基于CEEMDAN和卷积神经网络的配电网故障选线新方法  

A novel fault line selection method for distribution network based on CEEMDAN and convolutional neural network

作　　者：马红月 李温静 吴文炤 张楠 王婧 MA Hongyue;LI Wenjing;WU Wenzhao;ZHANG Nan;WANG Jing(State Grid Information&Telecommunication Group Co.,Ltd.,Beijing 102200,China)

机构地区：[1]国网信息通信产业集团有限公司信通研究院,北京102200

出　　处：《电测与仪表》2024年第10期97-103,共7页Electrical Measurement & Instrumentation

基　　金：国家重点研发计划资助项目(2020YFB0905900)。

摘　　要：文中提出一种基于自适应噪声的完全集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)和卷积神经网络(convolutional neural networks,CNN)的配电网故障选线方法。利用CEEMDAN算法分解各线路零序电流信号,得到各线路零序电流的内禀函数;将各线路内禀函数按顺序拼接到一起得到一个时频数据矩阵。时频数据矩阵囊括丰富的故障特征,可以反映当前系统工况;通过卷积神经网络自主地挖掘时频数据矩阵的故障特征向量,通过softmax函数输出故障线路编号,实现配电网故障选线。通过仿真实验表明,该方法不受过渡电阻、检测时延等因素影响,可准确、有效地识别故障线路。A fault line selection method for distribution network based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and convolutional neural network(CNN)is proposed.Firstly,the CEEMDAN algorithm is used to decompose zero-sequence current signal of each line to obtain the intrinsic function of zero-sequence current of each line.Secondly,the intrinsic functions of each line are spliced together in order to obtain a time-frequency data matrix containing abandunt fault features,which corresponds to the current system operating conditions.Finally,the fault eigenvector of the time-frequency data matrix are independently mined by CNN,and the fault line selection of distribution network is realized through the fault line number output of softmax function.Simulation experiments show that the method is independent of transition resistance,detection time delay and other factors,which can accurately and effectively identify the fault lines.

关 键 词：故障选线 CEEMDAN 卷积神经网络 数据矩阵 故障特征向量 

分 类 号：TM715[电气工程—电力系统及自动化]