基于核密度估计的非线性时间序列聚类

本文研究的是时间序列的聚类问题。由于现实世界中时间序列多数是非线性的，而现有的时间序列聚类问题大都是基于线性时间序列模型进行聚类的，本文提出了可以用于非线性时间序列的聚类方法。以时间序列的二维核密度估计之间的相似性作为非线性时间序列的距离度量，该距离度量方式是一种非参数的距离度量方法，考虑到了时间序列自相关结构的差异，能够粗糙地识别时间序列形状和动态相关结构的相似性。与理论研究结果相一致，我们的模拟实验结果也验证了这种距离度量的有效性。

Nonlinear Time Series Clustering based on Kernel Density Estimation

This paper addresses the problem of time series clustering. Most of the popular clustering methods are for the non-linear time series, yet the existing problem of time series chusteing is mostly clustered by linear time series. It proposes a cluster algorithm which can be used in nonlinear time series clustering. It defines the distance of nonlinear time series determined by the similarity of the two dimensional kernel density estimation, which is a kind of nonparametric distance measure. It considers the difference of the autocorrelation structure, and can recognize the similarity of the shape and the dynamic structure of time series. Analogy to the theoretical analysis, the experiments on simulated data testify the effeciveness of the distance.