基于数据挖掘的配电网故障风险预警  

Fault risk early warning of distribution network based on data mining

作　　者：刘科研[1] 吴心忠 石琛 贾东梨[1] LIU Keyan;WU Xinzhong;SHI Chen;JIA Dongli(China Electric Power Research Institute, Beijing 100192, China;School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China)

机构地区：[1]中国电力科学研究院,北京100192 [2]北京交通大学电气工程学院,北京100044

出　　处：《电力自动化设备》2018年第5期148-153,共6页Electric Power Automation Equipment

基　　金：国家电网公司科技项目(支撑精益化管理的配电大数据分析技术研究与基础平台开发)(52020116000G)~~

摘　　要：为了提高配电网风险预警的准确性,提出了基于数据挖掘的配电网故障关联因素分析与风险预警的方法。通过数据清洗、数据变换、数据集成和离群样本剔除,归纳配电网四大类共28个故障特征;采用改进的Relief-Wrapper算法进行故障关联因素分析,剔除了6个冗余特征,形成了由22个故障特征组成的最优故障特征子集;提出了兼顾故障发生频率和失电负荷比例的配电网故障风险指标和风险等级划分方法,采用基于径向基函数的支持向量机(SVM)方法和最优故障特征子集进行风险预警。对某市120条馈线配电网进行了风险预警算例分析,结果验证了所提方法的有效性。The fault correlation factor analysis and risk early warning method of distribution network are proposed based on the data mining to improve the accuracy of the warning. Four kinds of general categories including 28 fault features of distribution network are summed up through data cleaning, data transformation, data integration and outlier sample elimination. The fault correlation factors are analyzed by the improved Relief-Wrapper algorithm, which eliminates 6 redundant features to form the optimal fault feature subset composed of the rest 22 fault features. The fault risk indexes and risk level classification method for distribution network are proposed considering the fre- quency of failure and the proportion of electricity load loss, and the risks are warned by the radial basis function based SVM （ Support Vector Machine） and the optimal fault feature subset. The risk early warning of a distribution network with 120 feeders is analyzed ,and the results verify the effectiveness of the proposed method.

关 键 词：配电网 数据挖掘 故障关联因素 最优故障特征子集 风险预警 支持向量机 风险指标 

分 类 号：TM732[电气工程—电力系统及自动化]