Research on Transformer Condition Recognition Based on Acoustic Signal and One-dimensional Convolutional Neural Networks

Abstract

To solve the problem of transformer condition recognition, this paper propose a transformer condition recognition method based on acoustic signal and One-dimensional Convolutional Neural Networks(1D-CNN). In order to verify the effectiveness of 1D-CNN algorithm in the field of transformer condition recognition, a platform for acquisition of acoustic signals from a 500 kV transformer is built to carry out the acoustic signal acquisition test for four conditions of the transformer. The acoustic signal data sets are also made, and the 1D-CNN algorithm is used to calculate the recognition accuracy of the transformer conditions. According to test results, 1D-CNN algorithm, as a new structure of deep-learning algorithm, can properly classify acoustic signals of the transformer, and its classification accuracy is higher than those of FFT-BP, SVM, FFT-SAE and other algorithms. In order to explore the internal mechanism of 1D-CNN algorithm, in this paper, a t-SNE visual analysis is also conducted to reveal the performance of 1D-CNN algorithm.