Research And Application Of Key Technologies Of Knowledge Graph For Substation Inspection Operations

With the rapid growth of power grid business,higher requirements are put forward for the operation and inspection of substation equipment.At present,the content of substation equipment operation and inspection is complex and highly dependent on the professional and technical knowledge and experience accumulation of operation and inspection personnel.Making full use of the unstructured text accumulated in the power field,constructing,mining and utilizing the equipment knowledge with rich correlation in the power field,and pushing the auxiliary knowledge related to substation operation and inspection to the operation and inspection staff,is a work with important research significance and application value.Focusing on the specific requirements of substation operation and inspection,this paper studies the key technologies of knowledge extraction in the construction of domain knowledge graph,and realizes the knowledge push system for transformer operation and inspection based on the knowledge graph.so as to provide defect operation and inspection auxiliary knowledge for operation and inspection personnel.The specific research work is as follows:(1)In the field of knowledge extraction in substation operation and inspection,aiming at the task of named entity recognition,a bidirectional long-term and short-term memory network model based on the fusion of transformer domain vocabulary feature information is proposed.Combined with conditional random field to learn sequence label constraints,the performance of entity recognition is optimized.Aiming at the task of entity relation extraction,a method of entity relation extraction based on pre-training model fine tuning is proposed by using pre-training model instead of complex artificial feature extraction process,and good extraction effect is achieved in the experiment.(2)In the aspect of knowledge push for substation operation and inspection,a text recall method based on knowledge graph embedding is proposed,which combines defect scenarios and operation and inspection task information to obtain candidate texts.In the aspect of text relevance ranking,this paper studies and implements the relevance ranking method based on multi-model fusion.Specifically,the performance of the model is enhanced by fusing multiple text matching models,and the push text is sorted according to the correlation score.Besides,Experiments show that the graph embedding method has a better classification result and can be used to optimize the candidate recall text.(3)In the aspect of knowledge push system of substation operation and inspection,this paper analyzes the specific operation and inspection tasks and models the task process,and designs and implements the knowledge pushing system of substation operation and inspection based on the knowledge graph of substation operation and inspection,and verifies the knowledge pushing in the operation and inspection case.