Research On Knowledge Graph Construction And Applications Of Electrical Equipment Defect

With the comprehensive development of constructing smart grid,improving abilities of electricity related personnel in operation,maintenance and decision-making have become a research hotspot in the electric power field,so as to eliminate faults in time to ensure the safe and stable operation of the power system.The electrical equipment malfunction text related to equipment defects contains key information such as fault types,fault causes and equipment defects elimination method.However,such electrical equipment malfunction text has the difficult-processed issues of large volume,multi-source heterogeneity,as well as cluttered and redundant content.At present,there lacks efficient integration and utilization methods.In response to these issues,this paper uses natural language processing technology to conduct researches on both text quality improvement and information extraction,and then builds a knowledge graph of electrical equipment defects.The progress is presented as follows.Firstly,the quality improvement of structural malfunction text of electrical equipment is achieved.Above all,the malfunction text is preprocessed,and then the BI-LSTM algorithm is selected regarding the missings and errors of the malfunction text.Then,according to the characteristics of the malfunction text,UCNN is integrated to extract the semantic features between text words,and the Leaky Relu activation function is added to avoid the gradient disappearing.Consequently,the Foscal Loss function is reconstructed to solve the problem of data set imbalance.Finally,above three strategies are combined to propose an improved BI-LSTM algorithm,which improves the text quality of both defect types and defect levels in electrical equipment malfunction text.The experimental results show that the recall rate of the improved algorithm are increased by 9.07% and 3.58% respectively,compared with BI-LSTM.That is,the text quality of defect type and defect level is improved.Then,the research on information extraction of the unstructured malfunction text regarding electrical equipment is conducted.BERT is selected as the named entity recognition algorithm.Based on BERT,a BI-LSTM layer is utilized to further extract text semantic context information,and CRF is used to replace the output layer of BERT,which overcomes the local optimal problem of preferred word labels.Finally,combining the above two strategies,an improved BERT algorithm is proposed.And based on the rules,an entity relationship extraction process is designed to extract the entity relationship from above malfunction text.The experimental results show that the improved BERT algorithm has achieved the recall value of 91.28%,96.69%,99.03%,91.28%,96.69%,99.03% on seven defect entity,respectively.Consequently,the corresponding F1 values are 91.91%,98.57%,96.89%,98.02% and 95.52%.Compared with BERT,the overall precision and recall of entity extraction are improved by 0.94% and 0.95%,respectively.Finally,a knowledge graph of electrical equipment defects is constructed.The Neo4 j database is selected to store data,and a collective entity fusion strategy based on similarity propagation is proposed to realize knowledge fusion.Through the realization of the knowledge storage and update strategy,the knowledge graph of electrical equipment defects is formed.On this basis,the intelligent question answering based on defect knowledge graph is realized.Furthermore,the improved BERT algorithm based on entity recognition of questions is achieved,together with the improved BI-LSTM algorithm based intent recognition,and the templates based on query sentences generation.The test results show that the question answering system constructed in this paper can return targeted fault diagnosis and elimination strategies,and realize efficient operation and maintenance decisions of the power system,which has a superior application prospect.