【摘要】 时间序列是科研生产中广泛存在的重要的数据对象。由于时间序列形态复杂,对其进行相似性度量非常困难,因此设计高效的时间序列聚类算法成为一个很有价值的研究课题。提出了一种基于DTW度量和局部紧邻接图的聚类算法。新算法通过对DTW距离矩阵进行较小距离截断的办法分离出核心簇结构,进而确定聚类的簇数、划分核心点和边缘点,具有过程简单、物理意义明确的优点。在不同数据集上的仿真表明,新算法的聚类性能不仅超过了传统算法kmeans和hierarchy,同时优于最新的DP算法的DTW改进版。更多还原

【Abstract】 Time series is an important data type widely used in academia and industry. Due to the complex morphology,it is very difficult to measure similarity of time series. Therefore,designing efficient time series clustering algorithm becomes a valuable research topic. In this paper,a clustering algorithm is proposed based on DTW metric and local-close-link graph. The new algorithm separates the core clusters by truncating the DTW distance matrix with a small value,and then determines the number of clusters and divides the data set into core point set and border point set,which have the advantages of simple process and clear physical meaning. Simulation experiments on different data sets show that the clustering performance of the new algorithm not only surpasses the traditional algorithm k-means and hierarchy,but also outperforms the DTW improved version of the latest DP algorithm.更多还原