Partial discharge fault identification method for GIS equipment based on improved deep learning

Abstract

Aiming at the problems of large consumption of computational resources and insufficient data feature extraction in the current partial discharge fault identification process of GIS equipment, a partial discharge fault identification method of GIS equipment based on improved deep learning is proposed. Firstly, the audio information of GIS equipment is filtered by a simple power normalised cepstral coefficient (SPNCC). Secondly, the spatial correlation between audio data streams is obtained by a convolutional neural network, the temporal correlation of audio is obtained and the next time slice data stream is predicted by using bi‐directional long short‐term memory (BiLSTM) network, and the attention mechanism is designed to extract deeper data features. Finally, the partial discharge fault identification model of GIS equipment based on improved SPNCC‐CNN‐BiLSTM‐Multi‐att is established, which improves the accuracy of the partial discharge identification method of GIS equipment. Experiments show that when the number of iterations is 100, the accuracy, recall, and F1 value of the proposed GIS equipment partial discharge fault recognition method on the dataset are 0.876, 0.812, and 0.843, respectively.