Abstract: This paper reviews the current research status of time series similarity search, and proposes ordinal-transformation-based similarity search method, which can achieve fast trend similarity search among sub-series of different length. In addition to insensitivity to horizontal shifting and time-axis scaling, the proposed method has a lower false alarm ratio and a higher indexing efficiency. Experimental results show the proposed method is quite competitive in terms of speed and robustness.

Key words: time series, ordinal pattern, ordinal transformation, similarity search

摘要： 分析了时间序列相似匹配搜索的研究现状，提出了基于序变换的时间序列相似匹配搜索方法。该方法能够实现趋势相似的不同长度子序列的快速匹配搜索，同时具有对匹配序列的平移、时间轴的伸缩不敏感、误警率低，以及易于建立索引等优点。实验结果证明了该方法的有效性。

关键词: 时间序列, 序模式, 序变换, 相似性搜索

CLC Number: 

• TP311