测绘通报 ›› 2022, Vol. 0 ›› Issue (3): 107-110.doi: 10.13474/j.cnki.11-2246.2022.0086

• 技术交流 • 上一篇    下一篇

基于BP神经网络的昆明天顶湿延迟模型

丁仁军1, 王友昆1,2, 张君华1, 刘晨2,3   

  1. 1. 昆明市测绘研究院, 云南 昆明 650051;
    2. 武汉大学测绘学院, 湖北 武汉 430079;
    3. 桂林理工大学广西空间信息与测绘重点实验室, 广西 桂林 541004
  • 收稿日期:2021-04-12 出版日期:2022-03-25 发布日期:2022-04-01
  • 通讯作者: 王友昆。E-mail:beanflower@whu.edu.cn
  • 作者简介:丁仁军(1974-),男,硕士生,高级工程师,研究方向为城市信息化测绘和CORS运维。E-mail:393263948@qq.com
  • 基金资助:
    国家自然科学基金(41721003;41874033);昆明市卫星定位综合服务系统整合技术服务项目(JS2020-03);昆明市卫星定位综合服务系统扩展升级专项资金(JS2021-02);昆明市卫星定位综合服务系统整合建设及关键技术研究(昆测研202003);广西空间信息与测绘重点实验室资助课题(19-185-10-17)

Kunming zenith wet delay model based on a backpropagation neural network

DING Renjun1, WANG Youkun1,2, ZHANG Junhua1, LIU Chen2,3   

  1. 1. Kunming Surveying and Mapping Institute, Kunming 650051, China;
    2. School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China;
    3. Guangxi Key Laboratory of Spatial Information and Geomatics, Guilin University of Technology, Guilin 541004, China
  • Received:2021-04-12 Online:2022-03-25 Published:2022-04-01

摘要: 为了满足昆明市卫星定位综合服务系统(KMCORS)对高精度天顶湿延迟(ZWD)的需要,本文开发了适用于昆明地区的ZWD模型KM。KM模型是根据昆明探空站2015-2018年的探空资料,基于误差反向传播(BP)神经网络建立的,同时采用2019年的探空数据,验证了KM模型的预测性能。测试结果表明,与广泛使用的SA模型相比,KM模型的RMSE由4.0 cm降至2.2 cm,精度提升了45%;KM和SA模型的Bias分别为0和-3.1 cm。该结果表明KM模型对ZWD估计具有无偏性,而SA模型在高原区存在过度估计的问题,KM模型具有比SA经验模型更优的预测性能,其应用将有助于提升KMCORS的服务质量。

关键词: 昆明CORS;天顶湿延迟;BP神经网络;Saastamoninen模型

Abstract: For the high-precision zenith wet delay (ZWD) used in Kunming continuously operating reference stations (KMCORS),this paper developes the Kunming model (KM) suitable for the KM area.According to the sounding data of the KM sounding station from 2015 to 2018,the KM model is generated based on a backpropagation (BP) neural network.This study then validates the prediction performance of the KM model using the sounding data during 2019.Test results show that the RMSE of the KM model decreases from 4.0 cm to 2.2 cm compared with the widely used Saastamoninen (SA) model,indicating its 45% accuracy improvement.Additionally,the Bias of the KM and SA models are 0 and-3.1 cm,respectively,suggesting that the ZWD estimation of the KM model is unbiased,while the SA model has the problem of overestimation in the plateau area.In summary,the KM model has better prediction performance than the SA empirical model,and the application of the KM model will help to improve the service quality of KMCORS.

Key words: Kunming CORS;zenith wet delay;backpropagation neural network;Saastamoninen model

中图分类号: