摘要
该文将K近邻分类方法引入基于溶解气体法(DGA)的变压器故障诊断中。并针对经典K近邻方法(K-NN)存在的缺陷,提出根据待分类样本和近邻样本的距离来加权的改进方法。最后通过仿真验证了该方法的有效性。仿真表明,当K的取值在一定范围内时,经典的K-NN算法和加权的K-NN方法都具有比较高的准确性,且加权的K近邻方法比经典的K近邻法体现出了更好的性能。
This paper introduces the K-NN algorithm to lead-in DGA （Dissolved Gas Analysis） and develops a novel power transformer fault diagnosis method. To overcome the drawback of typical K-NN algorithm,a weighted K-NN method is presented by taking into account the distance between test pattern and its neighbor patterns. The simulation result shows that both typical K-NN and weighted K-NN can help to diagnose fault accuracy while the K is set properly. Besides,it also reveals that the weighted K-NN has better performance that the typical K-NN.
出处
《电气自动化》
2010年第5期59-61,80,共4页
Electrical Automation
基金
国家自然科学基金(50677062)
西藏自治区教科研究规划重点课题(藏教高[2007]6号文件)
关键词
电力系统
变压器故障诊断
模式识别
加权K-NN
DGA
power systems power transformer fault diagnosis pattern classification weighted K-NN DGA
