Inspection Text Classification of Power Equipment Based on TextCNN

AbstractA large number of text and reports about the power equipment are generated in power system, which consist of implicit information of operation condition and insulation status. With the development of convolutional neural network (CNN), the inspection text can be analyzed intelligently to improve the reliability of power system. In order to extract valuable information from inspection text for state evaluation of power equipment in local area, an information extraction model for inspection text based on TextCNN is proposed, improved and verified. First, the feature embedding of inspection text were performed by Word2Vec method. Secondly, the corpus were augmented with back translation method. Then, the TextCNN was adopted to classify the risk level of the power equipment or area involved in the inspection text. Finally, the classification results from the model were evaluated by classification accuracy, F1 score, confusion matrix and compared with the model based on BiLSTM and RCNN. The results demonstrated that the performance of TextCNN was the best among the three models on augmented dataset by back translation method with ACC and F1 scores of 0.9087 and 0.9099, respectively, which is the most suitable model among these three for recognition and classification of inspection text of power equipment.KeywordsInspection textInformation extractionBack translationTextCNN