一种新的DTW最佳弯曲窗口学习方法

摘要

The dynamic time warping is a classic similarity measure which can handle time warping issue in similarity computation of time series,and the DTW with constrained warping window is the most common and practical form of DTW. After systematically analyzing the traditional learning method for optimal warping window of DTW, we introduced time distance to measure the time deviation between two time series,and proposed a new leaning method for optimal warping window based on time distance. Since the time distance is an appurtenant of the DTW computation, the new method can improve DTW classification accuracy with little additional computatioa Experimental data show that the optimal DTW with best warping window gets better classification accuracy when the new learning method is employed. What is,.more,the classification accuracy is better than the ERP(Edit Distance with Real Penalty) and the LCSS(Lon-gest Common SubSequence),and is close to the TWEDCTime Warp Edit Distance).%时间序列相似性查询中,DTW(Dynamic Time Warping)距离是支持时间弯曲的经典度量,约束弯曲窗口的DTW是DTW最常见的实用形式.分析了传统DTW最佳弯曲窗口学习方法存在的问题,并在此基础上引入时间距离的概念,提出了新的DTW最佳弯曲窗口学习方法.由于时间距离是DTW计算的附属产物,因此该方法可以在几乎不增加运算量的情况下提高DTW的分类精度.实验证明,采用了新的学习方法后,具有最佳弯曲窗口的DTW分类精度得到明显改善,分类精度优于ERP(Edit Distance with Real Penalty)和LCSS(Longest Common SubSequence),接近TWED(Time Warp Edit Distance)的水平.