Fluctuating Interval Number Series Forecasting Based on GM (1,1) and SVM

摘要

In this paper, a forecasting method for the fluctuating interval number series is proposed. Firstly, the definition equation of GM (1, 1) is improved to suit the interval number series directly and the interval number series needs not to be transformed into real number series. The development coefficient of the definition equation is taken as an accurate number, but the grey input of the definition equation is taken as an interval number. By this method, the binary interval GM (1, 1) (BINGM (1, 1)) is proposed. In order to further make the model fitting to the fluctuating interval number series, the support vector machine (SVM) is introduced to amend the predicted results of BINGM (1, 1) and SVMBINGM (1, 1) is proposed. In the process of amendment, the interval series is transformed firstly to avoid the disorder of the boundary points of the interval series. The results of the application example show that the prediction accuracy of SVMBINGM (1, 1) is higher than BINGM (1, 1) for the forecasting of fluctuating interval series.