融合趋势信息的时间序列符号聚合近似方法  

Time series symbolic aggregate approximation method for fusion trend information

作　　者：黄俊杰 徐兴华[1] 崔小鹏[1] 康军[1] 杨皓翔 Huang Junjie;Xu Xinghua;Cui Xiaopeng;Kang Jun;Yang Haoxiang(National Key Laboratory of Science&Technology on Vessel Integrated Power System,Naval University of Engineering,Wuhan 430033,China)

机构地区：[1]海军工程大学舰船综合电力技术国防科技重点实验室,武汉430033

出　　处：《计算机应用研究》2023年第1期86-90,共5页Application Research of Computers

基　　金：国家自然科学基金项目;湖北省自然科学基金项目。

摘　　要：为解决符号聚合近似方法(SAX)表示时间序列时忽略序列局部趋势的问题,提出一种融合形态趋势信息的时间序列符号聚合近似方法。该方法以子序列段中的最大值和最小值以及它们两者之间的相对位置为依据,定义一种新的趋势指标来描述子序列段的趋势,并使用融合趋势指标的符号矢量来近似表示时间序列。针对所提表示方法,给出一种新的距离度量方法,并在UCR数据集和电机转矩数据集上进行分类实验,实验结果表明,所提方法在绝大多数数据集上获得了较SAX方法更高的分类准确率,能够有效弥补SAX方法表示时间序列时忽略局部趋势的不足。To solve the problem of losing trend information when representing time series with symbolic aggregate approximation method(SAX),this paper proposed a new time series symbolic aggregate approximation method integrating morphological trend information.Based on the maximum and minimum values in the subsequence and their relative positions,this method defined a new trend index to describe the trend information of the subsequence segments,and used the symbol vector integrating trend index to approximately represent the time series.For the proposed representation method,this paper gave a new distance metric and used it to conduct classification experiments on the UCR datasets and motor torque dataset.The experimental results show that the proposed method obtains higher classification accuracy than the SAX method on most datasets,and can effectively make up for the deficiency of losing local trend when the SAX method represents time series.

关 键 词：时间序列 符号化表示 符号化聚合近似 趋势信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]