自适应代价动态时间弯曲的多元时间序列相似性度量

Multivariate Time Series Similarity Measure for Dynamic Time Warping of Adaptive Cost

文章针对在多元时间序列动态时间弯曲度量中出现一对多情形容易产生距离相同的多条匹配路径导致无法确定最优路线的问题,提出一种自适应代价动态时间弯曲的多元时间序列相似性度量方法(ACM-DTW)。首先,多元时间序列按变量纵向排列把每个变量中数值列作为向量看待,计算向量之间的欧式距离作为两条多元时间序列间的基础距离矩阵;然后,自适应代价函数更新权重减少点列的重复使用次数,运用ACM-DTW度量多元时间序列间的相似性;最后,计算k近邻法在不同数据集上分类准确率。实验证明,ACM-DTW能有效地改进匹配时一对多情形从而实现较好的匹配结果,具有良好的准确性。

Aiming at the problem that in the dynamic time warping measurement of multivariate time series,the one-to-many case tends to produce multiple matching paths with the same distance,which causes failure to determine the optimal route,this paper proposes a method for multivariate time series similarity measurement of adaptive cost dynamic time warping(ACM-DTW).Firstly,the multivariate time series is arranged longitudinally according to the variables and the numerical columns in each variable are treated as vectors,and the Euclidean distance between vectors is calculated as the basic distance matrix between two multivariate time series.And then,the adaptive cost function updates the weight to reduce the number of reuse of the point columns,ACM-DTW used to measure the similarity between multivariate time series.Finally,the classification accuracy of k-nearest neighbor method on different data sets is calculated.Experiments show that ACM-DTW can effectively improve the matching of one-to-many cases to achieve better matching results with good accuracy.