一种改进支持向量域数据描述方法及其应用  

An Improved Support Vector Data Description Method and Its Application

作　　者：罗键[1] 庄进发[1] 李波[1] 吴长庆[1] 黄春庆[1] 

机构地区：[1]厦门大学信息科学与技术学院

出　　处：《厦门大学学报（自然科学版）》2009年第5期656-661,共6页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金(60704043);国家"211"工程(王艺之)项目"立体通信和信息集成技术"资助

摘　　要：针对支持向量域数据描述中的核参数选择及其决策边界规整问题,提出一种新的改进算法.该算法根据支持向量域数据描述本身的特点,利用非高斯性来测量核空间样本接近球形区域分布的程度,并根据此测量结果来优化核参数.当核参数选定之后,核空间样本可能存在分布不均匀的现象,对此,该算法应用核主元分析来进行规整,即通过尺度变换来调整各主轴的长度,以获得一个更合理的球形分界面.最后通过标准数据集和TEP故障诊断仿真以验证该算法,仿真实验结果表明了该算法的有效性.To address the problem of kernel parameter selection and regulating decision boundary in support vector data description （SVDD）, this paper proposes a new algorithm. According to the characteristic of SVDD, the proposed algorithm utilizes the non- Gaussian to measure how kernel samples approximate to a spherical area, and then optimize the kernel parameter. If the samples in feature space are with uneven distribution after the kernel parameter determined,KPCA（kernel principal component analysis） is employed to structure the hyper-sphere. The algorithm use KPCA to normalize,which is through the scale transform to adjust the length of the spindle in order to obtain a more reasonable spherical interface. The proposed approach is illustrated by the application to standard data and TEP. The results verified the feasibility and effectiveness of this method.

关 键 词：支持向量域描述 核主元分析 非高斯性 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]