电离层参数短期预报研究进展  

Development on Short-term prediction of Iononspheric Parameters

作　　者：孔庆颜[1] 冯静[1] 柳文[1] 凡俊梅[1] 

机构地区：[1]中国电波传播研究所青岛分所,电波环境特性及模化技术国家重点实验室,青岛266107

出　　处：《地球物理学进展》2010年第6期1968-1976,共9页Progress in Geophysics

基　　金：国家基础研究项目;国家自然科学基金项目(60771049)共同资助

摘　　要：对国内外电离层参数短期预报方法进行了综述,重点介绍了几种作者最新研究的电离层f_0F_2参数短期预报方法.包括基于混沌时间序列分析的电离层f_0F_2参数提前15min(分钟)准实时预测方法、基于人工神经网络技术的提前1h(小时)现报方法、提前1～3d(天)的神经网络预测方法、相似日短期预报方法以及综合预报模型方法.利用中国垂测站多年的观测数据对各种算法的预测精度进行了评估,并与国内外相关算法进行了定性或定量比较,各种预报方法都在前人的预报精度基础之上有了一定的提高.其中提前15min(分钟)预测方法平均相对误差小于4%,平均绝对误差小于0.2MHz,可以用于实时性和精度要求较高的短波系统;提前1小时预报方法在太阳活动高年平均预测相对误差小于6%,均方根误差小于0.6MHz,太阳活动低年平均预测相对误差小于10%,均方根误差小于0.5MHz,平均相对误差比前人研究的自相关方法提高3个百分点左右;对于提前1～3d(天)短期预报,综合预报模型方法充分利用了神经网络方法、自相关方法以及相似日方法的优点,获得了高于任何一种单一方法的精度,对于中国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)在不同太阳活动性条件下的历史数据进行了精度测试,提前1天和提前3天预测的平均相对误差分别小于10%和小于15%,达到了国内先进水平.此外,该方法还可以综合更多预报方法,具有进一步提高预报精度的潜力.文中提出的针对不同尺度进行电离层参数预测的方法具有一定的理论基础,且精度高、易实现,对从事电离层短期预报算法研究及相关专业的学者具有一定的参考价值.We review the ionospheric parameter short-term prediction methods and present several methods developed by the authors, including quasi-real time prediction （15-minute ahead）based on chaotic time series analysis, nowcasting （one-hour ahead） using artificial neural networks and short-term prediction （1 to 3 day ahead）, which is realized by several algorithms like neural network, similar-day and integrated model. By using years of data form ionospheric observation stations in China, the prediction accuracy of all the methods is examined, compared with other existent methods qualitatively or quantitatively. All the prediction methods developed by authors can reach high accuracy and are better than that of predecessorsr. The average relative error for 15 min ahead prediction is less than 4%, and the absolute one is less than 0.2 MHz. Such precision is high enough for those short wave systems which need high accuracy and real time. The method to forecast foF2 one-hour ahead can also reach high accuracy. During high solar activity, the average relative error is less than 6% and RMS less than 0. 6 MHz, while during solar minimum, the average relative error less than 10% and RMS less than 0.5 MHz and the average relative error for all the conditions is 30//oo lower than that of the autocorrelation one. As to forecasting one-to-three day ahead, the integrated model has the highest precision because it takes full advantage of the neural network method, similar-day method and the autocorrelation one. The method is tested with data of nine inland vertical stations （Haikou, Guangzhou, Chongqing, Lhasa, Lanchow, Beijing, Urumchi, Changchun, Manchuria）, covering one solar periods （1976~ 1986） and the average relative errors for one day and three day ahead are less than 10% and 15% respectively, keeping ahead in China. Besides, the integrated model can also integrate other methods, having potential to improve the precision. All the methods mentioned in the article are all easy to be implemented, with high acc

关 键 词：短期预报 电离层 人工神经网络 FOF2 综合模型 

分 类 号：P352[天文地球—空间物理学]