基于OLPP符号表示的时间序列分类算法  

TIME SERIES CLASSIFICATION BASED ON OLPP SYMBOLIC REPRESENTATION

作　　者：武天鸿 翁小清[1] Wu Tianhong;Weng Xiaoqing(College of Information Technology,Hebei University of Economics and Business,Shijiazhuang 050061,Hebei,China)

机构地区：[1]河北经贸大学信息技术学院,河北石家庄050061

出　　处：《计算机应用与软件》2021年第1期303-312,共10页Computer Applications and Software

摘　　要：基于符号表示的时间序列分类方法是时间序列数据挖掘的关键技术。大部分现有方法主要针对单个时间序列样本进行符号表示,没有考虑样本间的近邻关系对符号化分类的影响。对此提出一种基于正交局部保持映射(Orthogonal Locality Preserving Projection,OLPP)的时间序列符号表示方法。使用OLPP对原始数据集进行维数约减,利用信息增益寻找维数约减后数据的最佳符号投影区间,采用多重系数分箱技术(Multiple Coefficient Binning,MCB)将降维后数据表示成符号序列。该算法在20个时间序列数据集上的分类效果好于已有方法,有效利用样本间的近邻关系能够显著提高算法的分类性能。Time series classification based on symbolic representation is the key technology of time series data mining.Most of the existing methods mainly aim at symbolic representation of single time series samples,and do not consider the influence of neighbor relationship between samples on symbolic representation.This paper proposes time series classification based on OLPP symbolic representation.The OLPP was used to reduce the dimension of the original data set,and then we used the information gain to find the best symbol projection intervals and discretization the reduced-dimensional data into a symbol sequence with multiple coefficient binning(MCB).The classification performance of our algorithm is better than that of the existing methods on 20 time series data sets,and the effective use of the nearest neighbor relationship between samples can significantly improve the classification performance of the algorithm.

关 键 词：时间序列分类 符号表示 正交局部保持映射 信息增益 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]