A Classification Model Of Power Equipment Defect Record Texts Based On Multi-head Attention RCNN Network

With the increasing scale of power system,the number of equipment has doubled,which greatly increases the inspection workload.At the same time,the increase of equipment operation time and operation frequency increases the probability of equipment defects.If these defects are not found and eliminated in time,it will endanger the operation safety of power system.According to the regulations,the equipment defects are divided into three levels.Each level of defects has different risks to the power grid operation,and the time limit for eliminating defects is also different.Power equipment defects are recorded and reported in the form of text.At present,a large number of reported equipment defect classification work still needs to be carried out manually,which is timeconsuming,labor-consuming,inefficient and limited by the knowledge reserve of transportation and inspection personnel.It is often in an embarrassing situation that it is difficult to accurately judge,and the accuracy of classification is affected,which affects the progress of defect elimination.There are a large number of idle historical transmission and transformation equipment defect records text in the power grid defect management system.If the natural language processing method is used to mine and analyze the defect records to achieve automatic classification,the speed and accuracy of defect grade judgment can be improved,and the maintenance efficiency can be improved.The description of electrical equipment defects originates from the colloquial expression of inspectors,without fixed format and non-standard expression,which is not uniform for the same defects.And different from the ordinary Chinese text,the defect description text is highly professional,so it is difficult to accurately understand and use.Because the traditional expert system can’t accurately understand the semantics of sentences by making rules,how to make full use of the historical defect records text to quickly classify and grade the new equipment faults has become an urgent problem.In view of this,this paper proposes a method of power equipment defect text classification based on Multi-Head Attention recurrent convolutional neural networks(MAT-RCNN).Firstly,the text is preprocessed,and the proper noun dictionary and stop word list are established to realize the correct segmentation of words and the elimination of redundant information.Taking the segmented text as corpus,the word vector of each word is trained by Word2 vec model.Then,considering that the word vector will not change with the context and RCNN network can not automatically focus on key words,we introduce the Multi-Head Attention mechanism to mine the internal relations of the text from multiple dimensions,and allocate the corresponding weights of different words to make the output word vector change according to the context,and highlight the important information by weighting,so that RCNN network can focus on the key information and integrate RCNN network,the optimization of network structure makes it better integrated with Multi-Head Attention mechanism.Finally,the Multi-Head Attention RCNN power equipment defect text classification model is constructed,and on this basis,the automatic grading of equipment defects is realized.After experimental analysis,compared with the traditional machine learning text classification model and CNN,RNN and other deep learning text classification models,the classification effect of the model is evaluated from multiple classification comprehensive evaluation index MF1,classification accuracy(ACC)and model time-consuming.The experimental results show that the proposed method is superior to RNN and other conventional methods in semantic learning ability,feature extraction effect and classification effect.Therefore,the method proposed in this paper makes full use of the historical defect data to realize the rapid classification and grading of power equipment defects text,improves the accuracy of grading,reduces the manpower allocation,and effectively improves the efficiency of power grid operation and maintenance,which has strong practical value.