分段聚合近似和数值导数的动态时间弯曲方法  

Dynamic time warping based on piecewise aggregate approximation and data derivatives

作　　者：李海林[1] 梁叶[1] 

机构地区：[1]华侨大学信息管理系,福建泉州362021

出　　处：《智能系统学报》2016年第2期249-256,共8页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(61300139);福建省中青年教育科研基金项目(JAS14024);华侨大学中青年教师科研提升计划项目(ZQN-PY220)

摘　　要：针对动态弯曲方法对时间序列数据相似性度量的质量和效率的局限性,本文提出一种基于分段聚合近似和数值导数的动态时间弯曲方法。该方法通过分段聚合近似将时间序列数据进行有效地降维,再结合数值导数对降维后的特征序列构建新特征序列,并且设计符合该特征序列相似性度量方法。实验结果分析表明,与传统动态弯曲方法相比,新方法具有较好的度量质量,能在时间序列数据挖掘中得到较好的分类效果,且在低维空间具有较高的分类效率,具有一定的优越性。Dynamic time warping （DTW） is often used to measure the similarity of time series data; however, it has efficiency and quality limitations. In this study, a novel DTW method combining piecewise aggregate approximation （PAA） and derivatives is proposed to measure the similarity of time series. The dimensionality of the time series data was effectively reduced by PAA, and the feature sequence was transformed into new sequences by combining the numerical derivatives after the dimensionality reduction. Furthermore, the DTW design corresponded to the sim- ilarity measurement method of the feature sequence. The experimental results demonstrate that the proposed method is superior because it has better measurement quality, obtains a better classification effect in time series data min- ing, and has high efficiency in lower dimensional spaces.

关 键 词：动态时间弯曲 时间序列 分段聚合近似 数值导数 相似性度量 分类 数据降维 特征表示 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] O211.61[自动化与计算机技术—计算机科学与技术]