基于SVD和低秩RBF神经网络的局部放电信号提取方法  

Partial Discharge Signal Extraction Method Based on SVD and Low Rank RBF Neural Network

作　　者：杨晓丽[1] 黄宏光[1] 舒勤[1] 张大堃 周电波 YANG Xiaoli;HUANG Hongguang;SHU Qin;ZHANG Dakun;ZHOU Dianbo(School of Electrical Engineering,Sichuan University,Chengdu 610065,China;Chengdu Ruichi Technology Co.,Ltd.,Chengdu 610023,China;Electric Power Research Institute of State Grid Sichuan Electric Power Company,Chengdu 610072,China)

机构地区：[1]四川大学电气工程学院,成都610065 [2]成都芮驰科技有限公司,成都610023 [3]国网四川省电力公司电力科学研究院,成都610072

出　　处：《高电压技术》2021年第10期3608-3616,共9页High Voltage Engineering

摘　　要：局部放电(partial discharge,PD)特高频(ultra high frequency,UHF)信号检测过程易受到白噪声和周期性窄带干扰的严重影响。为有效提取PD UHF信号、抑制干扰,提出一种基于奇异值分解(singular value decomposition,SVD)和低秩径向基函数(radical basis function,RBF)神经网络的去噪方法。首先,将染噪局部放电信号构造为Hankel矩阵,并奇异分解到特征矩阵空间;然后,把特征矩阵中奇异值突变点设为阈值,以去除窄带干扰;最后,采用RBF神经网络逼近去干扰后的PD信号,并采用Gaussian窗滤波以提取局放信号。所提方法与逆向分离(reverse separation,RS)和形态学小波综合滤波器(morphology wavelet filter,MWF)进行对比。从仿真和实测结果表明,该方法对周期性窄带干扰和白噪声有着强抑制作用,评价指标更为显著。The detection process of partial discharge(PD)ultrahigh frequency(UHF)signals is severely susceptible to white noise and periodic narrowband noise.In order to extract PD UHF signals and suppress noise effectively,a denoising method based on singular value decomposition(SVD)and low rank radial basis function(RBF)neural network is proposed.Firstly,the noisy PD signals are constructed as a Hankel matrix and SVD matrix is decomposed into a feature matrix space.Then,the singular value mutation point in the feature matrix is set as a threshold to remove the narrowband noise.Finally,RBF neural network is used to approximate the denoised PD signals,and Gaussian window filtering is used to extract the PD signals.The proposed method is compared with reverse separation(RS)and morphology wavelet filter(MWF).The simulation and field-detection results show that the method has a stronger inhibition effect on periodic narrowband noise and white noise,and the evaluation index is more significant.

关 键 词：局部放电 奇异值分解 神经网络 白噪声 周期性窄带干扰 高斯窗 

分 类 号：TM855[电气工程—高电压与绝缘技术] TP183[自动化与计算机技术—控制理论与控制工程]