太阳黑子数平滑月均值的混合预测模型

A hybrid prediction model of smoothed monthly mean sunspot number

提出一种太阳黑子数平滑月均值的混合预测模型.通过最大Lyapunov指数得到太阳黑子数平滑月均值时间序列的最大可预测周期,结果表明太阳黑子数平滑月均值序列的最大可预测周期为42个月.太阳黑子数平滑月均值时间序列中包含着线性与非线性的成分,利用自回归滑动平均模型对线性成分进行预测,将太阳黑子数平滑月均值的实际值与自回归滑动平均模型的预测值作差值得到仅含有非线性成分的残差序列,利用具有良好非线性预测能力的回声状态网络预测残差序列,并通过人工蜂群算法来确定回声状态网络预测模型的最佳参数.将自回归滑动平均模型预测值与回声状态网络预测的残差相加,得到太阳黑子数平滑月均值的最终预测值.通过第23太阳活动周的太阳黑子数平滑月均值的预测表明提出的预测模型具有较高的预测精度.同时,对第24太阳活动周的太阳黑子数平滑月均值进行了预测,结果表明第24太阳活动周将在2020年2月结束.

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.