The clustering algorithm based on multivariate time series with unequal length

The dynamic clustering analysis of time series is an important topic in data mining research. The existing clustering algorithms often use K-means to cluster low dimensional data. However, it is not very applicable to data clustering of multivariate time series with unequal length. By combining clustering analysis method and time series data mining technology, this paper presents a clustering algorithm based on multivariate time series with unequal length. The algorithm can help eliminate the subjective incompatibility between clustering results and transcendent knowledge, making clustering results match up to objective reality. The clustering results based on this algorithm are interval values, thus lowering the risks of clustering. An analysis on load variation of a thermal power unit shows the algorithm can remarkably improve the similarity accuracy and is applicable to load identification for stable operation of thermal power units. © 2011 by Binary Information Press.