Time series granulation and forecasting based on k-plane clustering

A special k-plane algorithm to cluster all data in a time series into a number of time windows of unequal-length in both time and numerical domains is designed here．In each time window, a linear fuzzy information granule is established, then piecewise linear granulation representation of original time series is obtained．We introduce a distance measure for two linear information granules of unequal size to construct prediction of granular time series based on fuzzy inference．The proposed forecasting could complete long-term prediction for time series with pseudo-period．