摘要
针对高维小样本数据在构造分类模型时容易产生过拟合现象及特征物理量间的非线性关系,采用核切片逆回归(KSIR)特征提取方法,首先采用核函数将样本数据从低维不可分空间映射到高维可分空间,然后结合类别先验信息进行切片分组,将映射样本向有效降维方向投影实现数据的综合降维。将KSIR与核主成分分析(KPCA)同时应用于轴承故障模式分类,结果表明:KSIR在选取合适参数后不仅适应数据间的非线性关系,而且能以更少、解释能力更强的特征向量取得更高的分类精度,较KPCA有更强的类间区分和特征提取能力。
The kernel sliced inverse regression, as a method for feature extraction, is introduced to solve the problems such as small sample data in high - dimension for building classification model prone to over - fitting and nonlinear rela- tionship between the physical characteristics. Firstly, the sample data from low -dimensional inseparable space is mapped into high -dimensional separable space with kernel function. Then the grouped data with priori information is projected to the effective dimension -reduction direction to reduce the data synthetic dimension. KSIR and KPCA are both applied to bearing fault pattern classification, the results show that when KSIR chose suitable parameters, it not only suits the nonlinear relationship between data but also uses fewer and more explicated characteristics of variables to acquire higher classification accuracy and shows better interclass classification and feature extraction performance than KPCA.
出处
《轴承》
北大核心
2014年第4期50-53,共4页
Bearing
基金
国防装备研究基金项目(ZJ201181)
关键词
滚动轴承
故障模式
核切片逆回归
特征提取
rolling bearing
fauh mode
kernel sliced inverse regression
feature extraction