基于稀疏表示的时间序列最近邻分类  

The nearest neighbor classifier based on sparse representation for time series

作　　者：杨冠仪 於志勇[1] 郭文忠[1] 黄昉菀[1] YANG Guanyi;YU Zhiyong;GUO Wenzhong;HUANG Fangwan(Fujian Provincial Key Laboratory of Network Computing and Intelligent Information Processing,College of Mathematics and Computer Science,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]福州大学数学与计算机科学学院,福建省网络计算与智能信息处理重点实验室,福建福州350108

出　　处：《福州大学学报（自然科学版）》2020年第2期152-159,共8页Journal of Fuzhou University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61772136,61672159);福建省中青年教师教育科研基金资助项目(JT180045)。

摘　　要：提出一种基于稀疏表示的时间序列最近邻分类模型,旨在通过提取时间序列的关键特征,去除冗余信息,达到减少噪声干扰的目的.该模型首先求解时序数据基于过完备字典的稀疏表示,然后利用非零系数及其对应的原子重构原始序列,最后利用基于距离的分类器进行分类.在18个时间序列公开数据集上的实验结果表明,最近邻分类模型能够提高传统的最近邻分类器的分类准确率.The dynamic time warping distance can effectively solve the problem of time series with unequal length and local deformation,but its capability to dispose noise is weak.For the above problems,in this paper we propose a nearest neighbor classifier based on sparse representation for time series,aiming at extracting the main features of the time series,removing redundant information and decreasing influence of the noise.First of all,the model obtains the sparse representation based on the over-completed dictionary.Then the non-zero coefficients and their corresponding atoms are used to reconstruct the original sequence.Finally,the distance based classifier is used to classify the reconstructed sequences.The experiments on the 18 benchmark datasets from the UCR archive show that nearest neighbor classifier can improve the classification accuracy of traditional nearest neighbor classifier.

关 键 词：时间序列分类 稀疏表示 最近邻分类器 欧氏距离 动态时间弯曲 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]