Research On Short Term Prediction Model Of The Atomic Clock Bias On Board GNSS Satellites

Global Satellite Navigation System(GNSS)can provide all-weather services for people's navigation,positioning and timing.For navigation and positioning,time synchronization is a key technology and plays a crucial role.Its accuracy directly affects the accuracy of navigation and positioning.As the time base for navigation signal formation and system ranging,the atomic clock in satellite is a core part of the payload of the GNSS navigation system,which performance relate to the accuracy of time-frequency transmission.Due to the sensitivity of the satellite atomic clock,and the uncertainty of space environment,it is affected by various factors,making it difficult to understand its complex variations.There is an urgent need for a model that can predict its temporal variation.For this reasons,this paper studies the prediction model for navigation satellite on-board atomic clocks bias,aiming at how to improve the accuracy and stability of the satellite clock bias prediction.The main content of the article is as follows:(1)In this paper,various common prediction models,such as quadratic polynomial model(QP),grey model(GM)and time series analysis method,are studied in detail.The satellite clock bias modeling experiment of each common model was carried out by using GPS satellite clock bias data and comparative analysis was carried out.(2)Based on the disadvantages of basic QP model,this paper consider the periodic term and random term of the satellite clock bias,the refined QP model is established by using SVM.The fitting error of the QP model is used to quadratic modeling by the SVM to obtain the error compensation term to improve the accuracy of QP model prediction value.GPS satellite clock bias data is used to verify that the refined model has a significant improvement in prediction accuracy for the basic QP model.However,its prediction accuracy depends on the fitting effect of the basic QP model meaning that refined model has certain limitations.(3)A combined prediction model,which is based on differential ensemble empirical model decomposition(EEMD),high-order polynomial model and SVM,is proposed.First,the satellite clock bias is differentiated to eliminate the trend term and reduce the effective number of digits of the satellite clock bias.Then the EEMD is used to decompose it into IMF subsequences with different frequencies and one trend component term with low frequency.The low frequency components is predicted by using high-order polynomial.Moreover the IMF sequences is predicted by using SVM which its parameter search method has been optimized by the particle swarm optimization algorithm.The improved SVM is used to predict the IMF sequences,and all the prediction values are reconstructed to obtain the final value.GPS satellite clock bias data is used to verify that the combined model.The results show that the combined model is feasible and effective.And its prediction accuracy is better than the refined QP model.(4)This paper use the combined model and common models to predict the short-term satellite clock bias data and analyze the results.The result show that the prediction accuracy of the combined model is better than the commonly used basic models for the satellite clock bias data of different global satellite navigation systems and different atomic clock types.It is feasible and universal in practical applications.