摘要
采用动量BP神经网络算法拟合高程求解,训练样本的量纲存在一定的差异性,为改善不稳定性,通过对样本进行标准化处理,得到了理想的训练效果。将该方法得到的计算结果与平面拟合、二次曲面拟合及其他神经网络方法计算的结果进行对比得出:标准化动量BP神经网络算法求解高程,精度可靠且稳定。
When solving fitting elevation by using common momentum BP neural network algorithm there will be difference between the training samples in dimension. In order to improve the stability we standardized the samples and obtained the ideal results. It is shown that from the comparison among the results with the calculation methods by surface fitting, quadric surface fitting, and other neural network methods, the accuracy by using the stand- ardization of BP neural network algorithm for elevation fitting is more reliable and stable.
出处
《大地测量与地球动力学》
CSCD
北大核心
2010年第1期123-125,共3页
Journal of Geodesy and Geodynamics
关键词
动量BP神经网络
大地高
正常高
高程异常
标准化
momentum backpropagation neural network
geodetic height
normal height
elevation anomaly
stand- ardization
