基于质心频率和BP神经网络的配网故障测距  

Fault Location in Distribution Networks Based on Centroid Frequency and BP Neural Networks

作　　者：周念成[1] 肖舒严 虞殷树 王强钢[1] Zhou Niancheng;Xiao Shuyan;Yu Yinshu;Wang Qianggang(State Key Laboratory of Power Transmission Equipment&System Security and New Technology Chongqing University Chongqing 400044,China;Ningbo Power Supply Company of State Grid Zhejiang Electric Power Company Ningbo 315000,China)

机构地区：[1]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆400044 [2]国网浙江省电力公司宁波供电公司,宁波315000

出　　处：《电工技术学报》2018年第17期4154-4166,共13页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金项目(51577018);重庆市科技计划项目(cstc2015jcyj BX0033)资助

摘　　要：故障距离越大,零序电流低频分量含量越高,电流频谱呈现出左移趋势;故障合闸角越接近90°,电流频谱高频分量含量越高;过渡电阻阻值越大,电流频谱低频分量含量越高。据此,提出了基于质心频率的频域故障测距方法。运用质心频率描述电流频谱整体分布特征,过渡电阻、故障距离、故障合闸角对电流频谱的影响通过质心频率刻画,运用BP神经网络模拟过渡电阻、故障距离、故障合闸角与频谱特征量质心频率之间的映射关系,完成测距。同时提出了应对质心频率-故障距离关系曲线不单调特殊情况的辅助判据,结合质心频率曲线与辅助判据确定最终故障位置。测距结果表明,该频域测距方法对混合线路及带分支馈线测距具有明显优势,能适应配网复杂的网架结构,并能够在训练样本数量不足且存在误差时准确测距。Low frequency component contents of zero-sequence currents increase with the growth of fault distance,which makes current spectrums appear left-shift tendency.In addition,the closer fault closing angle approaches to 90°,the more high frequency component contents of zerosequence currents are,which opposites to the case that transient resistance increase.Hereby,this paper proposes a frequency domain fault location method based on centroid frequency.Centroid frequency can be used to describle current spectrums entire distribution feature,further,the influence on current spectrums by transient resistance,fault distance and fault closing angle can be describled by centroid frequency.By utilizing BPNN(back propagation neural networks),the mapping relationship can be simulated between transient resistance,fault distance,fault closing angle and centroid frequency,which realizes fault loaction.In the case that centroid frequency fault distance curves are not monotonous,an assitant criteria is given,further,final resulsts are obtained by centroid frequency fault distance curves combined with the assitant criteria.Results prove that the method proposed by this paper has a distinct advantage than other frequency domain methods in the terms of fault location in hybrid line or feeder line with branches and apply to complex structures in distribution networks.Moreover,accurate fault location results can be obtained when training samples lack in quantity and accuracy.

关 键 词：中性点经消弧线圈接地系统 频域法测距 质心频率 BP 神经网络 

分 类 号：TM77[电气工程—电力系统及自动化]