基于平稳子空间分析和相对熵的分类算法  

Classification Algorithm Based on Stationary Subspace Analysis and Relative Entropy

作　　者：周晓园[1] 余旌胡[1] 

机构地区：[1]武汉理工大学理学院,湖北武汉430070

出　　处：《河南科技大学学报（自然科学版）》2016年第6期42-45,7,共4页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(114331011);中央高校基本科研业务费专项基金项目(2015-zy-115)

摘　　要：针对多维时间序列维数多、变量间关系复杂的特点,提出了一种基于平稳子空间分析和相对熵的分类算法。首先,利用平稳子空间分析法将多维数据分离为平稳子空间和非平稳子空间;其次,利用相对熵衡量平稳子空间的分布相似性;最后,进行真实数据集的分类。研究结果表明:平稳子空间分析和相对熵分类算法优于DTW算法和PCA-ED算法。Based on the characteristics that the multivariate time series were multivariate and had complex relationship among the variables,a classification algorithm based on stationary subspace analysis( SSA) and relative entropy( KL) was proposed. Firstly,stationary subspace analysis method was adopted to divide the multivariate data into the stationary subspace and the non-stationary subspace. Then,relative entropy was used to measure the similarity of distributions between the stationary subspaces. Finally,the classification of real dataset was carried out. The results show that the SSA-KL algorithm is better than the DTW algorithm and the PCA-ED algorithm.

关 键 词：多维时间序列 平稳子空间分析 相对熵 分类算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]