Research on Feature Extraction and Identification of Acoustic Defects of GIS Equipment Based on Deep Learning

Early detection and identification of internal defects in Gas Insulated Switchgear (GIS) equipment is of great significance for preventing power accidents. This study focuses on deep learning based acoustic defect feature extraction and recognition technology for GIS devices, aiming to improve the accuracy and efficiency of defect detection. In the study, a deep learning model combining Convolutional Neural Network (CNN) and Long Short Term Memory Network (LSTM), namely CNN-LSTM model, was adopted to fully utilize the advantages of CNN in image feature extraction and the ability of LSTM to capture long-term dependencies in processing time series data. By converting the acoustic signals of GIS equipment into spectrograms, CNN can automatically extract the local features of acoustic signals, while LSTM learns the temporal dynamic features of these features. Based on the features extracted from CNN-LSTM model, a defect identification algorithm is designed, which maps the feature vectors to the probability distribution of defect types through full connection layer and softmax function to realize automatic defect identification. Experimental results show that CNN-LSTM model can maintain high recognition accuracy in a variety of complex scenes, especially in mechanical noise background and low noise intensity. The application of this technology can promote preventive maintenance, reduce unplanned downtime, improve the stability of power grid, and assist in locating the location of defects, providing accurate guidance for fault diagnosis and maintenance.