Research On Automated Construction Technology Of Knowledge Graph For Power Transformer

Knowledge graph is also known as scientific knowledge graph,It has four characteristics of dynamic,spatiality,relevance and knowledge dependence.It can not only create visualization model and presents knowledge as the form of directed graph and get the relationship between them,but also complete the accurate query of large amount of information in a very short time through the powerful query function of computer search engine,and make statistical analysis of scattered information.Power transformer is a kind of important equipment in power system,it is also one of the most complex and expensive equipment and the key of energy transformation and transmission.The construction of power transformer knowledge graph can effectively organize the power transformer related knowledge,and realize the functions of knowledge storage,intelligent search,auxiliary decision-making and so on.This paper intends to research on automated construction technology of knowledge graph for power transformer,and build a knowledge graph for oil-immersed transformer to implement intelligent query and associated search of transformer-related knowledge,so as to lay a foundation for the establishment of intelligent operation and maintenance system of transformer.Based on the principle of automatic word segmentation algorithm for Chinese text in statistical language model,this paper establishes a statistical model of mutual information,reads professional corpus and general non-professional corpus respectively,and obtains their respective segmented-word lists.The word segmentation results obtained by the above unsupervised word segmentation model include some non-professional vocabulary and some unwanted non-word components,so we optimize the algorithm,adding product of left entropy and right entropy as an evaluation criteria for word aggregation,and then comparing the power-related word segmentation results with non-professional corpus’ s word segmentation results,which achieves the automatic construction of power-related professional lexicon.In the process of word segmentation before triple extraction,makes use of the built power-related professional lexicon as entity extraction result and to increase the accuracy and professionalism of word segmentation,uses the vector encoding method of combining character&word-mixing Embedding and position Embedding to incorporate semantic information into vector encoding,Then through the triple extraction model based on Dilated Gated Convolutional Neural Network combined with Self-Attention mechanism,we get the triples with correct entity and relationship.After getting the set of triples of power transformer through Triple extraction,this paper use the Neo4 j graph database to save the knowledge graph of power transformer,realized the visualization of transformer knowledge graph;The intelligent query and association search of transformer-related knowledge is realized based on py2neo;and the knowledge embedding technology based on Trans R algorithm is used to realize the inference of power transformer fault analysis and the query of treatment measures.