中国机械工程 ›› 2012, Vol. 23 ›› Issue (3): 264-269.

• 机械基础工程 • 上一篇    下一篇

基于最大树聚类的多超球体一类分类算法及其应用研究

刘丽娟;陈果
  

  1. 南京航空航天大学,南京,210016
  • 出版日期:2012-02-10 发布日期:2012-02-15
  • 基金资助:
    国家自然科学基金资助项目(50705042, 61179057); 航空科学基金资助项目(2007ZB52022) 
    National Natural Science Foundation of China(No. 50705042, 61179057);
    Aviation Science Foundation of China(No. 2007ZB52022)

Study on One-class Classification with Multi Hyper-spheres Based on Maximal Tree Clustering and Its Applications

Liu Lijuan;Chen Guo
  

  1. Nanjing University of Aeronautics and Astronautics, Nanjing, 210016
  • Online:2012-02-10 Published:2012-02-15
  • Supported by:
     
    National Natural Science Foundation of China(No. 50705042, 61179057);
    Aviation Science Foundation of China(No. 2007ZB52022)

摘要:

提出了一种基于最大树聚类的多超球体一类分类算法。首先应用最大树聚类算法将训练样本聚为多个子类,再对各子类分别进行一类支持向量机(one-class SVM,OC-SVM)分类器训练,得到由各子类对应的超球体形成的多超球体一类分类模型。分别将该方法应用于仿真数据、UCI标准数据集以及转子故障诊断三个实例中,结果表明了该方法的有效性。

关键词:

Abstract:

An one-class Classification with multi hyper-spheres based on maximal tree clustering algorithm was presented herein. The training samples were firstly clustered into several sub-classes by the maximal tree clustering algorithm, and then, the sub-classes data were trained separately using one-class SVM(OC-SVM) and the multi hyper-spheres classifying models were established. The new method was applied to the instances of the simulation data set, UCI data sets and the rotor faults diagnosis, and the results show the effectiveness of the new method.

Key words: one-class classification, maximal tree clustering, multi hyper-sphere, rotor, fault diagnosis

中图分类号: