基于代价敏感学习的配电网故障线路分类算法  

Faulty Feeder Classification Algorithm Based on Cost-sensitive Learning for Distribution Network

作　　者：张鑫[1] 周伟[1] 徐志宇[1] ZHANG Xin;ZHOU Wei;XU Zhiyu(School of Electrical and Information Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学电子与信息工程学院,上海201804

出　　处：《实验室研究与探索》2024年第11期46-51,81,共7页Research and Exploration In Laboratory

基　　金：同济大学第十八期实验教改项目(0800104322)。

摘　　要：为了实现配电网故障线路辨识,以真实录波数据为驱动,从大数据分析角度提出一种配电网单相接地故障定位方法。首先,利用故障波形关联维数、零序电流时频故障测度值等构建多维故障特征向量,反映各种单相接地故障工况;其次,设计一种基于代价敏感学习的接地故障分类器,解决非均衡数据集分类问题,并依托真实波形数据对算法中代价因子的选择进行优化;最后,形成一种基于代价敏感学习AdaCost算法的配电网单相接地故障定位方法。模拟电网实验结果表明,该方法能够有效提高故障线路识别准确率,而且不受故障类型、故障样本比例、中性点接地方式的影响,为配电网接地故障定位提供了一种解决方案。To achieve fault line identification in distribution network,the paper proposes a fault location method based on big data for distribution network.High dimensional combined characteristics in time-frequency domain are constructed by using correlation dimension algorithm and various time-frequency characteristic indices.Then,aiming at the unbalanced characteristics of the sample data set of single-phase-to-ground(SPG)fault,a faulty feeder identification based on AdaCost is proposed.This method trains cost factor based on true fault data,which can effectively reduce the false negative rate of fault samples.Finally,the proposed method is verified by physical model simulation.The simulation results show that the proposed method can effectively improve the accuracy of faulty feeder identification,and is not affected by the fault type,the proportion of fault samples and the neutral grounding modes.It provides a new solution for the fault location of SPG fault.

关 键 词：配电网 单相接地故障 代价敏感学习 非均衡数据集 故障定位 

分 类 号：TM761[电气工程—电力系统及自动化]