摘要
文献[17]利用神经网络技术提出一种提前1小时预报电离层临界频率f0F2的方法。在该工作的基础上对网络的输入进行优化,将预报的提前量扩展到24小时。网络的输入包括地方时、季节、前24小时的观测值以及相应前30天的滑动平均值。分别用海口和北京站的历史数据进行检验,分析预报误差在太阳活动高低年和不同季节的变化,并将结果同国际参考电离层(IRI)进行比较。结果表明:神经网络的预报结果能较好地符合实测数据,在海口和北京站比IRI更具有实用性。
By using artificial neural network (NN), previous paper brings out'a method for forecasting the ionospheric critical frequency, f0F2, an hour in ad- vance. Based on this study, this paper optimizes the inputs of the network, fulfills to forecast the hourly values of f0F2 24 hours in advance. The inputs include time, season, preceding 24-hour measured and 30-day mean moving values. Historical da- ta at Haikou and Beijing is used to construct and checkout the network respective- ly. The prediction error which varies with solar activity and season is studied, and results between NN and the international reference ionosphere (IRI) model are also compared by giving their root-mean-square (RMS) errors. The results indicate that the prediction of NN has good agreement with observed data and is superior to IRI in Haikou and Beijing.
出处
《电波科学学报》
EI
CSCD
北大核心
2009年第1期152-156,共5页
Chinese Journal of Radio Science