流式时间序列的实时相似度研究

Real time similarity of streaming time sequence

动态时间弯曲算法虽然适合度量时间序列的相似度,但是在大数据背景下,对于序列个数多、潜在长度可能是无穷、实时性要求高的流式时间序列,面临着算法简单、计算不简单的可计算问题。以Spark计算平台为基础,针对流式时间序列的特点,提出了一种流式动态时间弯曲算法,能实时计算动态时间序列近似值,误差可控、稳定,且具备大数据计算能力。最后通过实验验证了算法的可行性和稳定性。

Although the dynamic time warping algorithm is suitable for measuring time sequence sim ilarity, streaming time sequence has a large quantity of sequences, potential infinite length, and require rnent for high real-time performance in the big data background, thus facing problems of simple algo rithm and complex computation. We propose a new streaming dynamic time sequence algorithm accord ing to the features of streaming time sequence based on the Spark calculation platform, which can calcu late the approximate value of dynamic time sequence in real-time, and has controllable error, good sta bility, and ability of processing big data. Experimental results verify its feasibility and stability.