改进的多元时间序列符号化表示方法研究  

Research on Improved Data Symbolization of Multi-dimensional Time Series

作　　者：刘博[1] 郭建胜[1] 

机构地区：[1]空军工程大学装备管理与安全工程学院,陕西西安710051

出　　处：《计算机仿真》2015年第1期314-317,共4页Computer Simulation

摘　　要：符号化表示技术,可以有效降低时间序列的维度,是实现海量时间序列数据挖掘的基础。聚类结果的精度是运用聚类方法对多元时间序列进行符号化表示的关键。运用共同马氏距离虽然可以实现多元时间序列的符号化表示,但忽略了不同类样本数量对聚类结果的影响,从而降低了符号化表示的有效性。针对上述问题,提出一种改进共同马氏距离的多元时间序列符号化表示方法,通过改进共同协方差矩阵的计算方法,优化了聚类结果,从而提高了符号化表示的有效性。仿真结果表明,改进后的方法可以更加有效地对多元时间序列进行符号化表示。Symbolic approximation technology can reduce the dimension of time series effectively, which is the basis to achieve the data mining of massive time series. The accuracy of clustering results is the key to the symbolic approximation of multivariate time series using clustering methods. Although the use of Common Mahalanobis Distance can achieve the symbolic approximation of multivariate time series, it ignores the effect produced by the number of samples of different classes on clustering results, thus reduces the effectiveness of symbolic approximation. To over- come the defects, a new symbolic approximation of multivariate time series based on Improved Common Mahalanobis Distance is proposed. This method can optimize the clustering results, and improve the effectiveness of symbolic ap- proximation by improving the calculation method of common eovariance matrix. The experimental results show that the proposed method can realize more effective symbolic approximation for multivariate time series.

关 键 词：多元时间序列 协方差矩阵 马氏距离 聚类 向量 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]