Transformer Vibration Analysis Based on Double Branch Convolutional Neural Network

Abstract The power transformer is one of the important pieces equipment in grid system, and its normal operation related to safety reliability whole system. There are many factors influencing vibration operation, characteristics complex, so it difficult be directly used for state analysis. This paper proposes a method signal analysis based on continuous wavelet time-frequency graph. segmented samples signals selected by time-domain sample segmentation method, time sequence transformed transform obtain two-dimensional graph input into two-branch convolutional neural network, classification given features extracted from network. simulation data measured multiple measuring points shows that proposed has an average recognition accuracy 98.3%. work this can provide reference transformer.