时间序列相似性度量方法  

Method of time series similarity measure

作　　者：王燕[1] 安云杰 

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《计算机工程与设计》2016年第9期2520-2525,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61263019)

摘　　要：在时间序列相似性度量中,符号聚合近似(symbolic aggregate approximation,SAX)方法没有将符号化后的模式序列进一步处理,导致存在一定误差,为此提出将算术编码技术引用到SAX中,即将符号化序列转换为编码序列,实现时间序列在概率区间上的分析与度量;在计算序列间的相似度时采用分层欧式距离算法,综合考虑序列的统计距离和形态距离,由粗到细地进行筛选,达到序列整体趋势匹配以及细节拟合的目标。实验结果表明,该方法在不同的数据集上都有一定的可行性,具有较高的准确度和较好的鲁棒性。In time series similarity measure,symbolic aggregate approximation（SAX）does not further process the patterns sequence,which leads to certain error.To solve the problem,the arithmetic encoding technology was introduced into SAX.The symbol sequence was changed to encoding sequence,which was implemented to measure and analyze time series in probability interval.The hierarchical Euclidean distance algorithm was used in similarity measure of time series.Statistical and morphological distances of time series were considered,sequences were filtered from rough to subtle,and the overall trend of sequence was matched,and the goal of fitting was reached in detail,thus,the ultimate realization of time series similarity measure was achieved.Experimental results verify that using this method in different data sets has certain feasibility,and higher accuracy and better robustness.

关 键 词：时间序列 相似性度量 关键点对等 算术编码技术 符号化 分层欧式距离 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]