基于趋势特征的时间序列符号聚集近似表示方法  

Novel symbolic aggregate approximation approach for time series based on trend features

作　　者：张豹 应励志 余宇峰[2] ZHANG Bao;YING Lizhi;YU Yufeng(The 28th Research Institute,China Electronics Technology Group Corporation,Nanjing Jiangsu 210007,China;College of Computer and Information,Hohai University,Nanjing Jiangsu 211110,China)

机构地区：[1]中国电子科技集团公司第二十八研究所,南京210072 [2]河海大学计算机与信息学院,南京211110

出　　处：《计算机应用》2022年第S01期123-129,共7页journal of Computer Applications

基　　金：国家重点研发计划项目(2018YFC1508100)。

摘　　要：针对时间序列数据降维过程中易丢失趋势特征信息的问题,提出一种基于趋势特征的时间序列符号聚集近似表示方法,除保留各序列分段的均值特征外,采用分段的趋势距离因子及趋势形态因子共同描述序列趋势特征;并给出了满足下界密封性的距离度量方法,从而更好地表示具有不同趋势特征的时间序列。在公共数据集上的实验结果表明,该方法在分类误报率、降维比率等方面比符号聚集近似方法(SAX)和基于趋势距离的时间序列符号近似表示方法(SAX_TD)有10%以上的下降,并具有更好的下界密封性。实验结果证明,该算法在进行时间序列压缩的同时充分保留时间序列的趋势变化形态,从而提高时间序列数据挖掘的效率。In order to solve the problem of trend feature information loss in the process of time series dimensionality reduction,a new Symbolic Aggregate approXimation(SAX)approach for time series based on trend features was presented.The mean feature of each sequence segment was extracted,while the trend distance factor and trend shape factor of the segment were extracted to describe the sequence trend feature together.A distance measurement method that satisfies the lower bound tightness was givento better represent time series with different trend characteristics.The expeimental results on public dataset show that,the classification false positives rate and dimensionality reduction ratio were improved more than 10%over the SAX and Symbolic Aggregate approXimation based on Trend Distance(SAX_TD)method.It is proved that the proposed algorithm can retain the trend feature of the time series while compressing the time series and thereby improve the efficiency of time series data mining.

关 键 词：时间序列 趋势特征 符号聚集近似 下界 距离测量 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]