多元时间序列模式匹配方法研究

针对多元时间序列模式匹配的方法难以高效、准确地刻画序列相似程度的问题,在考虑变量的量纲和特征差异的基础上,对多元时间序列进行多维分段拟合;然后,选取各个变量维度上拟合线段的倾斜角和时间跨度作为模式的描述方式,提出一种基于动态时间弯曲(DTW)的多元时间序列趋势距离匹配方法.实验结果表明,所提出的模式匹配方法对由连续型变量组成、时间跨度较大且体现一个连续、完整动作过程的多元时间序列,具有较好的匹配效果.

Research on pattern matching method for multivariate time series

Common methods for matching multivariate time series can’t measure their similarity rapidly and accurately.
Multivariate time series are fitted with multidimensional piecewise method on the basis of considering feature difference of
different variables. Then the angle of inclination and time span of a fitting line segment in a certain variable dimension are
chosen as feature pattern. A pattern matching method based on dynamic time warping(DTW) is proposed for multivariate
time series. Finally, the experimental results show that the proposed method can measure the similarity of multivariate
time series rapidly and accurately, which are composed of continuous variables and can present a whole action process in a
comparatively long time.