用神经网络方法转换GPS高程

Conversion of GPS Height by Artifical Neural Network Method

本文提出用神经网络方法转换ＧＰＳ高程为正高或正常高，给出一种改进了的ＢＰ神经网络拓扑结构和算法，并用ＧＰＳ的实际定位资料构成４３ 个样本集作了计算分析，估算的精度达到厘米级。最后用神经网络方法与二次多项式曲面拟合大地水准面转换ＧＰＳ高程的方法作了比较，神经网络方法的精度优于二次多项式曲面拟合法，而且精度比较稳定，对已知样本点的数量要求较少。

At present, the coordinate system of GPS is the World Geodetic System 84.Positions determined by GPS receiver are put in geocentric coordinate or geodetic coordinate defined by WGS 84 ellipsoid, but in engineering application, the coordinates need to be converted for local coordinate system, and ellipsoidal height need to be converted for physical height, viz. orthometric height or normal height. Ellipsoidal height conversion to physical height is difficult, because this conversion bears on determination of an unknown Geoid, which relates to the distribution of matter in earth. In this paper a conversion method of GPS geodetic height to orthometric height or normal height using artificial neural network is proposed. The method avoids puzzle for determination of Geoid, establishes mapping relation from GPS ellipsoidal height to orthometric height or normal height by dint of neural network learning function. An improved BP neural network topological architecture and algorithm are also given in this paper. Using the neural network the GPS height in a survey area is converted to test, the results show that accuracy of conversion is about the order of 2cm. This method is also compared with fitting method that simulates Geoid (or Quasi Geoid) using quadric polynomial for conversion of GPS height. The accuracy of artificial neural network method is better than that of fitting method. The known sample amount in request is also much less.