基于极值点特征的时间序列相似性查询方法  

Time series similarity matching algorithm based on extreme points

作　　者：吴学雁[1,2] 黄道平[1] 莫赞[2] 

机构地区：[1]华南理工大学自动化科学与工程学院,广州510640 [2]广东工业大学管理学院,广州510520

出　　处：《计算机应用研究》2010年第6期2068-2070,2074,共4页Application Research of Computers

基　　金：广东省自然科学基金资助项目(6300278)

摘　　要：为了提高时间序列子序列匹配的准确度和效率,提出了基于极值点特征的时间序列相似性查询方法。首先识别出时间序列中的极值特征点,根据极值点使用多层次极值划分法对长序列进行划分;然后对划分得到的多层次子序列集使用改进的动态时间弯曲方法与查询序列进行相似性匹配;最后找到与查询序列最相似的子序列。实验表明,此方法在保证准确度的情况下大大提高了相似性搜索过程的效率。In order to improve the accuracy of time series subsequence similarity matching, this paper proposed time-series similarity matching algorithm based on the extreme points. First of all, the algorithm recognized the extreme points of the time series and used the multi-level extreme segmentation method to divide the long sequence. Then put forward an improved dynamic time warping method to do similarity matching between the multi-level subsequence set and the query sequence. At last found the most similar subsequence to the query sequence. Experiments reveal that this method greatly increases the efficiency of similarity matching with ensuring the accuracy.

关 键 词：时间序列 相似性查询 数据挖掘 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]