基于F-SAE网络的GIS设备局部放电噪声抑制方法  

Noise suppression method of partial discharge for GIS equipment based on F-SAE network

作　　者：张彦军 徐肃 张瑞强 刘轶 李智玲 卢霄霞 余传祥[2] ZHANG Yanjun;XU Su;ZHANG Ruiqiang;LIU Yi;LI Zhiling;LU Xiaoxia;YU Chuanxiang(Baotou Power Supply Branch of Inner Mongolia Electric Power(Group)Co.,Ltd.,Baotou 014000,China;State Key Laboratory of Power Transmission Equipment Technology(Chongqing University),Chongqing 400044,China)

机构地区：[1]内蒙古电力(集团)有限责任公司包头供电分公司,内蒙古包头014000 [2]输变电装备技术全国重点实验室(重庆大学),重庆400044

出　　处：《电工电能新技术》2025年第1期118-128,共11页Advanced Technology of Electrical Engineering and Energy

基　　金：内蒙古电力(集团)有限责任公司科技项目(nmgdlkjxm2021017)。

摘　　要：在进行气体绝缘金属封闭开关设备(GIS)早期局部放电(PD)监测时,信号容易受到噪声干扰,从而造成对PD故障的误判。为减少噪声对PD信号的影响,本文提出一种基于分数阶随机自动编码器(F-SAE)的噪声抑制方法。该方法以自动编码器为核心,采用Tchebichef矩函数转换一维PD信号的正交特性进行噪声抑制。该方法在网络反向传播过程中引入分数阶梯度下降以提升网络层之间的权重信息更新效率,并通过随机奇异值变换压缩权重信息,共同提升F-SAE网络对PD信号的降噪性能。本文对真实GIS设备下0.5 dB、1.5 dB和5 dB的三种染噪PD信号进行降噪,其中0.5 dB染噪PD信号去噪后的信噪比分别为5.48 dB、6.28 dB和6.92 dB以及均方根差百分比分别为42.51%、57.28%和58.14%。并且降噪性能皆优于所对比的小波变换算法和去噪自动编码器。When performing partial discharge(PD) monitoring of gas insulated metal-enclosed switchgear(GIS) in the early stages,the signal is susceptible to noise interference,which can cause misjudgment of PD faults.To reduce the influence of noise on PD signals,a noise suppression method based on fractional-stochastic auto-encode(F-SAE) is proposed by the paper.The AE played a core role in the method.And the Tchebichef moment function is used to transform the orthogonal characteristics of one-dimensional PD signals for noise suppression.The method of fractional-order gradient descent is introduced to improve the efficiency of updating weight information between network layers in the network backpropagation.In addition,the weight information of network is compressed by the random singular value decomposition,which jointly improves the noise reduction performance of the F-SAE network for PD signals.In this paper,three kinds of noise-dyed PD signals of 0.5 dB,1.5 dB and 5 dB are denoised under real GIS equipment,and the signal-to-noise ratios of the denoised 0.5 dB PD signals are 5.48 dB,6.28 dB,6.92 dB,and the percentage root-mean-square differences are 42.51%,57.28%,and 58.14%,respectively.And the denoising performances are better than that of the compared wavelet transform algorithm and denoising auto-encoder.

关 键 词：气体绝缘金属封闭开关设备 局部放电 噪声抑制 分数阶随机自动编码器 

分 类 号：TM855[电气工程—高电压与绝缘技术]