采用随机森林方法提高区域GNSS高程拟合精度  

Improve the Fitting Accuracy of Regional GNSS Elevation by Using Random Forest

作　　者：丁渃鹏 杨久东 张凌云 陈晓东[2] DING Ruo-peng;YANG Jiu-dong;ZHANG Ling-yun;CHEN Xiao-dong(College of Mining Engineering,North China University of Science and Technology,Tangshan Hebei 063210,China;Innovation Academy for Precision Measurement Science and Technology,Chinese Academy of Sciences,Wuhan Hubei 430071,China)

机构地区：[1]华北理工大学矿业工程学院,河北唐山063210 [2]中国科学院精密测量技术与创新研究院,湖北武汉430071

出　　处：《华北理工大学学报（自然科学版）》2024年第4期74-81,共8页Journal of North China University of Science and Technology：Natural Science Edition

基　　金：国家自然科学基金(42004075):基于地心天体位置潮汐诱发地震研究和新诱发机制提出;国家自然科学基金(41974023):月球固体潮汐理论推导及其数值计算。

摘　　要：本研究针对无实测重力数据和高分辨数字高程模型(DEM)数据下如何提高GNSS(Global Navigation Satellite System)高程拟合精度问题,通过采用反向传播(BP)神经网络与随机森林两种机器学习方法,与目前广泛采用的多项式二次曲面法的拟合结果进行了对比。研究结果表明,在地形起伏较为平缓的区域内,机器学习方法在外符合精度上至少有1厘米精度上的提升相比多项式二次曲面法或顾及地形起伏的多项式二次曲面法;在地形起伏较大的区域内,机器学习方法在外符合精度上均至少有1厘米精度上的提升。用均方根误差的标准差来衡量网络的稳定性,研究结果表明随机森林方法的标准差在试验区内优于BP神经网络。综上,机器学习方法可提升GNSS高程拟合精度、随机森林法的网络稳定性优于BP神经网络。This study focuses on how to improve the accuracy of GNSS elevation fitting without measured gravity data and high-resolution digital elevation model data.By using two machine learning methods of Back Propagation(BP)Neural Eetwork and Random Forest,the fitting results were compared with those of polynomial quadric surface method,which was widely used at present.The results show that in the area with gentle terrain undulation,the machine learning method has at least 1 centimeter accuracy improvement compared with the polynomial quadric surface method or the polynomial quadric surface method considering terrain correction.In the area with large terrain undulation,the machine learning method has an improvement of at least 1 centimeter accuracy in the external coincidence accuracy.The stability of the network is measured by the standard deviation of the Root Mean Square Error.The comparison of the results of the two machine learning methods in the study shows that the standard deviation of the results of the Random Forest method is better than that of the BP Neural Eetwork in the test area.The conclusion shows that the machine learning method can improve the accuracy of GNSS elevation fitting and the network stability of Random Forest method is better than that of BP Eeural Network.

关 键 词：GNSS高程拟合 机器学习 高程异常 GNSS水准高程 水准测量 

分 类 号：P228.4[天文地球—大地测量学与测量工程]