Mining Self-similarity in Time Series

Abstract

Self-similarity can successfully characterize and forecast intricate, non-periodic and chaos time series avoiding the limitation of traditional methods on LRD (Long-Range Dependence). The potential principals will be found and the future unknown time series will be forecasted through foregoing training. Therefore it is important to mine the LRD by self-similarity analysis. In this paper, mining self-similarity of time series is introduced. And the practical value can be found from two cases study respectively for seasonvariable trend forecast and network traffic.