Analysis of Transformer Audible Noise Based on Redundant Convolutional Encoder Decoder

During the field acquisition of transformer acoustic signals, it is easy to be interfered by the outside world. The collected acoustic signals are often mixed with many noise interferences from non-transformer bodies, especially when the acoustic signals are interfered by various noises or even submerged in extremely harsh environments. The measuring instrument cannot effectively carry out preprocessing, which will cause interference to the transformer acoustic signal online monitoring system, and may lead to the misjudgment of the transformer operating state. In this paper, a method named Redundant Convolutional Encoder Decoder (R -CED) is proposed to de-noising transformer-containing noise signals. Short time Fourier transform (STFT) is used to extract the features of clean sound signals and noisy signals. The transformer acoustic signal denoising model based on R -CED is constructed by reasonably selecting the number of convolutional network layers and the size of the convolutional kernel. Finally, the denoising model is used to denoise the acoustic signal of a 72.5 kV converter transformer, the results show that the model has a good denoising effect, and the method has reference significance for the transformer operating state monitoring based on acoustic signals.