The medium-term prediction of smoothed monthly sunspot numbers is examined on the basis of the theory of nonlinear dynamical systems in this paper. A local linear model is constructed using a technique called the inverse problem of chaotic dynamical systems, reconstructing the d-dimensional phase space, utilizing k neighbour points dosed to the current point in phase space on the basis of the achieved dynamical parameters of fractal dimension D and the largest Lyapunov exponent lambda(1) for the monthly mean sunspot numbers. The optimal parameters predicting monthly mean sunspot numbers, the embedding dimension d and the number of neighboring points k, are determined examining the mean error for the prediction of one time step ahead, to obtain an optimal nonlinear statistical model. To reduce the drop in the accuracy of the medium-term nonlinear prediction caused by chaotic behavior, the predictions for the later time are corrected or compensated by means of the largest Lyapunov exponent lambda(1) and the predictive error obtained by comparing the predicted with the observed values. The final results are obtained using the formula of the smoothed monthly sunspot numbers. Our predictions of 5 months ahead for the period from August 1989 to the present are compared with the observed values and the corresponding predictions given by SIDC. It is shown that this nonlinear predictive technique is feasible and has broad prospects. This technique can become a better candidate for medium-term prediction of the smoothed monthly sunspot numbers.
; Qin, Z
修改评论