Power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion

Abstract

The invention relates to the technical field of power system fault protection, in particular to a power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion, and the method comprises the steps: carrying out the semantic extraction of multi-source heterogeneous data to construct a defect knowledge graph, breaking through the limitation of a data island, and achieving the full-dimensional correlation analysis of equipment defect features; mapping the power grid secondary equipment defect knowledge graph to a Bayesian network framework, and quantifying the prior probability and conditional probability of defect occurrence based on historical data prior knowledge; and finally, providing accurate fault mode positioning by means of defect sub-graph search of the knowledge graph, and completing defect propagation path prediction by means of probability reasoning of the Bayesian network, so that the method can comprehensively utilize multi-source heterogeneous data of the secondary equipment of the power grid, and the fault propagation path prediction accuracy is improved. The complex relation and probability influence between the defect phenomenon and multiple layers of reasons are accurately described, the defect recognition accuracy and response timeliness are remarkably improved, and support is provided for safe and stable operation of a power grid.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system fault protection, and particularly relates to a power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion. BACKGROUND

[0002] With the development of smart grid and the continuous expansion of power grid scale, power grid secondary equipment (such as relay protection devices, automatic monitoring systems, communication equipment, etc.) plays an increasingly important role in the power system. The normal operation of these devices is the key to ensure the safe, stable and efficient operation of the power system. However, power grid secondary equipment may have various defects during long-term operation, such as misoperation, communication failure, etc. These defects not only affect the normal operation of the power system, but also may induce major safety accidents, resulting in huge economic losses and social impact.

[0003] In the traditional defect identification system, artificial experience judgment depends on the professional knowledge and field experience of operation and maintenance personnel. This mode has the problems of strong subjectivity and low efficiency, especially when facing massive device data, manual analysis is difficult to cover all potential defect scenarios. The expert system based on rule base conducts reasoning through pre-set logical rules, which reduces human intervention to a certain extent, but the construction of its rule base highly depends on the prior knowledge of domain experts, and it is difficult to adapt to the update iteration of equipment or the emergence of new defects. When new intelligent relay protection devices are put into use, their fault modes may be significantly different from traditional devices, and the static rule base cannot be quickly expanded, resulting in deviation of the identification result. Other statistical analysis models mostly use linear regression or clustering algorithms, which can process structured data, but have weak fusion ability for heterogeneous data and cannot capture the deep correlation characteristics of device defects. In addition, the defect data of power grid secondary equipment is widely sourced, including device alarm information, inspection records, online monitoring data, maintenance work order texts, etc. These data have significant differences in structure, format and semantics. The current methods are difficult to fully mine the effective components of various types of information when processing data, which may adversely affect the accuracy of defect identification. In addition, the existing defect identification models have the problems of insufficient data utilization, limited real-time updating capability, etc. in real-time defect data processing and defect reason reasoning, and are difficult to accurately obtain the complex relationship and probability influence between defect phenomena and multi-layer reasons.

[0004] Therefore, there is an urgent need for a power grid secondary equipment defect identification scheme that can comprehensively utilize the multi-source heterogeneous data of power grid secondary equipment, accurately depict the complex relationship and probability influence between defect phenomena and multi-layer reasons, and improve the real-time and accuracy of defect identification, so as to improve the safe and stable operation level of power grid. SUMMARY

[0005] In view of the above prior art deficiencies, the present application provides a power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion, which solves the problem of insufficient real-time and accuracy of defect identification for power grid secondary equipment due to the difficulty in accurately obtaining the complex relationship between defect phenomena and multi-layer causes.

[0006] To solve the above technical problems, the present application adopts the following technical solutions:

[0007] A power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion, comprising the following steps:

[0008] S1, transforming the defect data of the power grid secondary equipment into a structured knowledge graph, and constructing a power grid secondary equipment defect knowledge graph;

[0009] S2, mapping the power grid secondary equipment defect knowledge graph to a Bayesian network framework, and assigning values to the prior probability and conditional probability of the Bayesian network in combination with historical defect data, to construct a power grid secondary equipment defect Bayesian network;

[0010] S3, using defect subgraph search of the power grid secondary equipment defect knowledge graph and defect reasoning of the power grid secondary equipment defect Bayesian network to identify power grid secondary equipment defects.

[0011] As a preferred solution, the specific steps of step S1 include:

[0012] S101, modeling the ontology data structure of power grid secondary equipment defect knowledge, which includes entities, attributes and relationships of defect knowledge;

[0013] S102, extracting corresponding knowledge elements from the defect data of the power grid secondary equipment according to the modeling elements of the ontology data structure of the power grid secondary equipment defect knowledge, to obtain the ontology data of the power grid secondary equipment defect knowledge;

[0014] S103, data cleaning processing of the ontology data of the power grid secondary equipment defect knowledge, to unify the entity names so that the expressions for the same entity in each ontology data of multiple data sources are consistent.

[0015] As a preferred solution, step S101 specifically includes: defining the entities, attributes and relationships in the ontology data structure of the power grid secondary equipment defect knowledge according to the characteristics of the power grid secondary equipment and its defect identification requirements; using the defined ontology data structure as a semantic framework to describe the power grid secondary equipment defect knowledge, to complete the construction of the ontology data structure model of the power grid secondary equipment defect knowledge;

[0016] Wherein:

[0017] The entity includes the name of the power grid secondary equipment and the name of each component in the power grid secondary equipment;

[0018] The attribute includes a defect phenomenon, a defect cause, and a treatment measure; the defect phenomenon is used to describe an abnormal state or a fault performance of the power grid secondary equipment; the defect cause is used to describe various reasons for causing the defect phenomenon; and the treatment measure is used to describe a response measure for effectively treating the defect cause;

[0019] The relationship includes a whole and partial relationship, a cause and effect relationship, and a causal relationship; the whole and partial relationship is used to describe a hierarchical relationship between the power grid secondary equipment and its components; the cause and effect relationship is used to describe a relationship in which the defect cause directly or indirectly causes the power grid secondary equipment to have an abnormal state or a fault performance; and the causal relationship is used to describe a causal connection between the defect phenomenon and its potential defect cause.

[0020] As a preferred solution, step S102 specifically includes:

[0021] S1021, pre-training the original defect text by using a bidirectional encoding representation model BERT based on a Transformer architecture to convert it into a word vector representation containing semantic information, and capturing the context semantic relationship of each word in the text sequence;

[0022] S1022, inputting the word vector representation generated by BERT into a bidirectional long short-term memory network BiLSTM to extract context-related features and constructing a feature matrix accordingly;

[0023] S1023, taking the feature matrix as input data of a conditional random field CRF to train and generate semantic structure labels corresponding to entities, attributes, and relationships by calculating transition probabilities;

[0024] S1024, extracting the entities, relationships, and attributes in the defect text according to the semantic structure labels generated by CRF to form the basic units of the power grid secondary equipment knowledge graph.

[0025] As a preferred solution, step S103 specifically includes:

[0026] S1031, converting the entities in the ontology data of the power grid secondary equipment defect knowledge into word vectors and calculating the similarity between these word vectors;

[0027] S1032, when the similarity between two word vectors exceeds a set threshold, it is determined that the entities corresponding to the two word vectors belong to the same entity, and the entity name is unified.

[0028] As a preferred solution, the specific steps of step S2 include:

[0029] S201, mapping entities and attributes in the power grid secondary equipment defect knowledge graph into nodes of the Bayesian network, mapping relationships in the power grid secondary equipment defect knowledge graph into directed edges, and constructing a topology structure of the power grid secondary equipment defect Bayesian network;

[0030] S202, assigning values to the prior probability and conditional probability of the power grid secondary equipment defect Bayesian network in combination with historical prior data of power grid secondary equipment defect knowledge; the prior probability is a set value of preliminary probability estimation based on historical prior data; and the conditional probability calculation formula is:

[0031]

[0032] In the formula, π(U i ) represents other relationship nodes other than the parent nodes of node U i ; P(π(U i )|U i ) represents the probability of occurrence of node π(U i ) event under the condition that node U i event occurs; P(U i |π(U i )) represents the probability of occurrence of node U i event under the condition that node π(U i ) event occurs; P(U i ) and P(π(U i )) respectively represent the occurrence probability of node U i event and node π(U i ) event.

[0033] S203, processing the conditional probability by using a Noisy-OR model to represent the joint action of multiple defect causes on the defect phenomenon, and the calculation formula is:

[0034]

[0035] In the formula, x i represents a defect cause leading to defect phenomenon y i , and x represents a set of defect causes leading to defect phenomenon y i ; represents that defect phenomenon y i does not occur; P(y i |x) represents the probability of occurrence of defect phenomenon y i under the joint action of the set of defect causes; represents the probability of non-occurrence of defect phenomenon y i under the joint action of the set of defect causes.

[0036] As a preferred solution, the specific steps of step S3 include:

[0037] S301, using the ontology data structure information and semantic association of the power grid secondary equipment defect knowledge graph, according to the defect phenomenon, positioning various defect causes related thereto, as the preliminary identified defect causes;

[0038] S302, using the power grid secondary equipment defect Bayesian network to perform posterior probability reasoning on the preliminary identified defect causes, determining the defect cause with the maximum probability and the treatment measures related to the defect cause with the maximum probability;

[0039] S303, outputting the defect cause with the maximum probability and the corresponding treatment measures as the identification result, and using them to guide the maintenance and operation adjustment of the power grid secondary equipment.

[0040] As a preferred scheme, step S301 specifically includes:

[0041] S3011, inputting the component name of the defective power grid secondary equipment and the description information of the defect phenomenon;

[0042] S3012, performing knowledge extraction on the input description information, and analyzing and extracting keywords in the input text;

[0043] S3013, performing word matching on the extracted keywords and the entities and attributes of the ontology data structure in the power grid secondary equipment defect knowledge graph, determining the defect phenomenon of the component with the highest word matching degree in the power grid secondary equipment defect knowledge graph, positioning various defect causes related to the defect phenomenon, as the preliminary identified defect causes.

[0044] As a preferred scheme, step S302 specifically includes:

[0045] S3021, using the reverse reasoning mechanism of the power grid secondary equipment defect Bayesian network corresponding to the power grid secondary equipment defect knowledge graph, to calculate the posterior probability of various preliminary identified defect causes;

[0046] S3022, sorting the various preliminary identified defect causes according to the size of the posterior probability, selecting the defect cause with the maximum probability, and according to the relationship edge of the defect cause with the maximum probability, determining the treatment measures related to the defect cause with the maximum probability.

[0047] As a preferred scheme, after step S303, it further includes:

[0048] S304, after the maintenance and operation adjustment of the power grid secondary equipment, dynamically updating the power grid secondary equipment defect knowledge graph and the power grid secondary equipment defect Bayesian network according to the maintenance feedback and new operation data of the power grid secondary equipment, for subsequent power grid secondary equipment defect identification;

[0049] The step specifically comprises:

[0050] S3041、According to the maximum probability of the defect reason output as the identification result and the corresponding treatment measures, the power grid secondary equipment is maintained and operation adjustment is made;

[0051] S3042、If there is a new defect in the power grid secondary equipment after maintenance, the new defect data is constructed as the ontology data structure in the power grid secondary equipment defect knowledge graph, and is mapped to the nodes and edges in the power grid secondary equipment defect Bayesian network; if the power grid secondary equipment returns to normal operation after maintenance, the actual defect reason and the maintenance result are collected and fed back to the power grid secondary equipment defect Bayesian network, and the probabilities between the corresponding defect phenomenon, defect reason and treatment measures are updated;

[0052] S3043、Re-statistics the prior probability and conditional probability of the updated power grid secondary equipment defect Bayesian network.

[0053] Compared with the prior art, the present application has the following beneficial effects:

[0054] 1、The method of the present application realizes intelligent identification of power grid secondary equipment defects by fusing knowledge graph and Bayesian network. In the processing process, the defect knowledge graph is constructed by semantic extraction of multi-source heterogeneous data, breaking through the limitation of data island and realizing full-dimensional correlation analysis of equipment defect characteristics. Then, the power grid secondary equipment defect knowledge graph is mapped to the Bayesian network framework, and the prior probability and conditional probability of defect occurrence are quantified based on historical data prior knowledge. Finally, the precise fault mode positioning is provided by the defect subgraph search of the knowledge graph, and the defect propagation path prediction is completed by the probability reasoning of the Bayesian network. The combination of the two makes the method of the present application be able to comprehensively utilize the multi-source heterogeneous data of the power grid secondary equipment, have the hierarchical diagnostic identification ability from defect performance characteristic identification to defect occurrence reason, accurately depict the complex relationship and probability influence between defect phenomenon and multi-layer reasons, significantly improve the defect identification accuracy and response timeliness, and then provide information support for the maintenance and operation adjustment of the power grid secondary equipment, and help to improve the safe and stable operation level of the power grid.

[0055] 2、The method of the present application in the construction of power grid secondary equipment defect knowledge base adopts BERT pre-training model to obtain dynamic word vector, effectively captures the context semantic features of defect text; through BiLSTM bidirectional time sequence modeling to extract long distance dependency, constructs high-dimensional feature matrix to retain the complete defect description context; the semantic structure label decoding of entity-relation is realized by using CRF conditional probability constraint, which can effectively guarantee the accuracy of labeling; thus, while ensuring the accuracy of knowledge extraction, the construction of power grid secondary equipment defect knowledge ontology data is realized from structured, semi-structured and unstructured defect data, the cost of manual labeling is reduced, and the integrity and timeliness of the construction of power grid secondary equipment defect knowledge base is improved.

[0056] 3、The method of the present application constructs Bayesian network topology based on knowledge graph structured mapping, which not only retains the semantic correlation of power grid secondary equipment defect knowledge, but also introduces the uncertainty processing ability of probabilistic graph model, and after preliminary assignment of prior probability and conditional probability based on historical prior data, further combines Noisy-OR model to realize joint probability modeling of multiple cause and effect, effectively solves the parameter explosion problem faced by traditional Bayesian network in processing multiple concurrent defect reasons, accurately describes the complex relationship and probability influence between defect phenomenon and multiple reasons under the premise of ensuring reasoning efficiency, and enhances the analysis ability of power grid secondary equipment defect Bayesian network to complex defect scenarios.

[0057] 4、The method of the present application can also realize adaptive optimization of power grid secondary equipment defect identification through closed-loop feedback mechanism, by inputting maintenance feedback and real-time operation data feedback into Bayesian network, recalculating posterior probability and adjusting network parameters, improving the real-time optimization ability of defect identification, which helps to improve the accuracy of power grid secondary equipment defect identification in long-term operation scenario. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below with reference to the drawings, in which:

[0059] Figure 1 is the flow chart of the power grid secondary equipment defect identification method of the present application based on knowledge graph and Bayesian network fusion;

[0060] Figure 2 is an example diagram of power grid secondary equipment defect knowledge graph in the embodiment of the present application;

[0061] Figure 3 is an example diagram of part of nodes and edges of power grid secondary equipment Bayesian network in the embodiment of the present application;

[0062] Figure 4is a defect identification result example diagram of a power grid secondary equipment Bayesian network in an embodiment of the present application;

[0063] Figure 5 is a defect identification update result example diagram of a power grid secondary equipment Bayesian network after operation adjustment in an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0066] As shown in the drawings, Figure 1 The present application discloses a power grid secondary equipment defect identification method based on knowledge graph and Bayesian network fusion, comprising the following steps:

[0067] S1, transforming the defect data of the power grid secondary equipment into a structured knowledge graph, and constructing a power grid secondary equipment defect knowledge graph;

[0068] S2, mapping the power grid secondary equipment defect knowledge graph to a Bayesian network framework, and assigning values to the prior probability and conditional probability of the Bayesian network in combination with historical defect data, to construct a power grid secondary equipment defect Bayesian network;

[0069] S3, using defect subgraph search of the power grid secondary equipment defect knowledge graph and defect reasoning of the power grid secondary equipment defect Bayesian network to identify the defects of the power grid secondary equipment.

[0070] The method of the present application realizes intelligent identification of power grid secondary equipment defects by fusing a knowledge graph and a Bayesian network. In the processing process, the semantic extraction of multi-source heterogeneous data is performed to construct a defect knowledge graph, thereby breaking through the limitation of data islands and realizing full-dimensional correlation analysis of equipment defect characteristics. Then, the power grid secondary equipment defect knowledge graph is mapped to a Bayesian network framework, and the prior probability and conditional probability of defect occurrence are quantified based on historical data prior knowledge. Finally, the precise fault mode positioning is provided by the defect subgraph search of the knowledge graph, and the defect propagation path prediction is completed by the probability reasoning of the Bayesian network. The combination of the two enables the method of the present application to comprehensively utilize multi-source heterogeneous data of power grid secondary equipment, has the hierarchical diagnostic identification capability from defect performance characteristic identification to defect cause, thereby accurately depicting the complex relationship and probability influence between defect phenomena and multi-layer causes, significantly improving the defect identification accuracy and response timeliness, and further providing information support for the maintenance and operation adjustment of power grid secondary equipment, and helping to improve the safe and stable operation level of the power grid.

[0071] In specific implementation, the specific steps of step S1 include:

[0072] S101, modeling the ontology data structure of power grid secondary equipment defect knowledge, the ontology data structure including entities, attributes and relationships of defect knowledge.

[0073] Specifically, according to the characteristics of power grid secondary equipment and the defect identification requirements thereof, the entities, attributes and relationships in the ontology data structure of power grid secondary equipment defect knowledge are defined; the defined ontology data structure is used as a semantic framework for describing power grid secondary equipment defect knowledge, and the modeling of the ontology data structure model of power grid secondary equipment defect knowledge is completed.

[0074] The entities include the name of the power grid secondary equipment and the name of each component part in the power grid secondary equipment.

[0075] The attributes include defect phenomena, defect causes and disposal measures; the defect phenomena are used to describe abnormal states or fault performances of the power grid secondary equipment, the defect causes are used to describe various causes leading to the defect phenomena, and the disposal measures are used to describe the response measures for effectively disposing the defect causes.

[0076] The relationships include whole and partial relationships, causing relationships and causal relationships; the whole and partial relationships are used to describe the hierarchical relationship between the power grid secondary equipment and its component parts; the causing relationships are used to describe the relationship that the defect causes directly or indirectly cause the power grid secondary equipment to appear abnormal states or fault performances; and the causal relationships are used to describe the causal connection between the defect phenomena and the potential defect causes thereof.

[0077] S102, extract corresponding knowledge elements from the defect data of the power grid secondary equipment according to the modeling elements of the ontology data structure of the power grid secondary equipment defect knowledge, and obtain the ontology data of the power grid secondary equipment defect knowledge.

[0078] This step specifically includes:

[0079] S1021, pre-training the original defect text by using a bidirectional encoding representation model BERT based on a Transformer architecture, converting it into a word vector representation containing semantic information, and capturing the context semantic relationship of each word in the text sequence;

[0080] S1022, input the word vector representation generated by BERT into a bidirectional long short-term memory network BiLSTM, extract context-related features, and construct a feature matrix accordingly;

[0081] S1023, take the feature matrix as input data of a conditional random field CRF for training, and generate semantic structure labels corresponding to entities, attributes and relationships by calculating transition probabilities;

[0082] S1024, according to the semantic structure labels generated by CRF, extract entities, relationships and attributes in the defect text to form the basic unit of the power grid secondary equipment knowledge graph.

[0083] In this step S102, the BERT pre-training model is used to obtain dynamic word vectors, effectively capturing the context semantic features of the defect text; the BiLSTM bidirectional time series modeling extracts long-distance dependency relationships, and constructs a high-dimensional feature matrix to preserve the complete defect description context; the CRF conditional probability constraint realizes the semantic structure label decoding of entities-relationships, which can effectively guarantee the accuracy of labeling; Therefore, while ensuring the accuracy of knowledge extraction, the construction of the ontology data of the power grid secondary equipment defect knowledge is realized from the structured, semi-structured and unstructured defect data, reducing the cost of manual labeling, and improving the completeness and timeliness of the construction of the power grid secondary equipment defect knowledge base.

[0084] S103, data cleaning processing is performed on the ontology data of the power grid secondary equipment defect knowledge, and the entity names are unified, so that the expressions of the same entity in the ontology data of each data source are consistent.

[0085] This step specifically includes:

[0086] S1031, convert the entities in the ontology data of the power grid secondary equipment defect knowledge into word vectors, and calculate the similarity between these word vectors;

[0087] S1032, when the similarity between the two word vectors exceeds a set threshold, it is determined that the two word vectors correspond to the same entity, and the entity name is unified.

[0088] This step S103 realizes the entity alignment and cleaning processing of multi-source heterogeneous data through the semantic matching mechanism based on the word vector, which is used to solve the problem of inconsistent expression of the same entity in multiple data sources, eliminate the semantic ambiguity in the establishment stage of the knowledge graph, and provide data quality guarantee for the subsequent construction of the Bayesian network. In specific implementation, the word vector can be extracted by using deep learning technologies such as bidirectional encoding representation model BERT, and the word vector similarity can be calculated by using cosine similarity algorithm and the like, and the reference of the entity name is unified through similarity matching.

[0089] In specific implementation, the specific steps of step S2 include:

[0090] S201, mapping the entities and attributes in the power grid secondary equipment defect knowledge graph into the nodes of the Bayesian network, mapping the relationships in the power grid secondary equipment defect knowledge graph into directed edges, and constructing the topological structure of the power grid secondary equipment defect Bayesian network;

[0091] S202, combining the historical prior data of the power grid secondary equipment defect knowledge to assign values to the prior probability and the conditional probability of the power grid secondary equipment defect Bayesian network; wherein the historical prior data can include historical defect data, past defect records and processing experience, domain knowledge, etc.; the prior probability is a set value of the preliminary probability estimation based on the historical prior data; and the conditional probability calculation formula is:

[0092]

[0093] In the formula, π(U i ) represents other relationship nodes of the parent node of node U i ; P(π(U i )|U i ) represents the probability of occurrence of node π(U i ) event under the condition that node U i event occurs; P(U i |π(U i )) represents the probability of occurrence of node U i event under the condition that node π(U i ) event occurs; P(U i ), P(π(U i )) respectively represent the occurrence probability of node U i event and node π(U i ) event.

[0094] S203, the Noisy-OR model is used to process the conditional probability, so as to represent the joint action of various defect causes on the defect phenomenon, and the calculation formula is as follows:

[0095]

[0096] In the formula, x i represents one defect cause leading to the defect phenomenon y i , and x represents a set of defect causes leading to the defect phenomenon y i . represents that the defect phenomenon y i does not occur; and P(y i |x) represents the probability of the occurrence of the defect phenomenon y i under the joint action of the defect cause set. represents the probability of the occurrence of the defect phenomenon y i under the joint action of the defect cause set.

[0097] This step S2 constructs a Bayesian network topology based on the knowledge graph structured mapping, which not only retains the semantic correlation of the power grid secondary equipment defect knowledge, but also introduces the uncertainty processing capability of the probabilistic graph model. After the preliminary assignment of the prior probability and the conditional probability based on the historical prior data, the joint probability modeling of multiple causes and effects is further realized by combining the Noisy-OR model, which effectively solves the parameter explosion problem faced by the traditional Bayesian network in processing multiple concurrent defect causes. Under the premise of ensuring the reasoning efficiency, the complex relationship and probability influence between the defect phenomenon and the multi-layer causes are accurately described, and the analysis ability of the power grid secondary equipment defect Bayesian network for complex defect scenarios is enhanced.

[0098] In specific implementation, the specific steps of step S3 include:

[0099] S301, according to the defect phenomenon, the various defect causes related to the defect phenomenon are positioned by using the ontology data structure information and semantic correlation of the power grid secondary equipment defect knowledge graph, and the various defect causes are taken as the preliminary identified defect causes.

[0100] This step specifically includes:

[0101] S3011, input the component name of the defect power grid secondary equipment and the description information of the defect phenomenon;

[0102] S3012, knowledge extraction is performed on the input description information, and the keywords in the input text are analyzed and extracted;

[0103] S3013, match the extracted keywords with entities and attributes of the ontology data structure in the power grid secondary equipment defect knowledge graph, determine the defect phenomenon of the component with the highest word matching degree in the power grid secondary equipment defect knowledge graph, and locate various defect causes related to the defect phenomenon as the preliminary identified defect causes.

[0104] This step S301 first performs cross-dimension association analysis of defect phenomena and potential causes based on the semantic retrieval framework and entity association characteristics constructed by the power grid secondary equipment defect knowledge graph, provides a high-confidence candidate set for subsequent reasoning of the final defect cause, and helps to reduce the processing and computing complexity while ensuring the comprehensiveness of defect identification.

[0105] S302, using the power grid secondary equipment defect Bayesian network, the posterior probability of the preliminary identified defect causes is inferred to determine the defect cause with the maximum probability and the disposal measures related to the defect cause with the maximum probability.

[0106] This step specifically includes:

[0107] S3021, using the reverse reasoning mechanism of the power grid secondary equipment defect Bayesian network corresponding to the power grid secondary equipment defect knowledge graph, the posterior probability of the preliminary identified various defect causes is calculated.

[0108] S3022, according to the size of the posterior probability, the preliminary identified various defect causes are sorted, the defect cause with the maximum probability is selected, and the disposal measures related to the defect cause with the maximum probability are determined according to the relationship edge of the defect cause with the maximum probability.

[0109] This step S302 uses the reverse reasoning engine of the power grid secondary equipment defect Bayesian network to realize the deduction of the final defect cause through the posterior probability calculation, and determines the disposal measures corresponding to the final defect cause through the automatic association of the defect causes and the disposal measures reflected by the knowledge graph relationship edge, forming a closed-loop decision link of "probability reasoning-reason positioning-measure matching". It helps to improve the scientificity and pertinence of the maintenance and operation adjustment decision making for the power grid secondary equipment while ensuring the accuracy of the diagnosis.

[0110] S303, the defect cause with the maximum probability and the corresponding disposal measures are output as the identification result, and are used to guide the maintenance and operation adjustment of the power grid secondary equipment.

[0111] Therefore, the defect identification task for the power grid secondary equipment is completed.

[0112] In addition, the method of the present application can also have the ability to update the ability in real time according to the maintenance feedback and real-time operation data. Specifically, after step S303, it can also include:

[0113] S304, after the maintenance and operation adjustment of the power grid secondary equipment, the power grid secondary equipment defect knowledge graph and the power grid secondary equipment defect Bayesian network are dynamically updated according to the maintenance feedback and new operation data of the power grid secondary equipment, for subsequent power grid secondary equipment defect identification;

[0114] This step specifically includes:

[0115] S3041, according to the maximum probability defect reason and the corresponding treatment measures output as the identification result, the power grid secondary equipment is maintained and operated;

[0116] S3042, if there is a new defect in the power grid secondary equipment after maintenance, the new defect data is constructed as the ontology data structure in the power grid secondary equipment defect knowledge graph and is mapped to the nodes and edges in the power grid secondary equipment defect Bayesian network; if the power grid secondary equipment returns to normal operation after maintenance, the actual defect reason and the maintenance result are collected and fed back to the power grid secondary equipment defect Bayesian network, and the probabilities between the corresponding defect phenomena, defect reasons and treatment measures are updated;

[0117] S3043, the prior probability and the conditional probability of the updated power grid secondary equipment defect Bayesian network are re-counted.

[0118] Therefore, the power grid secondary equipment defect identification method of the present application realizes the adaptive optimization of power grid secondary equipment defect identification through a closed-loop feedback mechanism, re-calculates the posterior probability and adjusts the network parameters by feeding back the maintenance feedback and real-time operation data to the Bayesian network, improves the real-time optimization capability of defect identification, and helps to improve the accuracy of power grid secondary equipment defect identification in long-term operation scenarios.

[0119] Embodiment:

[0120] In order to verify the effectiveness of the method of the present application, the scheme of the present application is analyzed and verified by the following embodiment.

[0121] In this embodiment, 5000 secondary equipment defect text records of a power grid company are selected to form a knowledge graph, and a power grid secondary equipment defect knowledge graph containing 1265 nodes and 1564 relationships is constructed, as shown in Figure 2 In the example shown in Figure 2 The entities and attributes in the ontology data structure in the power grid secondary equipment defect knowledge graph are represented by dots of different colors, and the relationships between entities and attributes are represented by lines between the dots. Figure 2In the example graph, the name of the power grid secondary equipment is represented by a red dot, the name of each component of different levels in the power grid secondary equipment is represented by a pink dot and a blue dot, the defect phenomenon is represented by a yellow dot, the defect cause is represented by a green dot, and the treatment measure is represented by a gray dot. The connection between different dots represents the local relationship, the leading relationship and the causal relationship between them. The text information in the graph is not important. The overall form of the constructed power grid secondary equipment defect knowledge graph is mainly presented through these color dots and lines.

[0122] In the process of mapping the Bayesian network according to the power grid secondary equipment defect knowledge graph, the nodes and edges of the Bayesian network need to be established based on the entity, attribute and relationship information in the ontology data structure. Take the “fiber channel alarm” as an example to illustrate the power grid secondary equipment defect identification based on the fusion of knowledge graph and Bayesian network. According to the knowledge graph subgraph query display, the possible causes of the fiber channel alarm include: jumper connector loosening, tail fiber bending too large or connector contamination, fiber fusion quality problem, and multiplex interface device failure. The causes of the multiplex interface device failure include unstable power supply, incorrect multiplex mode configuration and link interruption. After the retrieval is completed, the knowledge graph is converted into a Bayesian network, and GeNIe or other Bayesian modeling tools can be used to model the retrieval results. Part of the nodes and edges of the power grid secondary equipment Bayesian network are shown in the example of FIG. 8. Figure 3 The above secondary equipment defect text data is used to parameterize the Bayesian network. The prior probability of the leaf node “fiber channel alarm” is 0.04, the prior probability of the intermediate node “multiplex interface device failure” is 0.07, and the prior probabilities of the other root nodes “jumper connector loosening”, “tail fiber bending too large or connector contamination”, “fiber fusion quality problem”, “unstable power supply”, “incorrect multiplex mode configuration” and “link interruption” are 0.05, 0.02, 0.02, 0.03, 0.04 and 0.04 respectively.

[0123] According to the Noisy-OR node model, each node is processed to determine the conditional probability distribution of the Bayesian network. Take the node y1 “multiplex interface device failure” and its three parent nodes x4, x5 and x6 as an example. The detailed data is shown in Table 1. At the same time, there may be unrecorded defect causes in engineering practice. Therefore, when calculating the conditional probability, it is assumed that the Leaky node always has a certain influence on the occurrence of defects, and the influence strength of the node on the defects is set to 0.05. For example, when x4, x5 and x6 do not occur, y1 may also occur, that is,

[0124] Table 1. Defect phenomenon y1 conditional probability calculation steps

[0125]

[0126] On the basis of obtaining all conditional probability distributions, quantitative evaluation is carried out by using Bayesian network inference. The state probability of the node "fiber channel alarm" is preset as 100%, and then the posterior probability of each defect reason is calculated by using the reverse inference mechanism of the Bayesian network. The defect identification result of the power grid secondary equipment Bayesian network is as shown in Figure 4 When the fiber channel alarm, the probabilities of occurrence of the reasons of the jumper connector loose, the fiber tail bending too large or the connector being contaminated, the fiber fusion quality problem, the link interruption, the unstable power supply and the multiplexing mode configuration error are 18%, 7%, 8%, 13%, 8% and 9% respectively. The greater the probability of occurrence of the defect reason is, the higher the detection priority is, so as to faster eliminate the defect. From Figure 4 , the posterior probabilities of the defect root causes are sorted as x1>x4>x6>x3>x5. Therefore, the detection order of the defect reasons is x1, x4, x6, x3 and x5. After the device operation adjustment and maintenance are carried out according to the detection order, if the information obtained determines that the node y1 multiplexing interface equipment failure occurs, the state value of the node is set as 100%, and at this time, the defect identification update result of the power grid secondary equipment Bayesian network is as shown in Figure 5 It can be known that the probability of the "link interruption" node in the root node is the highest, that is, it is the most possible defect reason.

[0127] The 1000 defect records are searched by using the LSI, the LDA, the single knowledge graph model and the power grid secondary equipment defect identification model fused with the knowledge graph and the Bayesian network respectively, and the search results of the models are counted, as shown in Table 2.

[0128] Table 2 Search result statistics of various models

[0129]

[0130] The comparison result shows that the fusion model of the knowledge graph and the Bayesian network is superior to the LSI, the LDA and the single knowledge graph model in terms of the accuracy, the recall rate and the F1 value. It can be seen that the method of the present application can accurately identify the key information, carry out knowledge and probability inference, realize modeling and analysis on the complex relationship between the defect phenomenon and the reasons by fusing the knowledge graph and the Bayesian network.

[0131] In summary, compared with the prior art, the present application has the following technical advantages:

[0132] 1. The method of the present application realizes intelligent identification of power grid secondary equipment defects by fusing knowledge graph and Bayesian network. In the processing process, the semantic extraction is performed on the multi-source heterogeneous data to construct the defect knowledge graph, thereby breaking through the limitation of data island and realizing full-dimensional correlation analysis of equipment defect characteristics. Then, the power grid secondary equipment defect knowledge graph is mapped to the Bayesian network framework, the prior probability and conditional probability of defect occurrence are quantified based on historical data prior knowledge, finally, the accurate fault mode positioning is provided by means of defect subgraph search of the knowledge graph, and the defect propagation path prediction is completed by means of probability reasoning of the Bayesian network, so that the method of the present application can comprehensively utilize the multi-source heterogeneous data of the power grid secondary equipment, has the hierarchical diagnostic identification ability from defect performance characteristic identification to defect cause, thereby accurately depicting the complex relationship and probability influence between the defect phenomenon and the multi-layer causes, significantly improving the defect identification accuracy and response timeliness, and further providing information support for the maintenance and operation adjustment of the power grid secondary equipment, and helping to improve the safe and stable operation level of the power grid.

[0133] 2. In the construction of the power grid secondary equipment defect knowledge base, the BERT pre-training model is adopted to obtain dynamic word vectors, so as to effectively capture the context semantic features of the defect text; the long-distance dependency relationship is extracted by means of BiLSTM bidirectional time sequence modeling, a high-dimensional feature matrix is constructed to retain the complete defect description context; the semantic structure label decoding of entity-relation is realized by means of CRF conditional probability constraint, so as to effectively ensure the accuracy of labeling; thereby, while ensuring the knowledge extraction accuracy, the knowledge elements are extracted from the structured, semi-structured and unstructured defect data to complete the construction of the ontology data of the power grid secondary equipment defect knowledge, the artificial labeling cost is reduced, and the integrity and timeliness of the construction of the power grid secondary equipment defect knowledge base are improved.

[0134] 3. The method of the present application constructs the Bayesian network topology based on the knowledge graph structured mapping, which not only retains the semantic correlation of the power grid secondary equipment defect knowledge, but also introduces the uncertainty processing ability of the probabilistic graph model, and after the preliminary assignment of the prior probability and conditional probability based on the historical prior data, the joint probability modeling of multiple causes is further realized by means of the Noisy-OR model, thereby effectively solving the parameter explosion problem faced by the traditional Bayesian network in processing multiple concurrent defect causes, accurately depicting the complex relationship and probability influence between the defect phenomenon and the multi-layer causes under the premise of ensuring the inference efficiency, and enhancing the analysis ability of the power grid secondary equipment defect Bayesian network to the complex defect scene.

[0135] 4、The method can also realize adaptive optimization of power grid secondary equipment defect identification through a closed-loop feedback mechanism, input maintenance feedback and real-time operation data feedback into the Bayesian network, recalculate the posterior probability and adjust the network parameters, improve the real-time optimization capability of defect identification, and help improve the accuracy of power grid secondary equipment defect identification in long-term operation scenarios.

[0136] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and not to limit the technical solutions, and those of ordinary skill in the art should understand that those who modify or equivalently replace the technical solutions of the present application without departing from the purpose and scope of the technical solutions should be covered in the scope of claims of the present application.

Claims

1. A method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network, characterized in that: The steps include: S1. Convert the defect data of power grid secondary equipment into a structured knowledge graph and construct a knowledge graph of power grid secondary equipment defects; S2. Map the knowledge graph of power grid secondary equipment defects to the Bayesian network framework, and assign the prior probability and conditional probability of the Bayesian network in combination with historical defect data to construct a power grid secondary equipment defect Bayesian network; S3. Identify defects in the power grid secondary equipment by using the defect subgraph search of the power grid secondary equipment defect knowledge graph and the defect reasoning of the power grid secondary equipment defect Bayesian network.

2. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 1 is characterized in that: The specific steps of step S1 include: S101. Modeling an ontology data structure of grid secondary equipment defect knowledge, wherein the ontology data structure includes entities, attributes, and relationships of the defect knowledge; S102. Extract corresponding knowledge elements from the defect data of the power grid secondary equipment based on the modeling elements of the ontology data structure of the power grid secondary equipment defect knowledge to obtain the ontology data of the power grid secondary equipment defect knowledge; S103: Perform data cleaning on the ontology data of the power grid secondary equipment defect knowledge, unify the entity names, and make the description of the same entity consistent in each ontology data of multiple data sources.

3. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 2 is characterized in that: Step S101 specifically includes: defining entities, attributes, and relationships in the ontology data structure of power grid secondary equipment defect knowledge based on the characteristics of power grid secondary equipment and its defect identification requirements; using the defined ontology data structure as a semantic framework for describing power grid secondary equipment defect knowledge, thereby completing the construction of the ontology data structure model of power grid secondary equipment defect knowledge; in: The entity includes the name of the power grid secondary equipment and the name of each component in the power grid secondary equipment; Attributes include defect phenomenon, defect cause, and treatment measures; the defect phenomenon is used to describe the abnormal state or fault manifestation of the power grid secondary equipment, the defect cause is used to describe the various causes that lead to the defect phenomenon; the treatment measures are used to describe the response measures to effectively deal with the defect cause; Relationships include overall and local relationships, causal relationships, and cause-and-effect relationships. The overall and local relationships are used to describe the hierarchical relationship between power grid secondary equipment and its components. The cause-and-effect relationship is used to describe the relationship in which the defect cause directly or indirectly leads to abnormal conditions or fault manifestations in power grid secondary equipment. The cause-and-effect relationship is used to describe the causal relationship between the defect phenomenon and its potential defect cause.

4. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 2 is characterized in that: Step S102 specifically includes: S1021. Use the Transformer-based bidirectional encoding representation model BERT to pre-train the original defect text and convert it into a word vector representation containing semantic information, capturing the contextual semantic relationship of each word in the text sequence; S1022. Input the word vector representation generated by BERT into the bidirectional long short-term memory network BiLSTM to extract context-related features and construct a feature matrix based on them; S1023, using the feature matrix as input data for training the conditional random field (CRF), and generating semantic structure labels corresponding to entities, attributes, and relationships by calculating transition probabilities; S1024. Based on the semantic structure labels generated by CRF, entities, relationships, and attributes in the defect text are extracted to form the basic units of the knowledge graph of power grid secondary equipment.

5. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 2 is characterized in that: Step S103 specifically includes: S1031. Convert entities in the ontology data of power grid secondary equipment defect knowledge into word vectors, and calculate the similarity between these word vectors; S1032. When the similarity between two word vectors exceeds a set threshold, it is determined that the entities corresponding to the two word vectors belong to the same entity, and the entity names are unified.

6. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 3 is characterized in that: The specific steps of step S2 include: S201. Mapping entities and attributes in the power grid secondary equipment defect knowledge graph into nodes of a Bayesian network, mapping relationships in the power grid secondary equipment defect knowledge graph into directed edges, and constructing a topological structure of the power grid secondary equipment defect Bayesian network; S202. Assign a priori probability and a conditional probability to the Bayesian network of power grid secondary equipment defects based on historical priori data of power grid secondary equipment defects; the priori probability is a set value of a preliminary probability estimate based on historical priori data; the conditional probability calculation formula is: In the formula, π(U i ) represents node U i Other relationship nodes other than the parent node; P(π(U i )|U i ) indicates that at node U i When an event occurs, node π(U i ) probability of an event; P(U i |π(U i )) represents the node π(U i ) event occurs when node U i The probability of an event; P(U i )、P(π(U i )) represent nodes U i Event, node π(U i ) the probability of the event occurring; S203. Use the Noisy-OR model to process the conditional probability to characterize the joint effect of multiple defect causes on the defect phenomenon. The calculation formula is: Where x i Indicates the defect phenomenon y i A defect cause, x represents the defect phenomenon y i The set of defect causes; Indicates defect phenomenon y i Did not occur; P(y i |x) represents defect phenomenon y i The probability of occurrence under the combined action of a set of defect causes; Indicates defect phenomenon y i The probability that a defect does not occur under the combined effect of a set of defect causes.

7. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 3 is characterized in that: The specific steps of step S3 include: S301. Using the ontological data structure information and semantic associations of the power grid secondary equipment defect knowledge graph, various defect causes related to the defect phenomena are located as the preliminary identified defect causes. S302: Using a Bayesian network for power grid secondary equipment defects, perform posterior probability reasoning on the initially identified defect causes to determine the defect cause with the highest probability and a treatment measure related to the defect cause with the highest probability; S303: Output the defect cause with the highest probability and the corresponding treatment measures as the identification result, and use them to guide the maintenance and operation adjustment of the power grid secondary equipment.

8. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 7 is characterized in that: Step S301 specifically includes: S3011. Input the component name of the defective power grid secondary equipment and description of the defect phenomenon; S3012, performing knowledge extraction on the input description information, analyzing and extracting keywords in the input text; S3013. Match the extracted keywords with the entities and attributes of the ontology data structure in the power grid secondary equipment defect knowledge graph, determine the defect phenomenon of the component with the highest word matching degree in the power grid secondary equipment defect knowledge graph, and locate various defect causes that are related to the existence of the defect phenomenon as the preliminary identified defect causes.

9. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 7 is characterized in that: Step S302 specifically includes: S3021. Calculate the posterior probabilities of various initially identified defect causes using the reverse reasoning mechanism of the power grid secondary equipment defect Bayesian network corresponding to the power grid secondary equipment defect knowledge graph; S3022. Sort the various defect causes initially identified according to the size of the posterior probability, select the defect cause with the highest probability, and determine the treatment measures related to the defect cause with the highest probability based on the relationship edge of the defect cause with the highest probability.

10. The method for identifying defects in power grid secondary equipment based on the fusion of knowledge graph and Bayesian network according to claim 7, characterized in that: After step S303, the following steps are also included: S304. After the grid secondary equipment is repaired and its operation is adjusted, the grid secondary equipment defect knowledge graph and the grid secondary equipment defect Bayesian network are dynamically updated based on the maintenance feedback and new operation data of the grid secondary equipment for subsequent grid secondary equipment defect identification. This step specifically includes: S3041. Repair and adjust the operation of the power grid secondary equipment based on the defect cause with the highest probability output as the identification result and the corresponding treatment measures; S3042. If new defects are found in the power grid secondary equipment after repair, the new defect data is constructed as an ontological data structure in the power grid secondary equipment defect knowledge graph and mapped to nodes and edges in the power grid secondary equipment defect Bayesian network. If the power grid secondary equipment resumes normal operation after repair, the actual defect cause and repair result are collected and fed back to the power grid secondary equipment defect Bayesian network to update the corresponding probability of defect phenomenon, defect cause, and treatment measure. S3043. Recalculate the updated prior probability and conditional probability of the power grid secondary equipment defect Bayesian network.

Images