Finding periodicity in pseudo periodic time series and forecasting

Abstract

In this paper, pseudo periodic time series are studied. A novel fuzzy granulation based approach is proposed for that. Periodicity discovery and forecast on pseudo periodic time series are concentrated on. Two definitions of periodicity, λ-Pseudo periodicity and generalized λ-Pseudo periodicity are given, and two corresponding algorithms for finding these two kinds of periodicity are designed. These studies are carried on by employing a similarity measure defined on granular value level. Furthermore, we study the forecasting about the development of pseudo periodic time series according to the proposed algorithms. Experiments are given to demonstrate the algorithms. The experimental results show the efficiency of the algorithms, and verify the rationality of the novel fuzzy granulation based approach for studying pseudo periodic time series.