Regional Ionospheric Modeling And Prediction Based On CORS

With the development of the Global Navigation Satellite System(GNSS),the application fields of high-precision navigation and positioning are becoming more and more widespread.As one of the most important errors in GNSS,ionospheric delay error has become the key to its application.At the same time,as the number of Continuously Operating Reference Stations(CORS)worldwide increases and observation quality improves,CORS observations are used to extract ionospheric information,and the distribution and changes of the inversion area or global ionosphere are monitored in real time and the ionosphere Short-term forecasting has become the focus of current research.This article uses regional CORS data to accurately extract Total Electron Contents(TEC)of the ionosphere,separate hardware delay errors,and establish a regional ionospheric delay correction model.During the model establishment process,a real-time ionospheric monitoring and modeling algorithm based on Kalman filtering and an optimization algorithm for ionospheric models based on Support Vector Machine(SVM)were proposed.In single and dual-frequency precision point positioning(PPP).An ionospheric TEC short-term prediction method based on Nonlinear Autoregressive Exogenous Model(NARX)with additional external input is also proposed,which has certain value in the practical application of GNSS.The main research contents and conclusions of this article are as follows:(1)The method of using high-frequency GNSS observations to extract high-precision TEC information is studied.Starting from the original GNSS observations,the causes and solutions of ionospheric delay errors and hardware delay errors are analyzed in detail.The experimental results prove that the TEC information extracted by the non-geometry model and the uncombined PPP model has the same form.The non-geometric model using phase-smoothing pseudorange can not only weaken the observation noise but also be simple to calculate,and can be applied to real-time ionospheric monitoring.The TEC information calculated using uncombined PPP has higher accuracy,but requires the support of precision products after the event,which can be applied after the event Solution of the ionosphere.(2)A regional ionosphere model was established based on CORS observations.First,a real-time ionospheric TEC modeling algorithm based on Kalman filtering is proposed.It receives CORS realtime data and forms phase-smoothed pseudorange non-geometric observations.It combines low-order spherical harmonic functions to solve the ionospheric model parameters and hardware delay at each epoch in Shanghai.The accuracy of this model is about 1.9TECU,which is 58.7% higher than the ionospheric products released by IGS,and the single-frequency PPP positioning accuracy is improved by 60.3%,which effectively speeds up the dual-frequency PPP initialization and reconvergence time.Secondly,using CORS ex-post data,using uncombined PPP algorithm to extract TEC,and combining the polynomial model to establish the ionosphere model in Western Europe,the accuracy is better than3.0TECU.(3)Combining SVM algorithm and traditional polynomial ionospheric model,a SVM-P based regional ionospheric fusion model is proposed.Based on the traditional polynomial model,the SVM regression algorithm is used to establish an error compensation model,which further improves the accuracy and effectiveness of the ionosphere model.The influence of different parameters on the function model is discussed,and the optimal SVM kernel function parameters are determined by the grid search method and the optimal correction effect is obtained.The ionospheric model in Jiangsu was established experimentally.The accuracy of the SVM-P model was 0.98 TECU,which was 17.3%higher than the polynomial model.In the single-frequency PPP positioning experiment,it was improved by more than 40% compared with the polynomial model.(4)A short-term prediction method of ionospheric TEC based on NARX is proposed.A NARX neural network model was established according to the characteristics of TEC changes.Using the historical TEC data and external input time parameters as network inputs,the TEC short-term forecast model was trained.Multiple sets of 2-day TEC prediction experiments were performed at high and low years of solar activity and at three different latitude observation stations,with Autoregressive Integrated Moving Average(ARIMA)models and the Center for Orbit Determination in Europe,CODE).The RMSs in 2011 with the NARX model are between 1.50 TECU and 3.72TECU;the RMSs in 2017 are between 1.05 TECU and 2.73 TECU.During the active period of the ionosphere,the accuracy of NARX prediction is 32.3% higher than that of ARIMA model and 43.5% higher than that of GIM.The accuracy of NARX is 20.7% higher than that of ARIMA model and 22.7% higher than that of GIM.