摘要
以广东省番禹区沙洲和石楼镇岛地区的1 657个高程点为样本点,把其分为A、B、C组各200个高程点,A+B组400个高程点,A+B+C组600个高程点作为训练数据集,在Matlab 7.1和ArcGIS 9.2平台上分别应用广义回归神经网络(GRNN)、普通克里格(O-Kriging)、广义回归神经网络残余Kriging方法(GRNNRK)进行高程估值和成图,最后计算出三种方法的均方根误差.结果表明,如果插值样本数据量不变,样本的空间分布格局对GRNNRK插值精度的影响不大,且其插值精度要优于GRNN和O-Kriging方法的插值精度.随着插值样本数据量的增加,三种方法的插值精度都有显著提高,但GRNNRK方法的插值精度仍优于另两种方法.这表明GRNNRK方法在地形高程预测中的应用是可行的.
In the paper, the 1 657 elevation points in the area of Shazhou and Shilou Town island in Fanyu District of Guangdong province were taken as sample points and split into group A, B, C each 200 points, A ＋ B 400 points and A ＋ B ＋ C 600 points separately as training data set. Then the authors estimated the surface elevation using General Regression Neural Network （GRNN）, Ordinary Kriging （O-Kriging）and General Neural Network Residual Kriging （GRNNRK）respectively and draw the maps of interpolation results on the basis of Matlab 7. I and ArcGIS 9.2 . At last, they calculated the root mean squared errors（RMSE） of the three methods. The result showed that if the number of sample points remains unchanged, the distribution pattern of sample points did not practically produce an effect on the precision of GRNNRK, and also the GRNNRK was superior to the methods of GRNN and O-Kriging. With the increasing of sample points, the precision of the three methods were also increasing, but GRNNRK was also superior to the other two methods. It showed that it was available to use the method of GRNNRK to estimate the surface elevation.
出处
《安徽大学学报（自然科学版）》
CAS
北大核心
2010年第5期21-26,共6页
Journal of Anhui University(Natural Science Edition)
关键词
广义回归神经网络
克里格
残余
地表高程预测
general regression neural network
Kriging
residual
surface elevation
