Granulation-based fuzzy clustering of large-scale time series

Abstract

The clustering of a group of large-scale time series with same length is a challenging problem. Facing with this problem, the existing clustering algorithms usually show high computation cost and low efficiency. In this paper, a granulation-based clustering method is proposed for this problem. In this method, each large-scale time series in the given group is firstly segmented into subsequences (segments or windows) according to some principle, and then in each window a fuzzy information granule is built for the subsequence included. After that, a granular time series corresponding to the processed large-scale time series is obtained. Processing all the original large-scale time series in the given group in same manner will result in a group of granular time series who have good fitness to the original group of time series and are the objects of our new granulation-based clustering method. We regard the clustering result of the group of granular time series as the cluster structure of the original group of large-scale time series. The simulation experiment shows good performance and high efficiency of the new clustering approach in revealing the cluster property of the original group of large-scale time series.