Abstract: This paper analyzes the problems of traditional Principal Component Analysis（PCA） when comparing the similarity of QAR data. The Kernel Principal Component Analysis（KPCA） based on EROS is proposed to deal with these problems. This paper introduces EROS method without vector treatment and adopts the kernel matrix of principal component analysis to reduce the dimension of QAR data. This paper gives classification on two groups of QAR data sets by using support vector products method with selecting different number of principal component, and compares it with SPCA and GPCA method. The results show that the proposed method used for QAR data has a good effect on classification.

Key words: Quick Access Recorder（QAR） data, Principal Component Analysis, kernel matrix, similarity

摘要： 叙述了传统的PCA方法在处理QAR数据相似性问题的不足，提出基于EROS的KPCA方法处理QAR数据之间的相似性问题。通过引入EROS方法而不需要对数据进行向量化，引入核矩阵对QAR数据进行主成分分析，可以有效降低数据的维数。选取两组QAR数据集，采用支持向量积方法，选用不同数目的主成分进行分类实验，同SPCA方法和GPCA方法进行比较，实验结果显示把该方法运用到QAR数据集，具有较好的分类结果。

关键词: 快速存取记录器（QAR）数据, 主成分分析, 核矩阵, 相似性