摘要： 在聚类过程中为保留数据的重要形态与趋势特征，提出一种基于形态特征的数据流聚类方法。在初始化阶段提取重要特征点表示序列分段，在在线更新阶段使用部分动态时间弯曲方法计算子序列距离，基于动态滑动窗口思想保证多条数据流中数据的同步，在用户触发聚类阶段提出数据流聚类方法。通过对仿真数据和实际股票数据的分析结果表明，在参数设置合理的情况下，该方法可以获得接近0.95的聚类演化精度。

关键词: 数据流, 聚类演化, 数据挖掘, 形态特征

Abstract: In order to retain shape and tend features during the clustering process, this paper proposes a data stream clustering method based on shape feature. In the initialization stage, the subsequence is represented with the important points. In the online update stage, Partial Dynamic Time Warping(PDTW) method is used to compute the distances between the subsequences and ensure the data synchronization using the dynamic sliding window. In the clustering stage triggered by the user, the data streams clustering method is proposed. Experimental results show that the shape-based clustering over data streams can get the evolution accuracy of 0.95 with the reasonable parameters.

Key words: data stream, clustering evolution, data mining, shape feature

中图分类号: 

• TP39