结合BP神经网络优化GNSS高程拟合  

Comprehensive Application of BP Neural Network in GNSS Elevation Fitting

作　　者：高振耀 丛晓明 Gao Zhenyao;Cong Xiaoming(Department of Geological Engineering,Qinghai University,Xining 810002,China;Qinghai Institute of Geological Mapping and Geographic Information,Xining 810012,China;Qinghai Provincial Key Laboratory of New Geographic Information Technology for Plateau Surveying and Mapping,Xining 810012,China;College of Geographical Sciences,Qinghai Normal University,Xining 810008,China)

机构地区：[1]青海大学地质工程系,西宁810002 [2]青海省地质测绘地理信息院,西宁810012 [3]青海省高原测绘地理信息新技术重点实验室,西宁810012 [4]青海师范大学地理科学学院,西宁810008

出　　处：《青海科技》2023年第3期83-90,共8页Qinghai Science and Technology

基　　金：青海省重点研发与转化计划项目(编号:2022-QY-224);青海省地矿局科研项目“高原重大滑坡协同监测技术研究”共同资助。

摘　　要：为了将GNSS测量得到的大地高精确、高速、稳定地转化为地面正常高,文章基于GNSS数据高程模拟实验,进行了不同拟合方法的对比,提出一种神经网络组合模拟方法。将基于神经网络的组合拟合算法的拟合精度与二次曲面拟合法、人工神经网络拟合法的拟合结果进行分析对比,结果表明:神经网络的组合拟合算法可结合二次曲面拟合法、人工神经网络拟合法两者的优势,降低两种单一方法自身导致的误差,网络性能稳定。并且当学习集训练中误差取2 mm时,工作中误差值最小,能够满足实际工程应用。In order to accurately,rapidly and stably convert the geodetic height measured by GNSS into the normal height of the ground,based on the GNSS data elevation simulation experiment,the comparison of different fitting methods is carried out,and a neural network combination simulation method is proposed.The fitting accuracy of the combined fitting algorithm based on neural network is analyzed and compared with the fitting results of quadratic surface fitting method and artificial neural network fitting method.The results show that the combined fitting algorithm of neural network can combine the advantages of quadratic surface fitting method and artificial neural network fitting method,reduce the error caused by the two single methods,and the network performance is stable.And when the error in the learning set training is 2mm,the error value in the work is the smallest,which can meet the practical engineering application.

关 键 词：GNSS 高程拟合 BP神经网络 拟合算法 

分 类 号：P228.4[天文地球—大地测量学与测量工程]