摘要
针对函数模型和BP(back propagation)神经网络等常用GPS高程拟合方法模型单一、拟合精度不高的问题,本文在上述模型高程异常拟合的基础上使用RBF(radial basis function)神经网络对拟合残差值进行二次拟合,对高程异常拟合值进行残差改正以提高拟合精度。从内外符合精度、拟合残差大小分布等方面对组合模型和单一模型拟合结果进行对比,结果表明:组合模型的拟合精度相较于函数模型有显著提高,但对BP神经网络的拟合结果改善不明显。
The common GPS height fitting methods such as function model and BP neural network are single and have low fitting accuracy.This paper uses RBF neural network to fit the residual value based on the above models,and corrects the residual value of the height anomaly fitting value to improve the fitting accuracy.The combined model and the single model are compared from the internal and external coincidence accuracy and the fitting residual size distribution.The results show that the fitting accuracy of the combined model is significantly higher than that of the function model,but the improvement of the BP neural network is not obvious.
作者
岳春芳
宋金元
YUE Chunfang;SONG Jinyuan(College of Hydraulic and Civil Engineering of Xinjiang Agricultural University,Urumqi 830052,China)
出处
《测绘地理信息》
2020年第1期20-22,共3页
Journal of Geomatics
基金
新疆教育厅创新项目(XJEDU2017T004).
关键词
GPS高程拟合
函数模型
神经网路
组合模型
GPS elevation fitting
function model
neural network
combination model
