Research And Implementation Of Grid Equipment Defect Case Analysis Based On Field Knowledge

With the development of intelligent technology,grid system has entered the stage of artificial intelligence.In power grid industry,a large number of equipment defect information is recorded in the form of text.Therefore,a method of extracting and analyzing the potential defect laws from those cases is significant for preventing faults in production and improving robustness of power system.Due to the strong professionalism in power grid field,general information extraction and text analysis methods are difficult to use directly.Therefore,how to organize the field knowledge reasonably to serve case text analysis,how to design a method of defect correlation analysis and defect level identification according to expression characteristics and structural patterns,how to modularize the implementation of those functions on a distributed framework to deal with the big data situation in practice,are all key issues in this paper.This paper designs and implements a method of extraction and analysis for grid equipment defect cases according to the above problems.And integrates those functions into the Power Equipment Data Analysis Platform based on Spark,achieving efficient analysis of large-scale power cases.In extraction and analysis stage,the target is divided into word-type information and sentence-type information.For word-type information,the field ontology knowledge base model is designed and constructed firstly for the structural representation of equipment and defect information.Then design the case text semantic framework.And with the introduction of slot filling,the key word extraction algorithm is designed and implemented.The case text with complex sentence structure can be extracted effectively.For sentence-type information,the category feature analysis is firstly carried out,then classify and extract the sentences with bag-of-word method and support vector machine(linear kernel).For word-type information,the association rules are used for defect correlation analysis.Firstly,the FP-Growth algorithm is distributed implemented based on Spark.In the Growth stage,the shared prefix and single-branch structure are optimized by splitting and reorganizing FP-Tree,which improve the performance of the algorithm generally.Then apply the algorithm to word-type information and mine defect association laws.For sentence-type information,the F-GCNN algorithm is designed and implemented considering the field professional characteristics(especially a large number of composite field vocabulary),to classify equipment defect level automatically.Fasttext is used to represent case text accurately.GCNN network realizes efficient classification of defect level.Finally,those extraction and analysis functions are integrated into the Power Equipment Data Analysis Platform based on Spark.