电离层f_0F_2参数提前24小时预测

Twenty-four Hour Ahead Prediction of f_0F_2

利用中国9个垂测站(海口、广州、重庆、拉萨、兰州、北京、乌鲁木齐、长春、满洲里)一个太阳周(1976—1986年)的数据资料,采用三层前向反馈神经网络(BP网络)实现了电离层F_2层临界频率(f_0F_2)参数提前24h预测。通过对f_0F_2参数的时间序列及其与日地活动之间进行相关分析,确定f(t) (当前时刻f_0F_2)、经过变换的F_(10.7)指数等5个参数作为神经网络的输入参数,并通过同时段训练法获得不同时刻的预测值,本文与自相关分析法进行了预测性能比较。结果表明,上述方法构建的神经网络可以达到较高的预测精度。针对暴时数据,对神经网络算法进行了改进,提高了神经网络法对暴时数据的适用性。

Using three-layer feed-forward back propagation neural networks, twenty-four hour ahead prediction of the critical parameter of ionospheric F2 layer （f0F2） is realized. The prediction model is developed based on 11 years of data （from 1976 to 1986） measured from China vertical station （Haikou, Guangzhou, Chongqing, Lhasa, Lanzhou, Beijing, Urumchi, ChangChun, Manchuria）. By analyzing time series correlation of foF2 and solar-terrestrial activity, five input parameters are determined. The same-time training method is selected and the prediction values within 24 hour can be obtained without changing the network frame. By comparing the prediction property of Neural Network （NN） method and the autocorrelation one （named Corr）, for quite data the NN method has higher accuracy except for summer data. While for the whole year data set, the Corr is better. In order to improve the applicability of the method for storm-time data, NN is corrected, and using two specified examples to explain the improvement in the article. After such modification, NN is better than Corr for the same test data as that used above.