论文部分内容阅读
传统BP神经网络在GPS高程转换中存在一定局限性,特别是在外推精度方面。为避免几何曲面模型不能贴合的情况及神经网络在训练时产生局部极小概率的问题,文中提出一种改进的BP神经网络高程异常拟合方法,即LS-SVW/BP神经网络组合模型,通过实例对该组合模型与BP神经网络、二次曲面拟合、平面拟合进行比较,说明了该组合模型用于GPS高程转换的可行性和优越性。
The traditional BP neural network has some limitations in GPS elevation conversion, especially in extrapolation accuracy. In order to avoid the non-conformity of geometric surface model and the local minimum probability of neural network in training, an improved BP neural network anomaly fitting method is proposed in this paper, which is LS-SVW / BP neural network combined model, An example is given to compare the combination model with BP neural network, quadric surface fitting and plane fitting. The feasibility and the superiority of the combined model for GPS elevation conversion are illustrated.