Research On The Construction Method Of Fault Knowledge Graph Of Substation Primary Equipment

With the rapid development of ultra-high voltage(UHV)power grids and new energy sources,troubleshooting primary equipment in substations requires personnel with advanced professional capabilities.However,presently,this task largely depends on the staff’s knowledge reserves and long-term experience accumulation,and entails repeated memorization and retrieval of vast amounts of information in textual form.Due to the lack of efficient and accurate power field knowledge extraction,organization and management technology,there is a problem that the accuracy and standardization of fault handling cannot be guaranteed.Aiming at this problem,a method for constructing a fault knowledge graph of substation primary equipment is proposed.This method will focus on analyzing the relevant data associated with primary equipment fault processing.It will leverage advanced technologies such as natural language processing and deep learning to identify and extract the entities and relationships embedded within the fault corpus.Once this information is obtained,the method will construct a comprehensive substation primary equipment fault knowledge graph.The graph will include primary device topology,device fault data,and fault handling methods,providing engineers with accurate and comprehensive fault diagnosis and processing solutions.This research project consists of three main challenges that need to be addressed.Firstly,the project needs to identify how to extract professional vocabulary from unstructured equipment failure texts and create domain-specific dictionaries containing embedded information.Secondly,the project needs to identify methods to automatically and accurately identify entity fragments related to electric power from equipment failure texts.These identified entity fragments will then be categorized by their types,and used as nodes in the knowledge graph.Lastly,the project needs to determine how to extract relationships between the identified entities and construct a substation primary equipment fault knowledge graph.This knowledge graph will represent the relationships between primary device topology,device fault data,and fault handling methods.By addressing these three key issues,the project aims to provide engineers with a comprehensive and accurate fault diagnosis and processing solution.To address the aforementioned three key issues,the research project involves the following three studies.Firstly,the project will clean the equipment failure corpus text and apply the N-Gram model,information entropy,and mutual information techniques to mine professional vocabulary.The word embedding model will be used to train a domain-specific embedding dictionary.Secondly,the project will propose a BERT-FLAT-CRF model that incorporates the domain-specific embedding dictionary information and introduces relative position codes.This model will enable the identification of entities and their types in the equipment fault text based on the thought structure of entity recognition.Lastly,the project will propose a fault knowledge graph construction method based on BiLSTM-ATT.This method will be built on top of the entity recognition task and will enable the construction of a comprehensive substation primary equipment fault knowledge graph.To verify the effectiveness of the proposed research method,the study utilized relevant corpus data such as the maintenance manual for primary equipment failure and troubleshooting questions and answers as experimental data.Firstly,the study constructed a device field embedding dictionary and combined the BERT-FLAT-CRF model with the dictionary information to identify entities in the equipment fault corpus.Comparative experiments were then conducted,and the results demonstrated that the model had the best effect on entity recognition.The precision rate,recall rate,and F1 value reached 86.2%,81.11%,and 83.58%,respectively.Secondly,the study utilized the BiLSTM-ATT model to extract the relationship between entities and conducted comparative experiments with three other models.The results showed that the model was superior to other models in all indicators in the relation extraction task of a device domain corpus.As a result,the model was able to automatically and accurately extract the relationship between entities,enabling the construction of a fault knowledge graph of substation primary equipment.