神经网络方法在太阳质子事件短期预报中的应用

在大量统计结果的基础上,深入研究了太阳质子事件预报机理．总结了质子事件爆发与太阳活动区面积、位置、McIntosh结构、磁结构以及前两天活动区爆发耀斑事件数目之间的关系．然后,在神经网络的基础上建立了太阳质子事件短期预报模型,并对2000年以后12个未参加训练的样本进行测试,结果对事件预报的准确率为83％．此外,我们还利用该模型对2002年1-4月发生的几次质子事件进行了预报试验,结果发现,这期间发生的6次事件都被预报．其中3次质子事件系统预报提前了3天,两次事件预报提前了2天,一次事件提前1天预报．

SHORT TERM FORECAST OF SOLAR PROTON EVENT

Solar Proton Events (SPE) are one kind of the major, severe space environmental events that induces errors even failure of the satellites on the orbit. In this paper, a model of SPE forecasting employing artificial neural network technology by Space Environment Prediction Center (SEPC) is introduced. The strength of solar event is related with releused from the solar burst. The size of area of the sunspot indicates the maximum strength of local magnetic field. The Mclntosh classification of the sunspot indicates the structure of the root area of the magnetic rope from the sunspot regions. The joint applying of the Mclntosh classification and the magnetic type enable us to describe the structure of the sunspot. In this paper, 10.7cm radio and X-ray fluxes were used, and the Artificial Neural Network (ANN) is employed. The digitized morphology data are then arranged and "feed" into the neural network. Through training the model based on ANN algorithm a forecast model was constructed. With it, the forecast of SPE 1-3 days ahead can be made. Statistics shows that the accuracy of our forecast is about 80%.