A novel forecasting model for the long-term fluctuation of time series based on polar fuzzy information granules

The long-term fluctuation of time series is generally composed of a large number of shortterm behaviors with various dynamical characteristics, where kinds of fluctuation patterns in different periods change mutually. In this paper, we propose a novel method to construct fuzzy information granules in polar coordinates and achieve the prediction of longterm fluctuation of time series on the basis of the short-term fluctuation patterns. Firstly, time series are divided into segments by means of the sliding time windows, and fuzzy information granules are defined based on the regression models to indicate the fluctuation patterns of segments of time series. The transfers among different information granules form a dynamical network containing rich inference information. Next, the constructed networks are analyzed to capture the transfer characteristics of fuzzy information granules. The results show that only a few types of fuzzy information granules and fuzzy relation groups play the key role in the fluctuation mechanism, which always have specific targets. Hence, according to the distribution of the transfer probability, a prediction scheme on the granularity level can be established. By utilizing both synthetic and real-life data sets, examples are shown to illustrate the effectiveness and feasibility of the proposed scheme. (C) 2019 Elsevier Inc. All rights reserved.