传统的BP神经网络在GPS高程异常拟合应用有其一定的局限性,特别是在对于外推高程异常值方面,传统的BP神经网络的不足表现得尤为明显.针对高程异常的特性,既有趋势性也有随机性,结合BP神经网络的优点,提出一种改进型的BP神经网络高程异常拟合方法,利用最小二乘配置法综合考虑高程异常的趋势性和随机性的特点,采用BP神经网络方法对包括最小二乘配置法的模型误差的综合误差进行优化减弱,最后可得到新的高程异常.通过实例,将文中提出的新方法与曲面拟合法以及传统BP神经网络拟合法在内插和外推2方面进行比较分析,结果表明文中提出的新方法拟合效果最佳.%Traditional BP neural network has its limitations in the application of fitting GPS height anoma ly. Especially in regard to extrapolate the value of height anomaly, the lack of traditional BP neural net work has been particularly evident. Because of the characteristics of height anomaly, which has tendency and randomicity, combined with the advantages of BP neural network, an improved BP neural network height anomaly fitting method is presented in this paper. Least squares collocation method for comprehen sive consideration the tendency and randomicity of height anomaly is used and then get a new height anom aly by BP neural network fitting height anomaly errors. The new method presented in the paper has been demonstrated that the accuracy of this fitting model is better than surface fitting mode and traditional BP neural network fitting mode through comparative analysis in interpolation and extrapolation.
展开▼
