基于改进深度残差网络的GIS局部放电在线监测白噪声干扰抑制

针对GIS局部放电(partial discharge, PD)监测中背景白噪声较多、GIS局部放电信号干扰较大的问题，应用改进深度残差网络设计一种新的GIS局部放电在线监测白噪声干扰抑制方法。进行局部放电在线监测中白噪声、局部放电脉冲信号的多尺度特性分析，在局部放电脉冲染噪信号中提取白噪声信号。加入感知损失，设计由生成图像网络与损失网络构成的改进深度残差网络，对白噪声信号波形图像实施超分辨率重建。通过SN-EMD算法提取白噪声信号波形图像的模态域特征。通过构建复小波滤波器组，对模态域特征实施滤波处理，实现GIS局部放电在线监测中的白噪声干扰抑制。实验测试结果表明，设计方法去噪后的信噪比最高可达97.22 dB,干扰抑制前后信号的幅值相对误差最高可达63.20 dB,干扰抑制前后信号相关系数一直大于0.75,完成GIS局部放电在线监测白噪声干扰抑制。

White Noise Interference Suppression of GIS Partial Discharge On-line Monitoring Based on Improved Depth Residual Network

Aiming at the problems of more background white noise and greater interference of GIS partial discharge signal in GIS partial discharge monitoring, a new white noise interference suppression method for GIS partial discharge on-line monitoring is designed by using the improved depth residual network. The multi-scale characteristics of white noise and PD pulse signal in PD on-line monitoring are analyzed, and the white noise signal is extracted from the PD pulse noise signal. Adding perceptual loss, an improved depth residual network composed of generated image network and loss network is designed to reconstruct the waveform image of white noise signal with super-resolution. The modal domain features of white noise signal waveform image are extracted by sn-emd algorithm. By constructing complex wavelet filter bank and filtering the characteristics of modal domain, the white noise interference suppression in GIS partial discharge on-line monitoring is realized. The experimental test results show that the signal-to-noise ratio of the design method after denoising can reach 97.22 dB, the relative error of the signal amplitude before and after interference suppression can reach 63.20 dB, and the signal correlation coefficient before and after interference suppression has been greater than 0.75. The white noise interference suppression of GIS Partial Discharge Online Monitoring is completed.