基于IForest和SVD的直流传感器去噪方法  

DC sensor signal denoising method based on isolation forest and singular value decomposition

作　　者：周军[1] 巩森 吴瑜坤 王岩 ZHOU Jun;GONG Sen;WU Yukun;WANG Yan(School of Electrical Engineering,Northeast Electric Power University,Jilin 132012,China;State Grid Jincheng Power Supply Company,Jincheng 048000,China;State Grid Jilin Electric Power Company Jilin Power Supply Company,Jilin 132012,China)

机构地区：[1]东北电力大学电气工程学院,吉林吉林132012 [2]国网晋城供电公司,山西晋城048000 [3]国网吉林省电力公司吉林供电公司,吉林吉林132012

出　　处：《电气应用》2022年第11期91-96,共6页Electrotechnical Application

基　　金：吉林省科技厅科技计划重点研发项目(20180201010GX)。

摘　　要：直流系统接地故障主要采用非接触式的直流漏电流检测方法。针对直流漏电流传感器灵敏度和稳定度不能同时提高、噪声干扰大的特性,提出了一种基于孤立森林-奇异值分解原理的去噪方法,在保证信号灵敏度的前提下,提高了信号检测的稳定性。首先采用孤立森林方法去除信号中离群点,提高信号的信噪比;使用奇异值分解,并重构去除信号中的白噪声;利用指数加权平均值算法平滑信号,进一步抑制干扰。仿真及实验结果表明,该方法与传统的低通滤波器方法和经验模态分解算法相比,去噪后信号信噪比更高,均方根误差更小,信号的稳定性明显提高。The ground fault of the DC system mainly adopts the non-contact DC leakage current detection method.Aiming at the characteristics that the sensitivity and stability of the DC leakage current sensor cannot be improved at the same time,and the noise interference is large,this paper proposes a denoising method based on the isolated forest-singular value decomposition principle,which improves the stability of signal detection.This method uses the isolated forest method to remove outliers in the signal and improves the signal-to-noise ratio of the signal;uses singular value decomposition and reconstruction to remove the white noise in the signal;uses the exponential weighted average algorithm to smooth the signal and further suppress the noise.Simulation and experimental results show that compared with the traditional low-pass filter method and empirical mode decomposition algorithm,this method has higher signal-to-noise ratio,smaller root mean square error,and more significant denoising effect.

关 键 词：直流小电流 噪声抑制 隔离森林 奇异值分解 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TM721.1[自动化与计算机技术—控制科学与工程]