摘要
针对二次曲面模型在地形起伏较大区域用于GPS高程转换中存在较大的模型误差的问题,该文构建了二次曲面-RBF神经网络组合的GPS高程转换模型,组合模型中用二次曲面拟合高程异常中的中长波项,用RBF神经网络来泛化高程异常去除中长波后的残余项,并进行了二次曲面模型、RBF神经网络模型及二次曲面-RBF组合模型的实测数据GPS高程转换、比较分析与精度评定。实例结果表明:该组合模型比二次曲面模型的转换精度提高了22%,比RBF神经网络模型的转换精度提高了40%,该组合模型的转换方法可行、精度优于单一模型。
Aiming at big model error of quadric surface for GPS conversion in the region of terrain fluctuation,the combination model of quadric surface and RBF neural network for GPS height was established,in which quadric surface was used to fit the medium-long term,and RBF neural network to fit the residual after long wave being removed in abnormal height.On the basis of project data,the quadric surface,RBF neural network,and combination model were used in doing GPS height conversion,and transformation accuracy of GPS height was analyzed and evaluated.The project example showed that the combination model accuracy was improved by 22%than that of quadric surface model,by 40%than that of RBF neural network model,and transformation method based on combination model was feasible and had higher accuracy.
出处
《测绘科学》
CSCD
北大核心
2018年第2期34-38,共5页
Science of Surveying and Mapping
基金
中国煤炭工业协会2015年度科学技术研究指导性计划项目(MTKJ2015-313)
关键词
组合模型
GPS高程
RBF神经网络
二次曲面拟合
高程异常
combination model
GPS height
radial basis function neural network
quadratic surfacefitting
abnormal height
