| 基于改进的BP神经网络构建区域精密对流层延迟模型 |
| |
| 引用本文： | 肖恭伟,欧吉坤,刘国林,张红星.基于改进的BP神经网络构建区域精密对流层延迟模型[J].地球物理学报,2018,61(8):3139-3148. |
| |
| 作者姓名： | 肖恭伟 欧吉坤 刘国林 张红星 |
| |
| 作者单位： | 1. 山东科技大学, 测绘科学与工程学院, 山东青岛 266590;2. 中国科学院测量与地球物理研究所, 大地测量与地球动力学国家重点实验室, 武汉 430077 |
| |
| 基金项目： | 国家自然科学基金（41574015）资助. |
| |
| 摘 要： | 利用神经网络算法挖掘海量数据的规律已成为科技发展的一种趋势，本文针对卫星信号的天顶对流层延迟进行建模.对流层延迟是影响卫星定位精度的重要因素之一，建立精密区域对流层模型对高精度定位有着重要的意义.对区域测站对流层延迟数据的分析，考虑到实时建模中传统BP（Back Propagation）神经网络计算量大，易出现"过拟合"现象、不稳定等因素，通过改进的BP神经网络建立了区域精密对流层模型.详细介绍了新模型的建立过程，并与常用的对流层区域实时模型进行了对比.还讨论了建模测站数目对预报精度的影响.相比现有的其他对流层延迟模型，基于改进的BP神经网络构建的区域精密对流层延迟模型无论在拟合和预报方面都有较好的精度，且随着测站数目的增加模型精度趋于平稳.改进的模型参数较少，可以进行实时的区域精密对流层延迟改正；需要播发的信息量小，适用于连续运行参考站系统（Continuously Operating Reference Stations，CORS）的应用.研究表明：改进的BP神经网络模型能够更好的充分利用大规模历史数据描述卫星信号对流层延迟的空间分布情况，适用于实时大区域精密对流层建模.基于日本地区2005年近1000多个测站的NCAR（National Center Atmospheric Research）对流层数据进行区域对流层延迟建模，结果表明改进的BP神经网络模型在拟合和预报精度上都有较大提升，RMSE（Root Mean Square Error）分别为：7.83 mm和8.52 mm，而四参数模型拟合、预报RMSE分别18.03 mm和16.60 mm.
|
| 关 键 词： | BP神经网络 区域对流层延迟 CORS 拟合模型 |
| 收稿时间： | 2017-09-04 |
Construction of a regional precise tropospheric delay model based on improved BP neural network |
| |
| XIAO GongWei,OU JiKun,LIU GuoLin,ZHANG HongXing.Construction of a regional precise tropospheric delay model based on improved BP neural network[J].Chinese Journal of Geophysics,2018,61(8):3139-3148. |
| |
| Authors: | XIAO GongWei OU JiKun LIU GuoLin ZHANG HongXing |
| |
| Institution: | 1. Geomatic Science and Engineering, Shandong University of Science and Technology, Shandong Qingdao 266590, China;2. State Key Laboratory of Geodesy and Earth's Dynamics, Institute of Geodesy and Geophysics of CAS, Wuhan 430077, China |
| |
| Abstract: | Using the neural network algorithm to excavate massive data has become a trend of technological development. Zenith tropospheric delay of satellite signal is modeled in this paper. The tropospheric delay is one of the important factors that affect the accuracy of satellite positioning, so establishing a precise regional troposphere model is of great significance to high-precision positioning. Based on the analysis of the tropospheric delay data from regional stations, considering the large amount calculation of traditional Back Propagation (BP) neural network in real-time modeling, easy overfitting and instability, we use the improved BP neural network to establish this model. The derivation process of the new model is presented in detail, and the new model is compared with common regional real-time models of troposphere. The influence of the number of modeling stations on the prediction accuracy is also discussed. Compared with other existing tropospheric delay models, this new model has good accuracy of both fitting and prediction, and the accuracy of the model is stable as the number of stations increases. The parameters of the improved model are less, and can be used in real-time regional precision troposphere modeling. The amount of information needed to broadcast is small, and is applicable to Continuously Operating Reference Stations (CORS) network. Research has shown that the improved BP neural network model can better make full use of large-scale historical data to describe the spatial distribution of the tropospheric delay of satellite signals, and can be applied to real-time region precision troposphere modeling. Based on the National Center Atmospheric Research (NCAR) troposphere data for nearly 1000 stations in Japan in 2005, regional tropospheric delay modeling results show that the improved BP neural network model has advantages in fitting and prediction accuracy, with Root Mean Square Errors (RMSE) of 7.83 mm and 8.52 mm, respectively, while the four-parameter model's RMSEs are 18.03 mm and 16.60 mm, respectively. |
| |
| Keywords: | BP neural network Regional tropospheric delay CORS Fitting model |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《地球物理学报》浏览原始摘要信息 |
| 点击此处可从《地球物理学报》下载免费的PDF全文 |
|
