidence output of the large language model, as two different sources of confidence, are independent in the cascaded structure and cannot collaboratively influence the final decision within a unified confidence framework. Because the technical convention in this field of fusion vector alignment and semantic alignment typically employs "cascaded filtering and sequential decision-making," it is often impossible to conceive of extracting the distance distribution statistical features of the vector dimension and the decision confidence level of the semantic dimension as a first confidence parameter and a second confidence parameter, respectively, and then generating a fused confidence parameter within a unified confidence space through weighted fusion, thereby driving the generation of dual alignment results. Compared with existing technologies, this embodiment uses a comparison mechanism that extracts the first confidence parameter from the vector distance distribution features and the second confidence parameter from the large language model confidence level, and then performs weighted fusion. This allows the uncertainty of the two dimensions to be quantified and integrated within the same confidence framework, rather than through sequential filtering.

[0042] When the comparison results between the vector dimension and the semantic dimension differ, i.e., one dimension points to an existing case while the other points to a new case, Figure 1 The embodiments shown do not provide differentiated handling strategies for this edge case, and simply categorizing it into any of the conclusions may lead to incorrect filing.

[0043] Therefore, referring to Figure 3 Another embodiment of the present invention provides a method for archiving a fault knowledge base, based on the above. Figure 1 The illustrated embodiment of the fault knowledge base archiving method further includes steps S500-S600, wherein: S500: Establish a verifiable association between the fault cases to be archived and the existing cases matched by the comparison results of the vector dimension; S600. Supplement the case semantic vector of the fault case to be archived to the case vector set of existing cases matched by the semantic dimension comparison results.

[0044] Among them, the relationship to be verified refers to the temporary association marker established between the failure case to be archived and the existing case, which represents that the two are highly similar in the vector dimension but has not yet been confirmed in the semantic dimension. This relationship needs to be further verified in the subsequent observation period. The case vector set refers to the set of multiple case semantic vectors associated with the same existing case. The same existing case may be associated with case semantic vectors from different versions or different sources at different times.

[0045] Secondly, when the comparison result of the vector dimension points to an existing case, but the comparison result of the semantic dimension points to a case that does not belong to an existing case, the management device establishes a verification relationship between the case identifier of the fault case to be archived and the case identifier of the existing case matched by the vector dimension, and persistently records the verification relationship and the comparison details of the vector dimension that triggered the relationship in the preset fault knowledge base.

[0046] Subsequently, when the comparison result in the vector dimension points to an existing case but the comparison result in the semantic dimension points to an existing case, the case semantic vector of the fault case to be archived is appended to the case vector set of the existing cases matched in the semantic dimension. This enriches the feature coverage of the existing case in the vector space, enabling the existing case to respond to more diverse fault description inputs in subsequent retrieval and matching.

[0047] Through the above technical solution, this embodiment performs different processing operations on the two edge cases where the vector dimension and semantic dimension diverge. When the vector matches but the semantics do not match, a relationship to be verified is established to preserve clues. When the semantics match but the vector does not match, a new vector is added to the existing case to expand its vector coverage. This ensures that the divergence results of the dual-channel comparison are effectively utilized rather than discarded, thereby avoiding information loss in the edge cases of single-channel comparison.

[0048] This embodiment utilizes the potential discrepancies arising from dual-channel comparison as the basis for triggering differentiated knowledge base update operations. Existing technologies typically directly accept consistent conclusions when vector and semantic comparison results align. For cases where the two are inconsistent, a conservative strategy is adopted, directly categorizing them into a manual review queue or classifying them as new cases. This often leads to the neglect of potential correlation clues identified in the vector dimension or vector coverage gaps identified in the semantic dimension. Because the field defaults to considering inconsistent comparison results as low reliability and abandoning automatic processing in automatic knowledge base archiving, it is often impossible to consider marking cases with vector matching but semantic mismatch as unverified associations and feeding the vector features of cases with semantic matching but vector mismatch into the vector set of existing cases. Compared to existing technologies, this embodiment uses the direction of discrepancies in the comparison results as the basis for differentiated processing. During vector dimension matching, unverified associations are established to retain tracking clues; during semantic dimension matching, new vectors are added to existing cases to expand the matching boundaries. This ensures that all outputs of the dual comparison, including discrepancies, generate corresponding knowledge base update operations.

[0049] If the observation period is too short, low-quality cases may have their permissions adjusted before they have accumulated enough search matching data for effective quality assessment. If the observation period is too long, high-quality new cases may not be able to obtain normal search participation weight in a timely manner, affecting the real-time response capability of the knowledge base.

[0050] Therefore, referring to Figure 4 Another embodiment of the present invention provides a method for archiving a fault knowledge base, based on the above. Figure 1 In the embodiment shown, the preset observation period is determined by steps S410-S440, wherein: S410. Match the device model of the target device corresponding to the fault case to be archived with the device reliability level in the preset device reliability level mapping table, and determine the first observation time based on the device reliability level. S420. Match the fault type of the fault case to be archived with the fault occurrence frequency level in the preset fault frequency mapping table, and determine the second observation duration based on the fault occurrence frequency level. S430. Match the fault level of the fault case to be archived with the maintenance urgency level in the preset urgency mapping table, and determine the third observation period based on the maintenance urgency level. S440. Set the maximum value among the first observation duration, the second observation duration, and the third observation duration as the preset observation period.

[0051] Among them, equipment reliability level refers to the equipment stability level comprehensively evaluated based on indicators such as the target equipment's historical fault-free operating time, mean time between failures, and maintenance response time; the preset equipment reliability level mapping table refers to the pre-built correspondence table between equipment models and equipment reliability levels; the first observation period refers to the observation period for new cases determined based on the equipment reliability level. The higher the equipment reliability level, the longer the first observation period. Fault cases of high-reliability equipment usually have a low occurrence frequency, requiring a longer observation period to accumulate sufficient matching reference data. Conversely, the observation period can be shortened for frequent fault cases of low-reliability equipment; the fault occurrence frequency level refers to the frequency level determined based on the historical occurrence frequency statistics of this fault type in the preset fault knowledge base; the preset fault... The fault frequency mapping table refers to a pre-built table that corresponds to fault types and fault frequency levels; the second observation period refers to the observation period for new cases determined based on the fault frequency level. The higher the frequency level, the shorter the second observation period, allowing new cases of high-frequency fault types to accumulate sufficient matching references for quality assessment within a short period; the maintenance urgency level refers to the urgency level determined based on the degree of impact of the fault on the operation of the target equipment; the preset urgency mapping table refers to a pre-built table that corresponds to fault levels and maintenance urgency levels; the third observation period refers to the observation period for new cases determined based on the maintenance urgency level. The higher the urgency level, the shorter the third observation period, requiring maintenance cases of high-urgency faults to verify their quality as soon as possible before being put into use.

[0052] Secondly, the management device extracts the device model of the target device from the failure cases to be archived, performs a matching query in the preset device reliability level mapping table using the device model as an index, and obtains the device reliability level corresponding to the device model; then, it queries the corresponding first observation time in the preset reliability-duration mapping table using the device reliability level as an index, where high reliability devices are mapped to a longer observation time and low reliability devices are mapped to a shorter observation time.

[0053] Subsequently, using the fault type and fault level as inputs in the same manner, the fault occurrence frequency level is matched in the preset fault frequency mapping table and the second observation time is queried; the maintenance urgency level is matched in the preset urgency mapping table and the third observation time is queried. The high frequency fault type and the high urgency fault level both correspond to a shorter observation time.

[0054] Finally, the first observation duration, the second observation duration, and the third observation duration are compared, and the maximum value among the three is set as the preset observation period for the fault case to be archived, ensuring that the length of the observation period is the most conservative value among the three dimensions, that is, meeting the most stringent observation duration requirements among all dimensions.

[0055] Through the above technical solution, this embodiment maps the three dimensions of equipment model, fault type and fault level to the corresponding reliability level, frequency level and urgency level respectively, and converts each into an observation period and takes the maximum value to set the preset observation period. This makes the observation period no longer a fixed parameter, but dynamically adapted to the equipment attributes, fault characteristics and urgency of different cases. This allows low-frequency fault types of high-reliability equipment to obtain a sufficiently long verification period to ensure the sufficiency of the evaluation, and high-frequency emergency fault types to obtain a shorter verification period for rapid deployment.

[0056] This embodiment employs a dynamic observation period setting method that independently maps three dimensions—equipment reliability, failure frequency, and maintenance urgency—to observation durations and then takes the maximum value of each. Existing technologies typically use fixed durations or single-dimensional adjustment strategies for setting observation or review periods for knowledge base entries. For example, they might set a uniform number of observation days based on the knowledge base administrator's experience, or set differentiated observation periods solely based on the credibility of the case source. This approach involves linearly adjusting the observation duration from a single dimension. However, this single-dimensional strategy cannot simultaneously accommodate the differentiated requirements for observation period durations from three independent dimensions: inherent equipment attributes, failure statistical characteristics, and business urgency. Because the technical convention in knowledge base management typically uses fixed values ​​or single-factor adjustments for setting verification periods, it is often impossible to conceive of a dynamic setting method that converts each of the three dimensions—equipment reliability, failure frequency, and maintenance urgency—into an independent mapping table, converting them into observation durations, and then taking the maximum value of each. Compared with the prior art, this embodiment uses a multi-dimensional dynamic observation period setting mechanism that maps three independent paths—device model, reliability level, first observation duration, fault type, frequency level, second observation duration, fault level, urgency level, and third observation duration—and takes the maximum value. This allows the observation period to simultaneously adapt to three independent dimensions: device attributes, fault statistical characteristics, and business urgency.

[0057] Simply counting the number of citations cannot distinguish between the two different situations of "being searched frequently but with low matching quality" and "being searched moderately and with high matching quality". Incomplete evaluation dimensions may lead to inaccurate decision-making on permission adjustment.

[0058] Therefore, referring to Figure 5 The present invention also provides a method for archiving a fault knowledge base, based on the above. Figure 1 The embodiment shown adjusts the participation permissions of the fault case in the preset fault knowledge base, including steps S450-S460, wherein: S450. Each time a fault case in the first archived state is retrieved and matched, the matching context information of the current match is recorded. The matching context information includes the fault description text that triggered the current retrieval, the matching similarity score between the fault case and the fault description text that triggered the current retrieval, and the sorting position of the fault case in the recommended case sequence of the current retrieval. S460. During the preset observation period, a matching quality score is generated based on the accumulated matching context information. The matching quality score reaching the preset matching quality threshold and the cumulative number of citations of the fault case reaching the preset citation threshold are used as trigger conditions to switch the fault case from the first archive state to the second archive state.

[0059] Among them, the matching context information refers to the retrieval environment data associated with the event of a fault case in the first archive state being matched during a retrieval matching process. This includes the content of the fault description text that triggered the current retrieval, used to trace the original text of the retrieval scenario in which the case was matched. The matching similarity score is used to measure the vector similarity between the case and the retrieval input, as well as the ranking position in the recommended case sequence. It is used to characterize the relative recommendation priority of the case among multiple candidate cases in the current retrieval. The matching quality score is a comprehensive quality evaluation index calculated by a preset scoring algorithm based on all the matching context information accumulated within the preset observation period. This index comprehensively reflects the frequency of matching, the average level of matching similarity, and the recommendation ranking distribution characteristics of the case in each retrieval during the observation period.

[0060] Secondly, each time a fault case in the first archive state responds to a retrieval request and is included in the retrieval matching results, the management device records the fault description text that triggered the retrieval, records the cosine similarity or Euclidean distance value between the case semantic vector of the fault case and the retrieval vector of the fault description text that triggered the retrieval as the matching similarity score, and records the sequence number position of the fault case in the recommended case sequence of all candidate cases in the current retrieval, which is arranged in descending order of similarity.

[0061] Subsequently, during the preset observation period, each matching event is accumulated and recorded in the above manner, forming a sequence of matching context information for the fault case within the preset observation period. At the end of the preset observation period, a matching quality score is performed on the accumulated matching context information sequence: the matching similarity score and ranking position of each match are obtained one by one, and a preset weighted algorithm is used. For example, higher weights are assigned to matches within the first K positions and lower weights are assigned to matches after the Kth position. The weighted average score is then calculated as the matching quality score.

[0062] Finally, the matching quality score is compared with the preset matching quality threshold, and the cumulative number of citations of the fault case within the preset observation period is compared with the preset citation threshold. When both conditions are met, the archiving status of the fault case is switched from the first archiving status to the second archiving status. Fault cases in the second archiving status are added to the case retrieval matching of the preset fault knowledge base with the same retrieval participation weight as existing cases.

[0063] Through the above technical solution, this embodiment records matching context information sequentially during the observation period. The matching context information includes the original text that triggers the retrieval, the matching similarity score, and the recommended ranking position. Based on the accumulated matching context information, a matching quality score is generated. The matching quality score and the cumulative number of citations are simultaneously satisfied as the state switching trigger condition. This makes the decision to switch from the first archive state to the second archive state dependent on the evaluation results of both matching quality and matching frequency. This avoids the problem of low-quality, high-frequency matching cases being mistakenly promoted or high-quality, low-frequency matching cases being mistakenly retained based solely on the number of citations.

[0064] This embodiment triggers the switch from the first archived state to the second archived state using a dual condition of matching quality score and cumulative citation count. In existing knowledge base entry lifecycle management, state upgrades or downgrades typically employ a single-dimensional trigger condition, such as automatic upgrade based solely on a citation count threshold or manual review. This approach uses a single-index threshold. However, low-quality cases in the first archived state may be frequently searched but not actually matched due to keywords in the fault description text similar to high-frequency search requests. A single citation count threshold cannot distinguish between "true high-frequency matching" and "superficial high-frequency matching," leading to a disconnect between permission adjustment decisions and actual case quality. Because the common practice in knowledge base entry permission management is "upgrade upon reaching a citation count threshold," it is often overlooked to incorporate the contextual information of each match, including matching similarity scores and recommended ranking positions, into the quality assessment to form a matching quality score, which, together with the cumulative citation count, constitutes a dual trigger condition for state switching. Compared with the prior art, this embodiment accumulates matching quality score and recommendation ranking position matching context information by recording matching similarity score and recommendation ranking position one by one to generate a permission adjustment mechanism with the cumulative number of references as a dual trigger condition, so that the state switching decision is based on the quantitative evaluation of matching quality.

[0065] As the number of fault cases in the second archived state continues to grow, there may be implicit relationships between the cases during the retrieval and matching process. For example, two different fault cases may be frequently hit at the same time in the same retrieval scenario. This co-occurrence relationship itself is a kind of related information in the knowledge base that has not yet been structured and utilized. There is a lack of means to discover and model such relationships.

[0066] Therefore, referring to Figure 6 Another embodiment of the present invention provides a fault knowledge base archiving method, based on the above. Figure 5 The illustrated embodiment of the fault knowledge base archiving method further includes steps S700-S900, wherein: S700. For at least two fault cases in the second archive state that are jointly cited in the same recommended case set during the same retrieval and matching process in the preset fault knowledge base, establish a co-citation association edge between their case identifiers. S800. Based on the cumulative number of times the fault cases corresponding to the co-referenced edges are co-referenced in the subsequent retrieval and matching process, update the co-reference strength value of the co-referenced edges. S900. Using the case identifiers of each fault case in the second archive state in the preset fault knowledge base as vertices and the common reference association edges as edges, construct a common reference association topology graph.

[0067] Here, a co-citation association edge refers to an undirected association edge connecting two faulty cases in the second archive state that are simultaneously included in the recommended case set during the same retrieval matching process. This association edge represents the co-occurrence relationship between the two faulty cases in the retrieval matching dimension. The recommended case set refers to the sequence of candidate cases sorted by matching similarity after case retrieval matching for a retrieval request. The co-citation strength value is a quantitative representation of the cumulative number of times a faulty case associated with a co-citation association edge is co-cited. The higher the co-citation strength value, the higher the frequency of the two faulty cases being hit simultaneously in the actual retrieval. The co-citation association topology graph refers to an undirected graph structure with each faulty case in the second archive state as a vertex and each co-citation association edge as an edge. This topology graph reflects the co-citation relationship network among all officially archived cases in the knowledge base.

[0068] Secondly, when performing case retrieval and matching each time, the management device obtains the recommended case sequence output for that retrieval, traverses the fault cases in the second archive state in the recommended case sequence, forms a pair for any two different fault cases in the second archive state, and checks whether there is a co-referenced association edge between the case identifiers of the two cases: if there is no co-referenced association edge, a new co-referenced association edge is established and the initial co-reference strength value of the edge is set to one; if there is already one, the co-reference strength value of the edge is incremented by one.

[0069] Subsequently, in the subsequent retrieval and matching process, each retrieval performs the establishment or accumulation operation on the co-citation association edges between the fault case pairs that are in the second archive state according to the above method, so that the co-citation strength value of each co-citation association edge is dynamically updated as the retrieval and matching activities accumulate, reflecting the actual co-citation frequency between each fault case pair.

[0070] Finally, using the case identifiers of each fault case in the second archived state in the preset fault knowledge base as graph vertices and each co-referenced edge as graph edges, the co-reference strength value of each edge is used as the weight attribute of that edge to construct a co-referenced association topology graph and persist it in the preset fault knowledge base. This co-referenced association topology graph can be used as a reference by the retrieval module when performing case retrieval matching. For example, when a user's search hits a fault case in the second archived state, the high co-reference strength adjacent cases of that case can be recommended as extended references based on the co-referenced association topology graph.

[0071] Through the above technical solution, this embodiment establishes co-citation association edges for fault cases in the second archive state that are jointly cited in the same retrieval process, and updates the co-citation strength value based on the cumulative number of subsequent co-citations to construct a co-citation association topology graph. This enables the originally isolated fault cases in the knowledge base to automatically form a co-citation relationship network through actual retrieval behavior. This network can serve as a structured basis for recommending related cases in subsequent retrievals, thereby reversibly structuring the retrieval behavior data into related topology assets within the knowledge base.

[0072] This embodiment uses the actual retrieval behavior of fault cases being co-cited during the retrieval matching process as the trigger event. It establishes co-citation association edges between case identifiers and updates the co-citation strength value based on the cumulative number of co-citations to construct a co-citation association topology graph. Existing technologies typically employ manual annotation based on case attributes to construct case association relationships in fault knowledge bases. For example, administrators manually establish association links between cases based on fault type or equipment model. This approach relies on static relationship definitions based on domain expert knowledge. However, static annotation methods cannot reflect the frequency and patterns of combined citations of each case in actual retrieval use and require continuous manual maintenance. The contradiction between maintenance costs and relationship coverage becomes particularly prominent as the knowledge base grows. Co-citation analysis methods have been used in bibliometrics to reveal the research topic associations between documents. However, in the field of fault knowledge base management, because the associations between knowledge base cases are usually defined manually based on technical attributes, the co-citation data generated by retrieval behavior is not considered a structured relational asset. This often leads to the inability to consider transforming co-citation events during the retrieval matching process into quantifiable association edges between cases and constructing a topology structure. Compared with existing technologies, this embodiment transforms the knowledge base case relationship network from static manual annotation to dynamic retrieval behavior-driven automatic construction by defining cases that jointly hit the same set of recommended cases in the same search as co-citation case pairs, establishing co-citation association edges, and constructing a co-citation association topology graph by accumulating co-citation strength values.

[0073] Different model architectures and judgment categories have varying comparison accuracy and computational efficiency in different fault case scenarios. The lack of specific implementation path constraints may lead to insufficient feasibility of technical solutions in specific scenarios.

[0074] Therefore, referring to Figure 7 Another embodiment of the present invention provides a method for archiving a fault knowledge base, based on the above. Figure 1 The embodiment shown describes a vector dimension comparison method, specifically step S1000, where: S1000 employs at least one of residual network architecture, visual transformer architecture, and convolutional network architecture to perform vector encoding on archived fault cases.

[0075] The semantic dimension comparison method is specifically step S1100, where: S1100 employs at least one of rule engines, expert systems, and Bayesian networks to perform semantic consistency comparisons between archived fault cases and existing cases.

[0076] Among them, the residual network architecture refers to a convolutional neural network architecture that achieves stable training of deep networks by introducing cross-layer identity mapping connections. Its residual connection mechanism allows gradients to be directly backpropagated across multiple intermediate layers, preserving multi-scale semantic features in fault text feature extraction. The visual transformer architecture refers to a transformer network architecture that incorporates the global dependencies between positions in the input sequence into the modeling based on a self-attention mechanism. It can capture long-distance contextual associations when processing long text fault descriptions. The convolutional network architecture refers to a network architecture that extracts local features by stacking convolutional layers and pooling layers. It is effective in detecting local key patterns in fault descriptions. It possesses rapid identification capabilities; vector encoding refers to the fixed-dimensional numerical vector output at the fully connected layer at the end of the network after processing the fault description text and maintenance instruction text through the aforementioned network architecture; rule engine refers to a judgment system that performs structured condition matching on two input fault case texts based on preset if-then logical rules; expert system refers to a judgment system that performs logical reasoning based on a pre-built fault domain knowledge rule base and inference engine; Bayesian network refers to a probabilistic graphical model that uses a directed acyclic graph to represent the probabilistic dependencies between fault attributes, and calculates the posterior probability of semantic consistency between two fault cases through Bayesian inference.

[0077] Secondly, after receiving the fault cases to be archived, the management device selects at least one model based on a preset model selection strategy. For example, it selects a convolutional network architecture for small texts to improve encoding speed, a visual transformer architecture for large texts to capture long-distance semantic dependencies, or a residual network architecture to extract image features from image-based fault attachments. The fault description text and maintenance guidance text of the fault cases to be archived are input into the selected network architecture for forward inference, and the output vector of the fully connected layer at the end of the network is taken as the case semantic vector of the fault case to be archived. In one optional implementation, a convolutional network architecture is used when the character length of the fault description text is less than 512 characters, and a visual transformer architecture is used when the character length exceeds 2048 characters. For the semantic comparison model, a specific scenario example is provided: for equipment types with standardized fault code systems (such as OBD-II, IEC 61850), a rule engine is used to perform exact matching; for complex faults lacking standardized fault codes, a Bayesian network is used for probabilistic inference.

[0078] Subsequently, in the semantic dimension comparison stage, a strategy is selected based on the preset semantic comparison model. For example, for fault types with clear fault codes and structured diagnostic processes, a rule engine is used for precise matching; for fault types that require comprehensive inference of causes from multiple fault phenomena, an expert system is used for reasoning; and for fault attribution with probability uncertainty, a Bayesian network is used for probabilistic inference. At least one of the rule engine, expert system, and Bayesian network is selected. The fault description text and maintenance guidance text of the fault case to be archived are input into the selected model and the corresponding text of the existing case to perform semantic consistency comparison, and the semantic dimension consistency comparison output is obtained.

[0079] Through the above technical solutions, this embodiment provides three optional encoding model implementation paths for vector dimension comparison: residual network architecture, visual transformer architecture, and convolutional network architecture; and three optional comparison model implementation paths for semantic dimension comparison: rule engine, expert system, and Bayesian network. This allows each of the two dimensions of the dual comparison to have multiple replaceable specific implementation schemes, thereby ensuring the feasibility and applicability of the technical solution under different fault case characteristics and data conditions.

[0080] This embodiment provides multiple optional specific implementation model categories for both vector dimension comparison and semantic dimension comparison. Existing technologies in automatic fault knowledge base archiving scenarios typically employ a single, fixed model for vector encoding and semantic comparison, such as using a general text encoder for vectorization and a rule engine for semantic matching by default. This approach covers all fault case types with a single implementation method. However, different fault cases vary significantly in text length, structuring level, and attribution complexity, making it difficult for a single model implementation path to maintain both comparison accuracy and computational efficiency in all scenarios. Since the art typically tends to use a uniform default implementation in model selection for knowledge base archiving systems to reduce system complexity, it is often not considered to provide multiple optional model categories as alternative paths for vector encoding and semantic comparison in the claims. Compared to existing technologies, this embodiment provides a multi-path implementation scheme by offering three alternative architectures for vector encoding (residual network architecture, visual transformer architecture, and convolutional network architecture) and three alternative decision model categories for semantic comparison (rule engine, expert system, and Bayesian network).

[0081] Reference Figure 8 This invention also proposes a fault knowledge base archiving system, which includes a data acquisition module, a comparison module, a storage module, and an adjustment module, wherein: The acquisition module is used to acquire fault cases to be archived, which include fault description text and maintenance guidance text. The comparison module is used to identify whether the fault case to be archived belongs to the existing cases based on the dual comparison results of the fault case to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions. The storage module is used to store the fault case to be archived in the preset fault knowledge base in the first archive state when the dual comparison results both indicate that the fault case to be archived does not belong to the existing cases. Fault cases in the first archive state are added to the case retrieval and matching of the preset fault knowledge base in a restricted participation mode. The adjustment module is used to adjust the participation permission of the fault case in the preset fault knowledge base within a preset observation period based on the retrieval and matching reference of the fault case in the first archive state.

[0082] It is worth noting that, since the fault knowledge base archiving system of the present invention is based on the above-mentioned fault knowledge base archiving method, the embodiments of the fault knowledge base archiving system of the present invention include all the technical solutions of all embodiments of the above-mentioned fault knowledge base archiving method, and the technical effects achieved are exactly the same, which will not be repeated here.

[0083] The present invention also proposes a management device, which includes: a memory, a processor, and a fault knowledge base archiving program stored in the memory and executable on the processor, the fault knowledge base archiving program being configured to implement the fault knowledge base archiving method described above.

[0084] It is worth noting that since the management device of the present invention is based on the above-mentioned fault knowledge base archiving method, the embodiments of the management device of the present invention include all the technical solutions of all embodiments of the above-mentioned fault knowledge base archiving method, and the technical effects achieved are exactly the same, so they will not be repeated here.

[0085] The present invention also proposes a cloud server, which includes a management device as described in the above embodiments.

[0086] It is worth noting that since the cloud server of the present invention is based on the above-mentioned management device, the embodiments of the cloud server of the present invention include all the technical solutions of all the embodiments of the above-mentioned management device, and the technical effects achieved are exactly the same, which will not be repeated here.

[0087] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for archiving a fault knowledge base, characterized in that, The fault knowledge base archiving method includes: Collect fault cases to be archived, which include fault description text and maintenance guidance text; Based on the comparison results of the fault cases to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions, it is determined whether the fault cases to be archived belong to existing cases. In response to the dual comparison results both pointing to the fact that the fault case to be archived does not belong to the existing cases, the fault case to be archived is stored in the preset fault knowledge base in the first archiving state, and the fault case in the first archiving state is added to the case retrieval matching of the preset fault knowledge base in a restricted participation manner. During the preset observation period, based on the retrieval and matching of fault cases in the first archived state, the participation permission of the fault case in the preset fault knowledge base is adjusted.

2. The fault knowledge base archiving method as described in claim 1, characterized in that, The process of identifying whether the fault case to be archived belongs to an existing case includes: Based on the vector distance distribution characteristics between the case semantic vector of the fault case to be archived and the vector of the initially screened hit case, the first confidence parameter is extracted; Based on the confidence score output by performing a semantic consistency comparison between the fault cases to be archived and existing cases using a large language model, a second confidence score parameter is extracted. A weighted fusion is performed on the first confidence parameter and the second confidence parameter to generate a fused confidence parameter; The fusion confidence parameter is compared with a preset fusion judgment threshold to generate the dual comparison result.

3. The fault knowledge base archiving method as described in claim 1, characterized in that, The fault knowledge base archiving method also includes: Establish a verification-ready association between the fault cases to be archived and the existing cases matched by the comparison results of the vector dimension; The case semantic vector of the fault case to be archived is added to the set of case vectors of existing cases matched by the comparison results of the semantic dimension.

4. The fault knowledge base archiving method as described in claim 1, characterized in that, The method for determining the preset observation period is as follows: Based on the device model of the target device corresponding to the fault case to be archived, match the device reliability level in the preset device reliability level mapping table, and determine the first observation period based on the device reliability level. The fault type of the fault case to be archived is matched with the fault occurrence frequency level in the preset fault frequency mapping table, and the second observation duration is determined by the fault occurrence frequency level. The fault level of the fault case to be archived is matched with the maintenance urgency level in the preset urgency mapping table, and the third observation period is determined based on the maintenance urgency level. The maximum value among the first observation duration, the second observation duration, and the third observation duration is set as the preset observation period.

5. The fault knowledge base archiving method as described in claim 1, characterized in that, Adjusting the participation permissions of the fault case in the preset fault knowledge base includes: Each time a fault case in the first archived state is retrieved and matched, the matching context information of that match is recorded. The matching context information includes the fault description text that triggered the current retrieval, the matching similarity score between the fault case and the fault description text that triggered the current retrieval, and the sorting position of the fault case in the recommended case sequence of the current retrieval. During the preset observation period, a matching quality score is generated based on the accumulated matching context information. The matching quality score reaching a preset matching quality threshold and the cumulative number of citations of the fault case reaching a preset citation threshold are used as trigger conditions to switch the fault case from the first archive state to the second archive state.

6. The fault knowledge base archiving method as described in claim 5, characterized in that, The fault knowledge base archiving method also includes: For at least two fault cases in the second archive state that are jointly cited in the same recommended case set during the same retrieval and matching process in the preset fault knowledge base, a co-citation association edge is established between their case identifiers; The co-citation strength value of the co-citation associated edge is updated based on the cumulative number of times the fault case corresponding to the co-citation associated edge is co-cited in the subsequent retrieval and matching process; Using the case identifiers of each fault case in the second archive state in the preset fault knowledge base as vertices and the common reference association edges as edges, a common reference association topology graph is constructed.

7. The fault knowledge base archiving method as described in claim 1, characterized in that, The method for comparing the vector dimensions is as follows: Vector encoding is performed on the fault cases to be archived using at least one of residual network architecture, visual transformer architecture, and convolutional network architecture. The semantic dimension comparison method is as follows: At least one of rule engines, expert systems, and Bayesian networks is used to perform a semantic consistency comparison between the fault cases to be archived and existing cases.

8. A fault knowledge base archiving system, characterized in that, The fault knowledge base archiving system includes: The data acquisition module is used to acquire fault cases to be archived, which include fault description text and maintenance guidance text. The comparison module is used to identify whether the fault case to be archived belongs to the existing cases based on the dual comparison results of the fault case to be archived and each existing case in the preset fault knowledge base in both vector and semantic dimensions. The storage module is configured to, in response to a situation where both comparison results indicate that the fault case to be archived does not belong to an existing case, store the fault case to be archived in the preset fault knowledge base in a first archived state, and the fault case in the first archived state is added to the case retrieval matching of the preset fault knowledge base in a restricted participation manner. An adjustment module is used to adjust the participation permission of a fault case in the preset fault knowledge base based on the retrieval and matching reference of a fault case in the first archived state during a preset observation period.

9. A management device, characterized in that, The management device includes: a memory, a processor, and a fault knowledge base archiving program stored in the memory and executable on the processor, the fault knowledge base archiving program being configured to implement the fault knowledge base archiving method as described in any one of claims 1 to 7.

10. A cloud server, characterized in that, Includes the management device as described in claim 9.

Images