一种基于始末距离的时间序列符号聚合近似表示方法

时间序列数据的特征表示方法是时间序列数据挖掘任务的关键技术,符号聚合近似表示(SAX)是特征表示方法中比较常用的一种。针对SAX算法在各序列段表示符号一致时无法区分时间序列间的相似性这一缺陷,提出了一种基于始末距离的时间序列符号聚合近似表示方法(SAX_SM)。由于时间序列有很强的形态趋势,因此文中提出的方法选用起点和终点来表示各个序列段的形态特征,并使用各序列段的形态特征和表示符号来近似表示时间序列数据,以将其从高维空间映射到低维空间；然后,针对起点和终点构建始末距离来计算两序列段间的形态距离；最后, 结合 始末距离和符号距离定义一种新的距离度量方式,以更客观地度量时间序列间的相似性。理论分析表明,该距离度量满足下界定理。在20组UCR时间序列数据集上的实验表明,所提SAX_SM方法在13个数据集中获得了最高的分类准确率(包含并列最大的),而SAX只在6个数据集中获得了最高的分类准确率(包含并列最大的),因此SAX_SM具有比SAX更优的分类效果。

Symbolic Aggregate Approximation Method of Time Series Based on Beginning and End Distance

The feature representation method of time series data is the key technology of time series data mining task,and the symbolic aggregate approximation (SAX) method is most commonly used in feature representation methods.A symbolic aggregate approximation method based on beginning and end distance (SAX_SM) was proposed because SAX algorithm can not distinguish the similarity between time series when the symbol is consistent in each sequence segment of time series.Time series data have a strong morphological trend,so the proposed method uses the beginning point and the end point to represent the morphological feature of each sequence segment,and then uses the morphological feature and representation symbol of each sequence segment to approximate the time series data,in order to map it from high-dimensional space to low-dimensional space.Next,in order to calculate the morphological distance between the two sequences,this paper constructed beginning and end distance based on the beginning point and the end point.Finally,to measure the similarity between time series more objectively,a new distance metric approach was defined by combining the beginning and end distance and the symbol distance.The theoretical analysis shows that the new distance measure satisfies the lower bound theorem.Experiments on 20 sets of UCR time series data sets show that the proposed SAX_SM method achieves the highest classification accuracy (including the largest side by side) in 13 data sets,while SAX only gets the largest classification accuracy in 6 data sets (including the largest side by side).Therefore,SAX_SM has better classification result than SAX.