基于多特征融合与优化支持向量机的小电流接地故障区段定位方法  

Location Method for Small Current Grounding Fault Segment Based on Multi-feature Fusion and Optimized Support Vector Machine

作　　者：杜政奇 王敬华 张新慧[1] 李晨朝 王洪庆 张莹 DU Zhengqi;WANG Jinghua;ZHANG Xinhui;LI Chenzhao;WANG Hongqing;ZHANG Ying(School of Electrical and Electronic Engineering,Shandong University of Technology,Zibo 255049,China;Shandong Kehui Power Automation Co.,Ltd,Zibo 255087,China)

机构地区：[1]山东理工大学电气与电子工程学院,淄博255049 [2]山东科汇电力自动化股份有限公司,淄博255087

出　　处：《电力系统及其自动化学报》2023年第9期103-111,共9页Proceedings of the CSU-EPSA

基　　金：淄博市张店区校城融合项目(2021JSCG0006)。

摘　　要：针对配电网小电流接地系统单相接地故障频发及发生高阻接地时故障特征信号微弱等问题,提出一种基于多特征融合与优化支持向量机的小电流接地故障区段定位新方法。首先,结合单相接地故障时主谐振分量能量最大原理确定变分模态分解的最优分解层数,分解故障线路上各区段的零序电流得到平稳的本征模态分量;其次,挖掘暂态信息中能充分体现故障检测点差异性的5种特征量,即衰减直流分量衰减速度、暂态主频分量峰值与稳态工频分量幅值比、暂态主频分量与衰减直流分量能量比、相邻检测点故障零序电流波形差异系数、相邻检测点暂态主频分量极性比,构造多维度特征向量,输入到经改进鲸鱼算法优化的支持向量机分类模型中训练测试。仿真结果表明,本文定位模型能够准确判别故障区段。Since the single-phase grounding faults occur frequently in a small current grounding system for distribution network and the fault characteristic signal is weak when a high-resistance grounding occurs,a novel method for locating small current grounding fault segment based on multiple-feature fusion and optimized support vector machine(SVM)is proposed.First,the optimal number of decomposition layers of variational mode decomposition(VMD)is determined by combing the principle of maximum energy for the main resonance component under single-phase grounding fault,and the zero-sequence current on each segment of the fault line is decomposed into the steady intrinsic mode function(IMF).Second,five characteristic quantities in the transient information that can fully reflect the difference between fault detection points are excavated,i.e.,the speed of decaying DC component,amplitude ratio of peak value of tran⁃sient high-frequency component to steady-state frequency component,energy ratio of transient high-frequency compo⁃nent to decaying DC component,difference coefficient of fault zero-sequence current waveform between adjacent detec⁃tion points,and the polarity ration of transient high-frequency component of adjacent detection points.These quantities are used to construct multi-dimensional feature vectors and input into the SVM classification model optimized by the im⁃proved whale algorithm for training and testing.Simulation results show that the proposed location model can accurately identify the fault segment.

关 键 词：暂态零序电流 变分模态分解 特征融合 支持向量机 区段定位 

分 类 号：TM77[电气工程—电力系统及自动化]