CN-121981219-A - Synchronous camera debugging fault diagnosis method based on knowledge graph and large language model

CN121981219ACN 121981219 ACN121981219 ACN 121981219ACN-121981219-A

Abstract

The invention discloses a synchronous camera fault diagnosis method based on a knowledge graph and a large language model, which comprises the steps of obtaining operation data of a synchronous camera, preprocessing the operation data to form a unified corpus and a standardized data set, encoding the corpus based on the unified corpus to obtain context perception vector characterization by using a BERT model, extracting sequence labels and relations based on the context perception vector characterization, outputting structured knowledge data, constructing a synchronous camera fault diagnosis knowledge graph based on the structured knowledge data, carrying out model training and optimization based on a pre-trained large language model and utilizing the standardized data set to understand and analyze a fault phenomenon described by natural language, outputting a semantic analysis result, and carrying out mixed reasoning by combining the output of the knowledge graph and the semantic analysis result to obtain the fault diagnosis result. The invention has the advantages of high diagnosis precision, high diagnosis efficiency and the like.

Inventors

ZHANG DONGQING
ZHOU JUN
JU YUNTAO
ZHANG CHAOFENG
ZHAO WENQIANG
ZHANG GUOHUA
WANG JIN
LV JIWEI
LI FEIYI
WANG ZHENGWEI
Tang hengwei

Assignees

国家电网有限公司直流技术中心
国网湖南省电力有限公司超高压变电公司
国网青海省电力公司电力科学研究院
国网青海省电力公司
国网湖南省电力有限公司
国家电网有限公司
清华海峡研究院（厦门）

Dates

Publication Date: 20260505
Application Date: 20251202

Claims (17)

1. A synchronous camera debugging fault diagnosis method based on a knowledge graph and a large language model is characterized by comprising the following steps: acquiring operation data of a synchronous camera, and preprocessing the operation data to form a unified corpus and a standardized data set; Based on the unified corpus, encoding the corpus by using a BERT model to obtain context perception vector representation, and based on the context perception vector representation, performing sequence labeling and relation extraction through a two-way long-short-term memory network and a conditional random layer, and outputting structured knowledge data; Constructing a fault diagnosis knowledge graph of the synchronous camera based on the structured knowledge data; Based on a pre-trained large language model, carrying out model training and optimization by utilizing the standardized data set, understanding and analyzing the fault phenomenon described by natural language, and outputting a semantic analysis result; and carrying out mixed reasoning by combining the output of the knowledge graph and the semantic analysis result to obtain a fault diagnosis result.
2. The knowledge-graph-and-large-language-model-based synchronous camera fault diagnosis method according to claim 1, wherein the specific process of preprocessing the operation data is: The running data is subjected to time stamp alignment and format unification, so that the consistency of the data in the time dimension is ensured; normalizing the time sequence monitoring data, removing abnormal values and unifying sampling rates; And performing word segmentation, denoising and term normalization processing on the unstructured text to form a unified corpus.
3. The knowledge graph and large language model based synchronous camera fault diagnosis method according to claim 1, wherein based on the unified corpus, the method uses the BERT model to encode the corpus to obtain context awareness vector representation, and based on the context awareness vector representation, the specific process of outputting structured knowledge data is as follows: Inputting the text corpus in the corpus into a pre-trained BERT model, and acquiring a context sensing vector representation H BERT of the characters by using a deep transducer encoder of BERT; Inputting the vector representation H BERT into a two-way long-short-term memory network to capture the two-way long-term dependency relationship in the text sequence and outputting a hidden state H BiLSTM ; splicing and fusing the vector representation H BERT and the hidden state H BiLSTM to obtain fused features ; Features after fusion Inputting the entity labeling sequence into a conditional random layer, and decoding to obtain the globally optimal entity labeling sequence by utilizing constraint relations among learning labels of the conditional random layer; and (3) coding the sentences containing the identified entities again, inputting the coding result into a TextCNN model, extracting local features of the sentences representing the entity relationship by TextCNN through a multi-scale convolution kernel, and sending the sentences into a classifier after maximum pooling and splicing to realize the judgment of the relationship, wherein the TextCNN model is a text convolution neural network model.
4. The knowledge-graph-and-large-language-model-based synchronous camera fault diagnosis method according to claim 1, 2 or 3, wherein the specific process of constructing the synchronous camera fault diagnosis knowledge graph based on the structured knowledge data is as follows: Based on the principle of domain ontology, a top-down ontology construction method is adopted to clearly define core concepts, attributes and relations in the field of camera fault diagnosis, so as to form an ontology hierarchical structure; the method comprises the steps of carrying out formal description on a body, carrying out visual modeling, filling the extracted entity and relation into the body as examples, and constructing a synchronous camera fault diagnosis knowledge graph; Introducing a fault mode, influence and hazard analysis method into the knowledge graph, grading the bottom event in the graph from three dimensions of severity S, occurrence degree O and detectability D, and calculating risk priority number RPN; And defining constraint relations among classes, and carrying out ontology logic consistency check.
5. The knowledge graph and large language model based synchronous camera fault diagnosis method according to claim 4, wherein a calculation formula of a risk priority number RPN is: 。
6. the knowledge-graph-and-large-language-model-based synchronous camera malfunction diagnosis method according to claim 1, 2 or 3, wherein the specific process of outputting the semantic analysis result is: When a user inputs a natural language query, matching the user query with the entities in the knowledge graph, and searching related entities and subgraphs as context information; performing domain-specific fine tuning on the large language model by using tuning camera fault data; And inputting the user query, the retrieved knowledge graph context and the trimmed model parameters into the large language model together to generate a semantic analysis result.
7. The knowledge graph and large language model based synchronous camera fault diagnosis method according to claim 1, 2 or 3, further comprising dynamically updating system knowledge through an online learning mechanism, and specifically comprising the steps of: designing an online learning mechanism, dynamically updating the parameters of a knowledge graph and a large language model by using new data, and realizing the continuous evolution of knowledge; Introducing a Bayes deep learning framework, carrying out uncertainty quantification on the diagnosis result, and evaluating the credibility of the result; And performing reliability scoring on each diagnosis result through posterior distribution calculation so as to enhance the reliability of the system.
8. A computer program product comprising a computer program which, when run by a processor, performs the steps of the method according to any one of claims 1-7.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, performs the steps of the method according to any of claims 1-7.
10. A computer system comprising a memory and a processor connected to each other, the memory having stored thereon a computer program which, when run by the processor, performs the steps of the method according to any of claims 1-7.
11. A knowledge graph and large language model based synchronous camera fault diagnosis system, comprising: the multi-source data acquisition module is used for acquiring the operation data of the synchronous camera and preprocessing the operation data to form a unified corpus and a standardized data set; the knowledge graph construction module is used for coding the corpus by using a BERT model based on the unified corpus to obtain context perception vector representation, and carrying out sequence labeling and relation extraction through a bidirectional long-short-term memory network and a conditional random layer based on the context perception vector representation to output structured knowledge data; the large language model processing module is used for carrying out model training and optimization by utilizing the standardized data set based on a pre-trained large language model, understanding and analyzing the fault phenomenon described by natural language and outputting a semantic analysis result; And the fault diagnosis reasoning module is used for carrying out mixed reasoning by combining the output of the knowledge graph and the semantic analysis result to obtain a fault diagnosis result.
12. The knowledge graph and large language model based synchronous camera fault diagnosis system according to claim 11, wherein the specific process of preprocessing the operation data in the multi-source data acquisition module is as follows: The running data is subjected to time stamp alignment and format unification, so that the consistency of the data in the time dimension is ensured; normalizing the time sequence monitoring data, removing abnormal values and unifying sampling rates; And performing word segmentation, denoising and term normalization processing on the unstructured text to form a unified corpus.
13. The knowledge-graph and large language model-based synchronous camera fault diagnosis system according to claim 11, wherein in the knowledge-graph construction module, the specific process of outputting structured knowledge data is as follows: Inputting the text corpus in the corpus into a pre-trained BERT model, and acquiring a context sensing vector representation H BERT of the characters by using a deep transducer encoder of BERT; Inputting the vector representation H BERT into a two-way long-short-term memory network to capture the two-way long-term dependency relationship in the text sequence and outputting a hidden state H BiLSTM ; splicing and fusing the vector representation H BERT and the hidden state H BiLSTM to obtain fused features ; Features after fusion Inputting the entity labeling sequence into a conditional random layer, and decoding to obtain the globally optimal entity labeling sequence by utilizing constraint relations among learning labels of the conditional random layer; and (3) coding the sentences containing the identified entities again, inputting the coding result into a TextCNN model, extracting local features of the sentences representing the entity relationship by TextCNN through a multi-scale convolution kernel, and sending the sentences into a classifier after maximum pooling and splicing to realize the judgment of the relationship, wherein the TextCNN model is a text convolution neural network model.
14. The knowledge-graph and large language model-based synchronous camera fault diagnosis system according to claim 11, 12 or 13, wherein the knowledge-graph construction module constructs a synchronous camera fault diagnosis knowledge graph based on the structured knowledge data by the following steps: Based on the principle of domain ontology, a top-down ontology construction method is adopted to clearly define core concepts, attributes and relations in the field of camera fault diagnosis, so as to form an ontology hierarchical structure; the method comprises the steps of carrying out formal description on a body, carrying out visual modeling, filling the extracted entity and relation into the body as examples, and constructing a synchronous camera fault diagnosis knowledge graph; Introducing a fault mode, influence and hazard analysis method into the knowledge graph, grading the bottom event in the graph from three dimensions of severity S, occurrence degree O and detectability D, and calculating risk priority number RPN; And defining constraint relations among classes, and carrying out ontology logic consistency check.
15. The knowledge-graph and large language model-based synchronous camera fault diagnosis system according to claim 14, wherein the calculation formula of the risk priority number RPN is: 。
16. The knowledge-graph-and-large-language-model-based synchronous camera fault diagnosis system according to claim 11, 12 or 13, wherein the large-language-model processing module outputs the semantic analysis result by the following specific processes: When a user inputs a natural language query, matching the user query with the entities in the knowledge graph, and searching related entities and subgraphs as context information; performing domain-specific fine tuning on the large language model by using tuning camera fault data; And inputting the user query, the retrieved knowledge graph context and the trimmed model parameters into the large language model together to generate a semantic analysis result.
17. The knowledge-graph and large language model based synchronous camera fault diagnosis system according to claim 11, 12 or 13, further comprising an online learning module for dynamically updating system knowledge through an online learning mechanism, and the specific process is as follows: designing an online learning mechanism, dynamically updating the parameters of a knowledge graph and a large language model by using new data, and realizing the continuous evolution of knowledge; Introducing a Bayes deep learning framework, carrying out uncertainty quantification on the diagnosis result, and evaluating the credibility of the result; And performing reliability scoring on each diagnosis result through posterior distribution calculation so as to enhance the reliability of the system.

Description

Synchronous camera debugging fault diagnosis method based on knowledge graph and large language model Technical Field The invention mainly relates to the technical field of power equipment detection, in particular to a synchronous camera fault diagnosis method based on a knowledge graph and a large language model. Background With the increasing demand for reactive support by new power systems, stable operation of the cameras has become critical. The synchronous regulator (Synchronous Condenser) is used as a key rotary standby reactive power source for guaranteeing the voltage stability of the power grid, and has an irreplaceable function in the power grid with ultra-high voltage direct current transmission and high new energy duty ratio. The system has the advantages of strong overload capacity, excellent fault ride-through capacity, good transient response performance and the like, so that the system can provide effective reactive support when a serious fault occurs in the system, and the stability of the power grid is obviously improved. The Knowledge Graph (KG) is used as a structural semantic network technology, so that complex correlations among fault phenomena, reasons, component systems and maintenance measures can be clearly expressed, and a Knowledge system which can be understood by a machine is constructed. In recent years, knowledge graphs have demonstrated potential in power equipment fault diagnosis, for example, intelligent fault positioning schemes of emerging communication integrate multi-source data through the knowledge graphs to realize precipitation and multiplexing of fault knowledge. Meanwhile, the breakthrough of the Large Language Model (LLM) in the aspect of natural language processing provides a new method for analyzing unstructured texts (such as maintenance records and technical manuals). LLM can understand fault description text, generate diagnosis suggestion, and reduce "illusion" problem by retrieving enhanced generation (RAG) framework, and promote diagnosis interpretability. The digital twin (DIGITAL TWIN) technology provides a dynamic context for fault diagnosis by mapping the physical device state in real time, thereby further enhancing the accuracy of analysis. Current camera fault diagnosis techniques still have significant limitations. First, the conventional method is highly dependent on expert experience, knowledge is scattered in the form of unstructured text in the historical report and case records, and systematic integration and iteration are difficult. Secondly, rule-based systems lack flexibility and cannot adapt to complex and varied fault scenarios. For example, traditional Fault Tree Analysis (FTA) and fault pattern, impact and hazard analysis (FMECA) rely heavily on manual modeling, making dynamic updates difficult. The application of the knowledge graph in the fault diagnosis of the camera is still in the primary stage, and the existing research focuses on knowledge association and visualization, and fails to deeply fuse a reliability analysis method and real-time data. The knowledge graph has high construction cost, lag in updating and lack of self-adaptive learning mechanism. In addition, conventional methods have difficulty in processing multi-modal data (e.g., vibration signals, temperature data text descriptions), resulting in insufficient extraction of fault features. While large language models perform well in text understanding, application alone may lead to inaccurate diagnostic decisions due to lack of domain knowledge. The existing system has the defects in real-time performance and interpretability. For example, the optimization method based on mixed integer programming is long in calculation time and cannot meet the requirement of on-line decision making, and the deep learning model is often used as a black box, so that a clear reasoning path is difficult to provide, and the trust degree of operation and maintenance personnel is reduced. These problems together restrict the improvement of fault diagnosis efficiency, and an integrated solution integrating multiple technical advantages is needed. Disclosure of Invention Aiming at the technical problems existing in the prior art, the invention provides a synchronous camera fault diagnosis method with high diagnosis precision and high diagnosis efficiency based on a knowledge graph and a large language model. In order to solve the technical problems, the technical scheme provided by the invention is as follows: a synchronous camera fault diagnosis method based on a knowledge graph and a large language model comprises the following steps: acquiring operation data of a synchronous camera, and preprocessing the operation data to form a unified corpus and a standardized data set; Based on the unified corpus, encoding the corpus by using a BERT model to obtain context perception vector representation, and based on the context perception vector representation, performing sequence labeling and relation e