Design of adaptive prediction system based on rough sets

Abstract

In this paper, a multiple prediction system using T-S fuzzy model is presented for time series forecasting. To design predictors with better performance especially for chaos or nonlinear time series, difference data were used as their input, because they reveal the statistical patterns and the regularities concealed in time series more effectively than the original data can. The proposed method consists of three major procedures. First, multiple model fuzzy predictors (MMFPs) are constructed based on the optimal difference candidates. Next, an adaptive drive mechanism (ADM) based on rough sets is designed for the selection of the best one among the multiple predictors according to each input data. Finally, an error compensation mechanism (ECM) based on the cross-correlation analysis is suggested in order to enhance further the prediction performances. Also we show the effectiveness of the proposed method by computer simulation for the various typical time series.