摘要
针对长时间序列,该文提出了一种新的能快速发现序列中时序模式的检索方法。首先将时间序列分成若干等长的子序列;接着从每个子序列中提取特征序列,该特征序列能够反映子序列中数据的变化趋势;然后根据每个特征序列将相应的子序列分配到一系列盒子中,使得不同盒子中的子序列因数据变化趋势不同而不相似,而在同一盒子中的序列由于数据变化趋势相同而有可能相似;最后通过计算每个盒子中任意两个子序列间的欧几里德距离来发现所有的模式。有关实验证明该算法是行之有效的。
In this paper a fast algorithm is presented for recovering time-series patterns in a long sequence.First,the sequence is segmented into same-length subsequences.Then a feature series is extracted from each subsequence to show its changing properties.Third,all the subsequences are distributed into a set of boxes according to their feature series,so that those in a box are possibly similar,while those in different boxes are impossibly similar.Finally the algorithm discovers all time -series patterns by computing Euclidean distance between any two subsequences in each box.The experiment results prove that it can be put into practice and work very efficiently.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第21期192-194,共3页
Computer Engineering and Applications
基金
国家自然科学基金重点项目(编号:69835001)
国家863高技术研究发展计划(编号:2001AA110464)资助
关键词
时间序列
时序模式
特征序列
欧几里德距离
Time sequence,time-series patterns,feature series,Euclidean distance