摘要
提出一种基于广义回归神经网络的电离层电子总含量建模的新方法。依据电子总含量的时空变化特性建立基于广义回归神经网络的区域电子总含量模型。结合实例,详细讨论训练样本的采样策略对网络模型性能的影响,并确定较优的模型光滑参数和采样策略。分别从理论和实例上与常用的多项式模型进行对比分析。结果表明在数据样本密集区域两者的精度相当,而在外推的空白区域内网络模型的精度优于多项式模型,验证网络模型的可行性和有效性。
A new idea is for model ionosphere vertical total electron content based on generalized regression neural network. Firstly, this new model is established according to time and space characteristics of vertical total electron content. Then, we have discussed how sampling strategy influences network model performance. An excellent sampling strategy and smooth coefficients were determined according to investigation scenario. Finally, the feasibility and availability of network model are proved based on the comparing analysis between network model and polynomial model. Results show that the accuracy of network model in ample data region is similar with polynomial model, and which is better than polynomial model in the extrapolated blank data region.
出处
《测绘学报》
EI
CSCD
北大核心
2010年第1期16-21,共6页
Acta Geodaetica et Cartographica Sinica
基金
国家863计划(2007AA12Z307)
关键词
电子总含量
广义回归神经网络
采样
模型精度
total electron content
generalized regression neural network
sampling
model accuracy
