摘要
由于DTW距离度量方法的计算时间和空间复杂度较高,不能满足大规模时间序列流中的相似性搜索要求,提出一种基于DTW的时间序列流相似性搜索方法。利用全局约束和时间序列标准化结合的方法提高搜索的精度,针对时间序列流中数据标准化方法计算代价过高问题,利用时间序列标准化和封袋逐步更新方法的下界距离,利用双循环缓冲区,存储查询序列的上下边界,进一步提高其数据读取和计算速度。实验结果表明,该方法与传统的静态时间序列搜索方法相比具有相同的准确度,但其搜索速度更快且DTW下界距离紧致性更好。
Due to the high computational time and space complexity of dynamic time warping(DTW),it can not satisfy the similarity search in large-scale time series stream.A DTW-based time series stream search method was proposed,which used global constraints and time series normalization methods to improve the accuracy of the search.To solve the problem of the high cost in data standardization in time series stream,the lower bounds that combined the time series normalization with the envelope update incrementally method was proposed.The upper and lower bounds of the query sequence were stored in double loop buffers to further improve the data reading and calculating speed.Experimental results show that the proposed method has the same accuracy compared with the traditional static time series search method,but its search speed is faster and DTWlower bound distance compactness is better.
出处
《计算机工程与设计》
北大核心
2017年第12期3291-3297,共7页
Computer Engineering and Design
基金
重庆市科委基金项目(2012jcsf-jfzhX0004)
