Research On The Forecast Of Solar Active Indexes Based On EMD Algorithm

Solar activity index reflects the overall level of the earth affected by the high frequency electromagnetic radiation from the sun. It is an important content of the solar activity forecast. With the further study of the forecast of the solar activity index, the requirement of the forecast accuracy is also getting higher and higher. The research object of this paper is the sunspots and solar10.7cm radio flux number-two important solar activity indexes. The time series values of them are strongly nonlinear and non-stationary characteristics. Therefore, this paper will use the ability of the EMD algorithm to decompose the non-stationary signal and artificial intelligence method to predict the ability of nonlinear time series to forecast and analysis the sunspots and solar10.7cm radio flux. It improves the prediction accuracy, and the significance application has the popular value. The main contents and progress in this paper is as following:(1) A new method based on Empirical Mode Decomposition (EMD) and Radial Basis Function (RBF) neural networks is proposed to forecast the monthly mean sunspot numbers from2002to2012. The results show that98%of the relative error of prediction is mostly smaller or equal to1.0, and95%of the absolute error can be controlled under-20～20, the maximum absolute error is less than35, the standard deviation is7.63. Comparing the three evaluating indicators:SD,RMSE,MAE, the results show that the prediction of the proposed method is more accuracy than the method of RBF neural network used by NSMC.(2) We use the EMD-SVM method to predict short-term F10.7from2012.1.1to2012.11.9.The results indicate:Each of mean relative error of next three days is2.5%、3.6%and4.3%; The new method has better performance than the single method in prediction accuracy; Comparing to the F10.7predictions released by NADC from2012.9.1to2012.11.9, the relative error of next three days respectively reduces5.99%、7.72%and8.10%.