Power Equipment Defect Text Mining Based on New Word Discovery and Feature Fusion

• Published in: 2021 International Conference on Power System Technology (POWERCON) pp. 1948 - 1952
• Main Authors: Sun, Lintao, Liu, Changbiao, Li, Wenyan, Zhang, Xuanzhe, Ai, Yunfei, Guo, Chuangxin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 08.12.2021
• Subjects: convolutional neural network; Deep learning; feature fusion; Maintenance engineering; power equipment; Semantics; Solid modeling; Text categorization; Text mining; Transforms
• ISSN: 2642-6226
• DOI: 10.1109/POWERCON53785.2021.9697625

With the intelligent development of power grid equipment operation and maintenance, how to effectively use a large number of defect text records has become an important issue. Since the text is complex unstructured data, it is difficult to effectively mine defect information. To solve this problem, the new word discovery method of solidification degree-degree of freedom is used in text preprocessing to extract the word features in the defective text; further, the word2vec word vector model is used to map the word features to a multi-dimensional vector space; finally based on feature fusion Constructed an attention mechanism to optimize the convolutional neural network defect text classification model. The analysis of the calculation example makes a comprehensive comparison and analysis of the attention mechanism optimized convolutional neural network based on new word discovery and feature fusion and the traditional neural network model. The proposed method has better semantic learning ability than the traditional deep learning method and can improve the classification accuracy, which is conducive to fully mining the defect text information.