Abstract: A new subsequence cluster algorithm is proposed to solve the trivial similarity and horizontal stretching problem. It smoothes time series through a’trous-smooth-filters, and generates trivial clusters from the obtained scale sequences, then clusters on the representative subsequences of scale sequences. The new algorithm solves the problem of trivial similarity, and it can find similar subsequences with difference length, so it can solves the problem of horizontal stretching. The experiment shows that the new algorithm is effective for subsequence clustering.

Key words: time series analysis, clustering algorithm, wavelet filter, horizontal stretching

摘要： 针对时间序列子序列聚类存在的平凡相似和水平伸缩等问题，提出了一种新的子序列聚类算法。它采用多孔平滑滤波器组对时间序列进行低通平滑处理，在所得到的多个尺度序列上生成平凡簇，然后将各个平凡簇的代表子序列作为数据样本进行聚类。新方法利用平凡簇克服了子序列聚类中的平凡相似问题，并且可以在时间序列上发现不等长的相似子序列，较好地解决了水平轴伸缩问题。实验结果证明新算法对于子序列聚类具有比较好的效果。

关键词: 时间序列分析, 聚类算法, 小波滤波, 水平伸缩