基于数据挖掘的孤岛检测整定阈值优化  

Optimization of threshold settings for islanding detection based on data mining

作　　者：谭啸风 许锦喜[2] 颜晟[3] 陈忻磊[3] 

机构地区：[1]上海交通大学电力传输与功率变换控制教育部重点实验室,上海200240 [2]泰州供电公司,江苏泰州225300 [3]上海市电力公司,上海200122

出　　处：《可再生能源》2014年第1期39-43,共5页Renewable Energy Resources

基　　金：国家高技术研究发展计划"863"课题(2012AA050803)

摘　　要：孤岛检测中整定判据及其阈值选取对于检测效果影响显著,然而孤岛检测整定值无固定原理公式。文章综合利用数据挖掘技术方法,从常规、多分辨率奇异熵判据中通过RELIEF算法筛选出最优特征判据,通过ROC评估,从C4.5,CART,SVM中选出最优分类算法;比较不同功率不平衡度条件下孤岛整定并加以试验验证。文章构建PSCAD仿真模型并对样本数据进行MATLAB预处理,最后3类算法的预测分类结果验证了结论。Eigenvalue criterion and its threshold settings greatly affects accuracy of islanding passive detection, however there is no prescriptive formula or principle to achieve the threshold settings. Ac- cording to the characteristics of islanding detection, data mining technique is employed to select most appropriate features. An estimate approach ROC is applied to select the best classification algorithm from algorithms such as C4.5、ART、SVM; Further, islanding passive detection under different power imbalances are compared. PSCAD model is established and sample data is preprocessed in MATLAB. Classification results of three algorithms above verify the proposal.

关 键 词：数据挖掘 阈值整定 孤岛检测 功率不平衡度 

分 类 号：TM615[电气工程—电力系统及自动化]