基于特征矩阵的多元时间序列最小距离度量方法

2016-01-15李海林,郭韧,万校基

智能系统学报 2015年3期

网络出版地址：http://www.cnki.net/kcms/detail/23.1538.tp.20150401.1459.001.html

李海林，郭韧，万校基

(华侨大学信息管理系，福建泉州 362021)

摘要：相似性度量是多元时间序列数据挖掘任务过程中一项重要的前期工作，度量质量直接影响到后期整个数据挖掘的性能和结果。利用主成分分析方法对数据集中的每个多元时间序列数据进行特征分析，提取其特征矩阵并且构建相应的新正交坐标系。通过夹角公式来度量2个正交坐标系之间距离，并且结合匈牙利算法计算它们之间的最小距离，进而实现了一种基于特征矩阵的多元时间序列最小距离度量方法。实验结果表明，与传统方法相比，新方法具有较好的相似性度量质量，提高了多元时间序列的数据挖掘效果。

关键词：多元时间序列；相似性度量；特征矩阵；最小距离；主成分分析；匈牙利算法；数据挖掘

DOI:10.3969/j.issn.1673-4785.201405047

中图分类号：TP301 文献标志码：A

收稿日期：2014-05-23. 网络出版日期：2015-04-01.

基金项目：国家自然科学基金资助项目(61300139); 福建省中青年教师教育科研项目(JAS14024); 华侨大学中青年教师科研提升资助计划项目(ZQN-PY220).

作者简介：

中文引用格式：李海林，郭韧，万校基. 基于特征矩阵的多元时间序列最小距离度量方法[J]. 智能系统学报, 2015, 10(3): 442-447.

英文引用格式：LI Hailin, GUO Ren, WAN Xiaoji. A minimum distance measurement method for a multivariate time series based on the feature matrix[J]. CAAI Transactions on Intelligent Systems, 2015, 10(3): 442-447.

A minimum distance measurement method for a

multivariate time series based on the feature matrix

LI Hailin, GUO Ren, WAN Xiaoji

(Department of Information Management, Huaqiao University, Quanzhou 362021, China)

Abstract：Similarity measurement is one of the most important preliminary works in the process of multivariate data mining. Its quality directly influences the performance and result of the later tasks of data mining. The data of every multivariate time series in dataset can be analyzed by the principal component analysis. The feature matrices are extracted to construct the corresponding new orthogonal coordinate systems whose distance can be measured by cosine value of the angles between two axes. Meanwhile, the Hungary algorithm is applied to the minimum distance computation of the two coordinate systems. In this way, the minimum distance measurement method for the multivariate time series based on the feature matrix is achieved. The results of experiment demonstrated that the proposed method has better quality of similarity measurement than the traditional ones and improves the effects of data mining for the multivariate time series.

Keywords：multivariate time series; similarity measurement; feature matrix; minimum distance; principal component analysis; Hungary algorithm; data mining