中文摘要：

对心电信号（ECG）这种高维的时间序列进行聚类,最重要的方面之一即进行特征提取。本研究提出利用自回归和移动平均（ARMA）模型拟合ECG信号,以拟合系数的欧氏距离为结构不相似测度征进行聚类。但此方法没有考虑样本数据的各维特征对聚类的不同贡献率,所以本文提出可以把首次聚类每维特征在聚类中的贡献率作为其权值,对每维数据加权后重新进行聚类。以MIT-BIH标准数据库中的正常窦性心率（NSR）和心室早期收缩（PVC）样本数据进行聚类分析,结果表明利用改进后的方法进行聚类的准确度达到93.10%,从而证明了所提方法的有效性。

英文摘要：

Feature extraction was one of the important parts in electrocardiogram（ECG） clustering.In this article,a simple autoregressive-moving average（ARMA） was applied to fit the ECG.we used the fitting coefficients＇ Euclidean Distance as dissimilarity to cluster ECG.However contribution of each dimension feature of data sample to the clustering was not considered in that algorithm.Therefore,we further took the contribution which was obtained by the firstly clustering as the weight of each dimension feature.After each dimension coefficient was weighted,we clustered these data again.The PVC and NSR data obtained from MIT-BIH Arrhythmia Database,was used for experimentation.The results showed that cluster precision reached to 93.10%,which proved effectiveness of the proposed method.