基于多维时间序列形态特征的相似性动态聚类算法

由于时间序列数据具有高维度、动态性等特点,这就导致传统的数据挖掘技术很难有效的对其进行处理,为此,提出了一种基于多维时间序列形态特征的相似性动态聚类算法(similarity dynamical clustering algorithm based on multidimensional shape features for time series,SDCTS).首先,提取多维时间序列的特征点以实现降维,然后,根据多维时间序列的斜率、长度和幅值变化的形态特征定义了一种新的时间序列相似性度量标准,进而提出无需人为给定聚类个数的多维时间序列动态聚类算法.实验结果表明,与其他算法相比,此算法对时间序列具有良好的聚类效果.

Similarity dynamical clustering algorithm based on multidimensional shape features for time series

Traditional data mining methods are difficult to deal with the high dimensionality and dynamics characteristic of the time series.Therefore, in this study, a similarity dynamical clustering algorithm based on multidimensional shape features for time series (SDCTS) was proposed.First, the feature points of multidimensional time series are extracted to realize dimensionality reduction.Second, a new similarity measure criterion is defined with the shape features (slope, length, and amplitude) of the obtained multidimensional time series, and thus a dynamical clustering algorithm of multidimensional time series is proposed without predefining clustering numbers.The experimental results demonstrate that the SDCTS algorithm improves the clustering accuracy for time series compared with other algorithms.