Author Login Editor-in-Chief Peer Review Editor Work Office Work

Computer Engineering ›› 2021, Vol. 47 ›› Issue (8): 62-68,77. doi: 10.19678/j.issn.1000-3428.0058855

• Artificial Intelligence and Pattern Recognition • Previous Articles     Next Articles

DTW Similarity Measure Based on Segmentation Features and Adaptive Weighting

LIU Miaomiao1, ZHOU Conghua1, ZHANG Ting2   

  1. 1. School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China;
    2. Wuxi Maternal and Child Health Hospital, Wuxi, Jiangsu 214002, China
  • Received:2020-07-07 Revised:2020-08-07 Published:2020-08-12

基于分段特征及自适应加权的DTW相似性度量

刘苗苗1, 周从华1, 张婷2   

  1. 1. 江苏大学 计算机科学与通信工程学院, 江苏 镇江 212013;
    2. 无锡市妇幼保健院, 江苏 无锡 214002
  • 作者简介:刘苗苗(1995-),女,硕士研究生,主研方向为机器学习、数据分析;周从华,教授、博士、博士生导师;张婷,博士。
  • 基金资助:
    江苏省重点研发计划项目(BE2016630,BE2017628);无锡市卫生计生委科研项目(Z201603)。

Abstract: Dynamic Time Warping(DTW) faces high time complexity in the similarity measure for raw multivariate time series, and DTW sometimes causes transition stretching and compression in the process of pursuing minimum bending distance. To address the problem, this paper proposes a new DTW method based on segmentation features and adaptive weighting for similarity measurement of multivariate time series. The raw time series is segmented in each variable dimension. Then the certain segments are selected, and their gradient, maximum and minimum values, as well as the time span are taken as the features of each segment. So the dimensions of the raw series can be significantly reduced to improve the computational efficiency. During the calculation of the optimal bending path by DTW, the adaptive cost weight is set for each point to limit the number of reused points in the bending path, and effectively improve the low measurement accuracy caused by the over-stretching or compression of the time series. So the optimal path can be obtained. Experimental results show that the proposed method can measure the similarity between multiple time series and provide excellent measurement results on multiple datasets.

Key words: multivariate time series, Dynamic Time Warping(DTW), similarity measure, segmentation feature, adaptive cost weight

摘要: 利用动态时间弯曲(DTW)技术在原始多元时间序列进行相似性度量时时间复杂度较高,且DTW在追求最小弯曲距离的过程中可能会出现过渡拉伸和压缩的问题。提出一种基于分段特征及自适应加权的DTW多元时间序列相似性度量方法。对原始时间序列在各个变量维度上进行统一分段,选取分段后拟合线段的斜率、分段区间的最大值和最小值以及时间跨度作为每一段的特征,实现对原始序列的大幅降维,提高计算效率。在DTW计算最佳弯曲路径的过程中为每个点设置自适应代价权重,限制弯曲路径中点列的重复使用次数,改善时间序列因过度拉伸或压缩所导致的度量精度低的问题,以得到最优路径路线。实验结果表明,该方法能很好地度量多元时间序列之间的相似性,在多个数据集上都能取得较好的度量结果。

关键词: 多元时间序列, 动态时间弯曲, 相似性度量, 分段特征, 自适应代价权重

CLC Number: