基于局部线性嵌入的时间序列聚类

摘要

时间序列聚类是时间序列数据挖掘中重要的研究内容之一。由于时间序列的维数比较大，直接对时间序列原始数据进行聚类性能不理想，如何有效的对时间序列进行维数约简，并且保持原数据集本质特征，是本论文的主要研究点。首先使用局部线性嵌入（LLE）对时间序列样本维数约简，在低维空间对维数约简后的数据进行聚类，然后将它的聚类性能与已有方法如主成分分析（PCA）、分段聚合近似（PAA）进行比较。实验表明，使用 LLE更能提高聚类性能。%Time series clustering is one of the important research contents in time series data mining .Owing to the large dimension of the time series ,directly to the original time series data were clustering performance is not ideal ,how to reduce the dimension of time series ,and keep the original data set of essential characteristics ,is the main research point of this thesis . The first ,using locally linear embedding (LLE) of time series sample dimensionality reduc‐tion ,in the low dimensional space of dimensionality reduction of data clustering ,then its clus‐tering performance with existing methods such as principal component analysis (PCA) ,piece‐wise aggregate approximation (P AA) were compared .Experiments show that using LLE can improve the clustering performance .