Four different approaches to robust fuzzy clustering of time series are presented and compared with respect to other existent approaches. These approaches are useful to cluster time series when outlying values are found in these time series, which is often the rule in most real data applications. A representation of the time series by using B-splines is considered and, later, robust fuzzy clustering methods are applied on the B-splines fitted coefficients. Feasible algorithms for implementing these methodologies are presented. A simulation study shows how these methods are useful to deal with contaminating time series and also switching time series due to fuzziness. A real data analysis example on financial data is also presented.

介绍了四种不同的时间序列鲁棒模糊聚类方法，并与其他现有方法进行了比较。当在这些时间序列中发现离群值时，这些方法对聚类时间序列很有用，这通常是大多数实际数据应用程序中的规则。考虑使用 B 样条表示时间序列，然后将稳健的模糊聚类方法应用于 B 样条拟合系数。介绍了实现这些方法的可行算法。模拟研究显示了这些方法如何用于处理污染时间序列以及由于模糊性而切换时间序列。还介绍了金融数据的真实数据分析示例。