Method for Continuous Construction of Incremental Thermal Power Plant Fault Knowledge Graph
LIU Yunjia
LIU Chen
Abstract：Given the diversity of equipment and large volume of rapidly updating fault documentation in thermal power plants,it's particularly important to efficiently extract knowledge from documents and integrate it into existing knowledge graphs while ensur-ing quality.To this end,an end-to-end framework for continuous construction of knowledge graphs is proposed.This framework cov-ers key steps such as text preprocessing,entity-relation extraction,knowledge fusion,and graph construction.In the entity-rela-tion extraction stage,a combination of pre-training models and deep learning models BERT-BiLSTM-GP(Global Pointer)is pro-posed to achieve joint extraction of entities and relations.In the knowledge fusion stage,to address the issue of non-standard fault text descriptions and sparse fault features,a model called KeyCoSENT(KeywordsAttention-BiLSTM-CoSENT)is proposed.This model introduces a keyword attention mechanism and multiple feature fusion by identifying keywords in sentences through a thermal power plant domain dictionary.The improved model is validated on a self-built thermal power plant fault dataset,increasing the F1 score by 1.1％compared to the original model,and by 4.7％compared to the SBERT model(Sentence-Bert).This method of con-structing high-quality knowledge graphs holds significant importance for better managing and maintaining thermal power plant equipment,and also provides an efficient and viable method for continuous knowledge construction.
Keywords：knowledge graphjoint entity-relation extractionattention mechanismCoSENT
Publication Date：2025-08-20
Online Publishing Date：2025-12-12(First online date of this platform, not the publication date of the document)
Pages：8( 2063-2069,2094 )