摘要
为了克服基于点距离的时间序列相似性搜索物理概念模糊和速度慢的缺点,提出时间序列的分段趋势序列(PTS)概念,并在此基础上提出一种变步长趋势子序列搜索算法.该算法基于时间序列分段线性表示理论,通过相似阈值和子序列间的趋势距离计算跳跃步长,从跳跃步长后开始的子序列进行下一次匹配,从而对全序列实现跳跃式搜索。理论分析和仿真结果表明,该算法对基于趋势表示的子序列搜索在时间和空间上都具有更优的性能,适用于时间序列的动态特征分析.
To overcome the shortcomings of concept indistinct and slow speed in time series similarity searching based on point distance, a piecewise trend sequence (PTS) and a variable step algorithm for sub-trend sequence searching based on PTS were proposed. The algorithm was founded on the theory of piecewise linear representation and calculated skip steps with similarity threshold and trend distance between sub-series. The next matching started after skip steps and skip-searching for whole series was realized. Theoretic analysis and simulation indicate that the algorithm has better performance for sub-trend searching in temporal and space, and is useful in time series dynamic feature analysis.
出处
《浙江大学学报(工学版)》
EI
CAS
CSCD
北大核心
2004年第12期1566-1569,共4页
Journal of Zhejiang University:Engineering Science
基金
国家"863"高技术发展计划资助项目(2001AA411210
2001AA413220)
国家"973"重点基础研究发展规划资助项目(2002CB31220304).
关键词
趋势序列
子序列搜索
数据挖掘
Computer simulation
Learning algorithms
Numerical analysis
Time series analysis