BP神经网络在县域GPS高程拟合中的应用

Application of BP neural network for height fitting in county area

本文针对GPS大地高与水准正常高相互转换的问题,根据实测GPS、水准数据,构建BP神经网络模型拟合新化县域GPS高程异常。通过测试样本拟合高程与正常高的比较及与常用的曲面拟合法对比发现,BP神经网络模型高程拟合最大残差为5.4cm,最小为0.5cm,拟合中误差为3.2cm,均优于曲面拟合法。通过与已有的湖南省区域似大地水准面精化结果比较,BP神经网络模型在新化县域拟合精度优于省级似大地水准面精度。因此,BP神经网络模型在该县域的GPS高程拟合是可行的。根据BP神经网络模型拟合的高程异常分布发现,新化县域GPS高程异常曲面具有西南、中部、北部高、东南及中部偏北柘溪水库库区低的分布特征,其中水车镇至金凤、油溪桥至官帽及大熊山林场均为高程异常高梯度带,柘溪水库库区及东南相邻冷水江市为高程异常低梯度带,这与该地区地形成镜像关系。高程异常最大值为19.35m,最小值为18.41m,拟合区域高程异常差值0.95m。

Targeting at the conversion of GPS ground high to the normal high,this paper constructs a BP neural network model,based on the measured GPS height and standard data,to fit the GPS height anomaly in Xinhua County area.By comparing the height of the test sample with the know normal height by the commonly used surface fitting method,it is observed that the maximum residual of the BP neural network model is 5.4 cm,the minimum is 0.5 cm and the standard deviation is 3.2 cm in the fitting,which are all superior to the results of the surface fitting method.Furthermore,compared with the results of the existing regions in Hunan Province,it is found that BP neural network model has better accuracy than that of the provincial level.Therefore,the BP neural network model of GPS height fitting is feasible in this county.The height anomaly distribution of BP neural network model manifests that the height anomaly surface of GPS in Xinhua County has the characteristics of low distribution in the Tuoxi reservoir area of southeast and central north,and high distribution in southeastern,central and northern parts.The high gradient zone of high anomaly covers Shuiche Town to Jinfeng Town,from Youxi Bridge to Guanmao and Daxiongshan forest,whereas the low gradient zone of high anomaly covers the Tuoxi reservoir area and the neighboring city Lengshuijiang in the southeast which fosters a mirror image relationship with the topography of the region.The maximum height anomaly is 19.35 m,and the minimum is 18.41 m,the range of height anomaly in this region is 0.94 m.