论文部分内容阅读
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。
In this paper, neural network method is used to convert GPS elevation to normal height or normal height. An improved BP neural network topology and algorithm are proposed. 43 sample sets are constructed by using the actual GPS positioning data. The accuracy of the estimation is up to Centimeter level. Finally, the neural network method and quadratic polynomial surface fitting GPS elevation geoid leveling method is compared, the accuracy of the neural network method is better than quadratic polynomial surface fitting method, and the accuracy is relatively stable, the number of known sample points Less demanding.