Models And Methods Of Data Processing For Multi-GNSS Real-time Precise Point Positioning

Real Time Precise Point Positioning(RT-PPP)is a research hotspot in the field of Global Navigation Satellite System(GNSS)and an important development direction of GNSS technique.Focusing on RT-PPP data processing model refinement and method optimization,the research is mainly carried out from three aspects:RT-PPP cycle slip detection and repair methods,real-time satellite clock offset estimation and prediction models,and real-time estimation and modeling methods of regional troposphere and ionospheric errors.And a real-time precision positioning service prototype system based on RT-PPP technology is developed.The main research contents and results are as follows:1)This paper analyzes the main difficulty in the detection and repair of cycle slip in RT-PPP,that is,the ionospheric delay having time-varying characteristics,which leads to the difficulty of repairing the cycle slip of narrow lane observation under the active condition of ionospheric delay.In this regard,an adaptive Kalman filter based on variance component estimation for the online modeling and prediction of epoch-differenced ionospheric delay(DID)is proposed.The DID prediction value is used to assist the cycle slip detection and repair.The experiment of real-time detection and repair of cycle slip in GNSS dual-frequency and tri-frequency signals with the method is performed,and the results show that: Compared with the traditional GF and MW cycle slip detection methods,the predicted DID value can effectively aid the detection and repair of small,large,continuous and insensitive cycle slips.In particular,the effect of the cycle slip repair on the narrow lane observation is more significant.2)Considering the influence of ground monitoring stations(distribution and quantity)on the accuracy and computational efficiency of GNSS satellite ultra-fast orbit determination and real-time clock offset estimation,a random optimization stations selection algorithm based on the spatial configuration of the monitoring station is proposed.The principle of the method is based on grid control theory and Monte Carlo random sampling.The select effect evaluation of the method is based on an index of station selection configuration.The algorithm can quickly and automatically select the station list with better geometric distribution and station quality.Experiments with 201 IGS stations shows that the proposed method canimprove the accuracy of GPS super-fast observation,forecast orbit and real-time clock offset by 17.15%,19.30% and 31.55% compared with the traditional grid method.At the same time,in the 100000 times random sampling experiments,when the number of stations is 10,50,and 90 respectively,the corresponding station selection time is lower than 2.22,6.65,and 14.15 min.3)For the problems of interruption and delay of real-time data streams in RT-PPP,this paper proposes an adaptive Kalman filter model based on variance component estimation for the ultra-short-term/short-term clock offset prediction.At the same time,considering the spatial correlation of satellite clock offset,a Kalman clock offset prediction model using inter-satellite correlation is studied.In order to verify the effectiveness of the proposed method,the prediction experiments with GBM final and CLK93 real-time products of 27 consecutive days are conducted.The result shows: Taking into account the correlation between satellites,better results can be obtained in the final clock offset prediction,such as: for the 6-hour BDS satellite clock offset predictions of GBM products,the method can increase the accuracy by about 50.00% compared with the traditional method(taking into account the period and trend items).Due to the weak correlation between satellites in real-time clock offset,the adaptive Kalman filter clock offset prediction model based on variance component estimation performs better in real-time clock offset prediction.For the1-minute BDS satellite clock offset predictions of CLK93 products,the prediction accuracy can be improved by 11.19% compared with the traditional Kalman filter predection method.4)Aiming at the characteristic that the zenith tropospheric delay(ZTD)parameter estimation in RT-PPP is susceptible to water vapor variation,an adaptive Kalman filtering method based on variance component estimation is proposed to improve the accuracy of real-time ZTD estimation.Based on the Beidou data processing center of China University of Mining and Technology(CUM),the ZTD solution experiment is carried out with the real-time estimated BDS/GPS clock offset products.The results show that:(1)The method can dynamically adjust the ZTD stochastic model to adaptively correct the error of the estimated parameters;(2)The method can effectively suppress of the ZTD estimation abnormality which is generally caused by unstable solution of other parameters,such as receiver clock error parameter,improve the accuracy of the ZTD in both real-time and final solution(especially in the real-time solution)whether the tropospheric variability is large orsmall.(3)The results show an improvement for real-time scheme with the new method over scheme without with average increasing rates of 20.7%(GC)and 20.2%(G),and an improvement for final scheme with the new method over scheme without with average increasing rates of 22.1%(GRCE),21.9%(GRC),18.4%(GR),15.9%(GC),15.2%(GE),and 12.1%(G).5)In order to realize ZTD real-time modeling,based on the real-time ZTD products above-mentioned,the machine learning methods(neural network method and support vector machine method)are used to perform regional real-time ZTD modeling.With the 5 consecutive days' BDS/GPS observation data of Hong Kong CORS network,the real-time ZTD model of the area is constructed.The accuracy of the constructed ZTD model is evaluated with reference to the four-parameter model.The results show that: The support vector machine modeling can achieve the ZTD modeling effect(mm level)equivalent to the four-parameter modeling;the average deviation and root mean square errors of neural network,support vector machine and four-parameter modeling are-2.25 mm and 9.17 mm,respectively;for the modeling of the sation at the average elevation surface of the measuring area,the support vector machine model has higher precision and stability than the four-parameter model.6)Aiming at the ionospheric delay error modeling based on RT-PPP technology,this paper constructs a global real-time ionospheric delay error model based on the spherical harmonic function model,and analyzes the characteristics of real-time ionospheric variations in small regions(longitude difference 5°,latitude difference2.5°)with a time resolution of 5 min,15 min,30 min,1 h and 2 h.The experimental results show that: the variation of the ionosphere in the latitude direction is greater than the variation in longitude;when the time resolution increases in multiples,the ionospheric variation shows the same trend.At the same time,in order to improve the accuracy of real-time ionospheric delay error extraction,this paper compares and analyzes the traditional carrier-to-code leveling(CCL)and the RT-PPP technologies.And the real-time modeling of regional ionospheric delay error is carried out by using neural network and support vector machine models with the 5 consecutive days' GPS observation data of Hong Kong CORS network.The results show that: RT-PPP technology has significant advantage over CCL technology in extracting real-time ionospheric delay errors.And artificial intelligence technology has high precision in real-time ionospheric modeling.7)In order to verify the correctness and effectiveness of RT-PPP dataprocessing model and method proposed in this paper,based on CUM platform,a real-time precision positioning service prototype system based on RT-PPP technology is designed and developed.With the real-time observation data streams of iGMAS and MGEX/IGS,and the real-time precision product data streams of CUM and CNES,the real-time location and atmospheric error enhancement service capabilities of the system are tested.The results show that: The system implements the main models and algorithms studied in this paper,and it is stable and reliable.