A hybrid prediction model of smoothed monthly mean sunspot number
Abstract
A hybrid prediction model of smoothed monthly mean sunspot number is proposed in this paper. The maximum predictable time span of smoothed monthly mean sunspot number is calculated by the maximum Lyapunov exponent method. The results show that the maximum predictable time span of smoothed monthly mean sunspot number is 42 months. The smoothed monthly mean sunspot number time series has linear and nonlinear components. Auto regressive moving average model is utilized to predict the linear component. The nonlinear error component can be obtained by the actual value of smoothed monthly mean sunspot number minus the prediction value by auto regressive moving average model. Then, echo state network with good nonlinear prediction ability is utilized to predict nonlinear error component. At the same time, artificial bee colony algorithm is used for parameters optimization of echo state network prediction model. The final prediction value of smoothed monthly mean sunspot number will be obtained by the prediction value of auto regressive moving average model added to the prediction value of echo state network. The simulation of the twenty-third solar cycle smoothed monthly mean sunspot number shows that the proposed model has higher prediction accuracy. At the same time, smoothed monthly mean sunspot number of twenty-fourth solar cycle is predicted, the results show that the twenty-fourth solar cycle will end in February 2020.