基于信号图像化和CNN-ResNet的配电网单相接地故障选线方法  

Single-phase ground fault line selection method for distribution network based on signal imaging and CNN-ResNet

作　　者：缪欣[1] 张忠锐 郭威 侯思祖[3] MIAO Xin;ZHANG Zhongrui;GUO Wei;HOU Sizu(Dongfang Electronics Co.,Ltd.,Yantai 264000,China;School of Electrical and Control Engineering,North China University of Technology,Beijing 100144,China;School of Electrical and Electronic Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]东方电子股份有限公司,山东烟台264000 [2]北方工业大学电气与控制工程学院,北京100144 [3]华北电力大学电气与电子工程学院,河北保定071003

出　　处：《中国测试》2024年第6期157-166,共10页China Measurement & Test

基　　金：国家重点研发计划(2018YFF01011900);北方工业大学科研启动基金项目资助(11005136024XN147-30)。

摘　　要：配电网发生单相接地故障时,零序电流呈现较强的非线性与非平稳性,故障选线较为困难,针对此问题,提出一种基于信号图像化和卷积神经网络-残差网络的配电网单相接地故障选线方法。首先,利用排列熵优化变分模态分解算法的参数,将零序电流信号分解成一系列固有模态函数;其次,引入新的数据预处理方式,将固有模态函数转成二维图像,获得零序电流信号的时频特征图;最后,利用一维卷积神经网络提取零序电流信号的相关性和特征,利用残差网络提取时频特征图的特征,将两个网络融合,构建混合卷积神经网络结构,实现故障选线。仿真与实验结果表明,该方法能够在高阻接地、采样时间不同步、强噪声等情况下准确地选择出故障线路,可满足配电网对故障选线准确性和可靠性的需求。When a single-phase grounding fault occurs in a distribution network,the zero-sequence current exhibits strong nonlinearity and non-stationarity,making fault line selection difficult.To address this issue,a distribution network single-phase grounding fault line selection method based on signal imaging and CNN�ResNet is proposed.Firstly,the parameters of the variational mode decomposition(VMD)algorithm are optimized using permutation entropy,decomposing the zero-sequence current signal into a series of intrinsic mode functions.Secondly,a new data preprocessing method is introduced,converting the intrinsic mode functions into two-dimensional images to obtain the time-frequency feature map of the zero-sequence current signal.Finally,a one-dimensional convolutional neutral network(CNN)is used to extract the correlation and features of the zero-sequence current signal,while a residual neural network(ResNet)is employed to extract the features of the time-frequency feature map.The two networks are merged to create a hybrid convolutional neural network structure,enabling fault line selection.Simulation and experimental results show that this method can eliminate the influence of manual features,accurately select the fault line under conditions such as high-resistance grounding,asynchronous sampling time,and strong noise,meeting the requirements of accuracy and reliability for fault line selection in distribution networks.

关 键 词：变分模态分解 卷积神经网络 残差网络 故障选线 排列熵 

分 类 号：TB9[一般工业技术—计量学] TM721[机械工程—测试计量技术及仪器]