Gnss Regional Atmospheric Delay Modeling And Precision Analysis Based On Artificial Neural Network

In GNSS observations,tropospheric delay and ionospheric delay are the major error source that affects positioning accuracy.The two delays are generally corrected by empirical models,which could not reach high accuracy due to the complex of parameters and the ambiguousness of internal mechanisms,thus inhibits the development of GNSS high-precision positioning services.In view of the artificial neural network's strong applicability,expandability and efficiency in nonlinear model processing,we applied it to GNSS regional tropospheric ZTD and ionosphere VTEC modeling.Then we analyzed the performance of the new model itself and its application in single point positioning and precision point positioning.The main work of the paper are as follows:1.A regional tropospheric delay model was established based on BP network.The CORS's data from Hong Kong and from Missouri,USA was used to build the troposphere prediction model,ZTDnet.Both data sets covered from the 1st day of 2014 to the 180 th day of 2018.The performance of Saastamonien,GPT2 w,UNB3m and ZTDnet in the two areas was evaluated.Results demonstrated that the root mean squared error of ZTDnet in the two areas was 4.98 cm and 5.40 cm.Compared with other models,ZTDnet reached higher accuracy especially in the low-latitude area.The research indicates that VTECnet performs stable and reliable in long-term prediction.Furthermore,taking the temperature into account,we optimized ZTDnet,whose accuracy was increased by 4.3%.2.A regional ionospheric delay model was established based on LSTM network.The CORS's data from Hong Kong and from Missouri,USA was used to build the ionospheric prediction model,VTECnet.Both data sets covered from the 1st day of2014 to the 180 th day of 2018.The performance of Klobuchar,Ne Quick,m Klobuchar,GIM,and VTECnet in the two areas was evaluated.Results demonstrated that the root mean squared error of VTECnet was 4.16 TECU and 1.76 TECU respectively.The accuracy of VTECnet is higher than Klobuchar,Ne Quick,and m Klobuchar.The research indicates that VTECnet performs stable and reliable in long-term prediction.3.A positioning approach based on regional atmospheric delay model was proposed.Using the observation data for 30 days from HKTP,a single frequency single point positioning scheme and a single frequency precision point positioning scheme were conducted for verification.The new model were compared with other three options,UNB3m+Klobuchar,Saastamonien+Klobuchar and no atmospheric delay correction.Results demonstrated that the new model,ZTDnet+VTECnet showed the highest accuracy,corresponding to the improvements of 11.0,7.0% for up component when comparing with the former two options in single point positioning,and of both16.2% for horizontal convergence rate and 11.1%,5.9% for north,east components in precise point positioning.