Search

CN-122021829-A - Method and system for constructing fault knowledge graph of water turbine speed regulating system

CN122021829ACN 122021829 ACN122021829 ACN 122021829ACN-122021829-A

Abstract

The invention provides a method and a system for constructing a fault knowledge graph of a water turbine speed regulating system, and belongs to the technical field of knowledge graphs. The method comprises the steps of obtaining a fault text of a hydraulic turbine speed regulation system, predefining entity types and relation types in the fault text, conducting knowledge extraction on the fault text through a named entity identification model and a relation identification model to obtain a basic triplet set, conducting knowledge extraction on a large language model with the basic triplet set input into fine tuning to obtain an inference triplet set, conducting knowledge updating on the basic triplet set through the inference triplet set to obtain a structured triplet set, storing the structured triplet set into a graph database, and conducting knowledge fusion in the storage process to obtain a fault knowledge map. The invention combines knowledge extraction based on the hybrid neural network model with knowledge reasoning based on the large language model, thereby ensuring the authenticity and reliability of fault knowledge extraction.

Inventors

  • ZHANG KEFEI
  • ZHANG TIANBAO
  • LIU SHENG

Assignees

  • 湖北工业大学

Dates

Publication Date
20260512
Application Date
20260130

Claims (10)

  1. 1. The construction method of the fault knowledge graph of the hydraulic turbine speed regulation system is characterized by comprising the following steps of: acquiring a fault text of a hydraulic turbine speed regulating system, and predefining entity types and relationship types in the fault text; Pre-training a named entity recognition model and a relationship recognition model based on a hybrid neural network according to the entity category and the relationship category; knowledge extraction is carried out on the fault text through the named entity recognition model and the relation recognition model, so that a basic triplet set is obtained; Inputting the basic triplet set into a fine-tuned large language model for knowledge extraction to obtain an inference triplet set; Carrying out knowledge updating on the basic triplet set through the reasoning triplet set to obtain a structured triplet set; and storing the structured triplet set into a graph database, and carrying out knowledge fusion in the storage process to obtain a fault knowledge graph.
  2. 2. The method for constructing the fault knowledge graph of the water turbine speed regulating system according to claim 1, wherein the entity categories comprise equipment, components, abnormal phenomena, abnormal reasons, processing measures, fault types, work projects, process requirements, working principles and key links, and the relation categories comprise occurrence, due, inclusion, belonging and adoption.
  3. 3. The method for constructing the fault knowledge graph of the hydraulic turbine speed regulating system according to claim 1, wherein the named entity recognition model comprises RoBERTa layers, a two-way long-short-term memory network layer and a conditional random field layer which are connected in sequence, and the relationship recognition model comprises RoBERTa layers and the two-way long-short-term memory network layer which are connected in sequence.
  4. 4. The method for constructing a fault knowledge graph of a hydraulic turbine speed regulation system according to claim 1, wherein the fine tuning method of the large language model is as follows: constructing a plurality of instruction samples based on the basic triplet set, and performing quality control processing on each instruction sample to construct a fine tuning data set, wherein the instruction samples comprise instruction texts and input and output texts; And according to the fine tuning data set, performing parameter optimization on the large language model by adopting a low-rank self-adaptive fine tuning technology to obtain a fine-tuned large language model.
  5. 5. The method for constructing a fault knowledge graph of a hydraulic turbine governor system according to claim 4, wherein said performing a quality control process on each of said instruction samples includes: Carrying out data cleaning on the instruction samples through a predefined cleaning rule to obtain cleaned instruction samples; carrying out confidence score on the data fluency, correlation and fact accuracy of the cleaned instruction samples, and removing samples with confidence scores lower than a preset confidence threshold value to form a data pool to be checked; Carrying out layered random sampling and correction on the samples in the data pool to be checked until the sampling passing rate reaches a preset passing rate threshold value, and obtaining a checked data pool; And calculating the key dimension distribution of the samples in the verified data pool, and expanding the samples according to the dimension distribution until the key dimension distribution is balanced.
  6. 6. The method for constructing a fault knowledge graph of a hydraulic turbine speed regulation system according to claim 1, wherein the step of performing knowledge updating on the basic triplet set through the reasoning triplet set to obtain a structured triplet set includes: Respectively calculating the semantic similarity between each inference triplet in the inference triplet set and each basic triplet in the basic triplet set, and correcting the basic triplet through the inference triplet if the semantic similarity is larger than a preset semantic similarity threshold value to obtain a corrected basic triplet set; And carrying out bidirectional matching on the corrected basic triplet set and the reasoning triplet set, reserving head entity-relation-tail entity triples consistent with the bidirectional matching, and removing noise and free triples to obtain a structured triplet set.
  7. 7. The method for constructing a fault knowledge graph of a hydraulic turbine speed regulation system according to claim 6, wherein the storing the structured triplet set in a graph database and performing knowledge fusion in the storing process to obtain the fault knowledge graph includes: mapping the head entity-relation-tail entity triples in the structured triplet set into nodes and edges in the graph database through the MERGE operation of the graph database; And calculating the comprehensive similarity between the existing nodes in the graph database and the entities of the head entity-relation-tail entity triples, and if the comprehensive similarity is larger than a preset comprehensive similarity threshold, fusing the entities of the head entity-relation-tail entity triples with the existing nodes to construct a fault knowledge graph.
  8. 8. The utility model provides a hydraulic turbine speed governing system fault knowledge map construction system which characterized in that includes: the data acquisition module is used for acquiring a fault text of the hydraulic turbine speed regulation system and predefining entity types and relation types in the fault text; the model training module is used for pre-training a named entity recognition model and a relationship recognition model based on the hybrid neural network according to the entity category and the relationship category; the knowledge extraction module is used for carrying out knowledge extraction on the fault text through the named entity recognition model and the relation recognition model to obtain a basic triplet set; the knowledge updating module is used for updating knowledge of the basic triplet set through the reasoning triplet set to obtain a structured triplet set; and the knowledge storage module is used for storing the structured triplet set into a graph database, and carrying out knowledge fusion in the storage process to obtain a fault knowledge graph.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the hydroturbine governor system fault knowledge graph construction method of any one of claims 1 to 7 when executing the program.
  10. 10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the hydraulic turbine governor system fault knowledge graph construction method according to any one of claims 1 to 7.

Description

Method and system for constructing fault knowledge graph of water turbine speed regulating system Technical Field The invention belongs to the technical field of knowledge graphs, and particularly relates to a method and a system for constructing a fault knowledge graph of a water turbine speed regulation system. Background The water turbine speed regulating system is used as core control equipment in power production, and the running state of the water turbine speed regulating system directly influences the stability of a hydroelectric generating set and the frequency quality of a power grid. Once the fault occurs, not only the unplanned shutdown of the unit can be caused, but also the chain reaction can be initiated, and the safety and stability of the whole power system are threatened. Therefore, the intelligent operation and maintenance and the accurate fault diagnosis of the water turbine speed regulating system are realized, and the intelligent operation and maintenance and the accurate fault diagnosis become key links for guaranteeing the safe production and improving the operation benefit in the hydropower industry. Along with the development of informatization and digitalization of the hydropower industry, a large amount of text data in the forms of fault records, maintenance reports, operation logs and the like are accumulated in long-term operation, and the text data contains rich experience knowledge of fault modes, reasons, disposal measures and the like. However, most of these text data are in unstructured or semi-structured form, are scattered in different systems and history files, are not yet systematically mined, integrated and utilized, and are difficult to directly serve fault analysis, decision support and knowledge inheritance. Currently, knowledge graph technology provides an effective way for realizing systematic organization and intelligent application of domain knowledge. In the electric power field, research attempts have been made to construct a spare operation and maintenance knowledge system by using a knowledge graph structure, but the knowledge graph structure is mainly concentrated on primary equipment such as a transformer, and the research on fault knowledge graphs of complex electromechanical liquid coupling systems such as a hydraulic turbine speed regulating system is still relatively lacking. The traditional fault knowledge graph construction method is mostly based on rules or statistical models, and has the problems of high dependence on quality of marked data, high calculation complexity, difficulty in adapting to fine-grained semantic relations in the professional field and the like. For example, some methods use improved sequence labeling models to extract joint entity relationships or map entity relationships to graph structures, and research has been conducted on extracting power equipment entities and relationships based on lightweight pre-training models (such as BERT) in combination with Pipeline (Pipeline) modes, and finally the power equipment entities and relationships are stored and applied in a graph database. Although the development of the electric power knowledge graph is promoted to a certain extent by the knowledge graph construction method, challenges such as dense technical terms, various fault descriptions, strong relationship implications and the like are faced when a large amount of unstructured data is oriented to a fault text of a water turbine speed regulation system, and time sequence and causality characteristics in a text context are often ignored, so that the integrity and accuracy of knowledge extraction are limited. In addition, the traditional method is low in calculation efficiency when processing massive historical texts, and engineering real-time requirements are difficult to meet. Therefore, aiming at the characteristics of the fault knowledge of the water turbine speed regulating system, a knowledge map construction method which can deeply fuse the prior knowledge in the field and give consideration to the knowledge extraction efficiency and accuracy is explored, and the method has important significance for realizing effective precipitation, intelligent inquiry and auxiliary decision-making of the fault knowledge and is also a key basis for promoting the intelligent transformation of the operation and maintenance of the hydropower industry. Disclosure of Invention The invention provides a method and a system for constructing a fault knowledge graph of a water turbine speed regulating system, which are used for solving the problem that fault text data of water turbine speed regulating system equipment in the prior art is not effectively mined and utilized, constructing the fault knowledge graph capable of more accurately carrying out fault identification, and providing accurate fault identification and decision support for the water turbine speed regulating system. In a first aspect, the present invention provides a method for con