Recognition of short-circuit discharge sounds in transmission lines based on hybrid voiceprint features

In recent years, with the rapid development of deep learning, more and more fields have introduced it to solve hot and difficult problems. For example, voiceprint and deep learning algorithm are combined to realize transformer status monitoring in power substation. In this work, We introduced deep learning voiceprint recognition technology into the power transmission scenario and proposed a method for detecting abnormal short-circuit discharge sound in transmission lines based on hybrid voiceprint features. First, basic voiceprint features are extracted from the collected sounds, and then the hybrid voiceprint features of the sounds are obtained based on the proposed hybrid voiceprint construction method. Finally, the hybrid voiceprint features are used to train the ResNet neural network model with SE module and Attentive Statistics Pooling (ASP). It realizes continuous and effective detection of abnormal short-circuit discharge in transmission lines through the voiceprint characteristics of sound. Experimental results show that the performance is good on the short-circuit discharge sound dataset of self-built transmission lines. It can effectively improve the safety level of power grid equipment operation.