Research On The Theories And Algorithms Of Earch Rotation Paramters Prediction

Earth Orientation Parameters(EOP)play a key role in the fields of geophysics,astronomy and so on.EOP mainly include: Precession,Nutation,Polar Motion,Length of Day(LOD),the change of earth rotation(UT1-UTC).Polar Motion,LOD and UT1-UTC are also known as the Earth Rotation Parameters(ERP).EOP are the important parameters to achieve mutual conversion between the celestial reference frame and the Earth reference frame.High-precision and real-time EOP play a key role in practical applications,such as deep space exploration,high-precision space navigation and positioning,spacecraft tracking,Satellite Laser Ranging(SLR),Lunar Laser Ranging(LLR)and so on.The rapid development of modern space geodetic techniques(such as VLBI,SLR,GPS and DORIS)significantly improved the observation accuracy of EOP.The International Earth Rotation and Reference System Service(IERS)releases the EOP products by combining the solutions of VLBI,SLR,GPS and DORIS.This processing often needs a few days or even longer to get its final solution and is hard to be performed in real time.Forecast of EOP should be done to meet the needs of the space navigation and positioning.The thesis focuses on the theories and algorithms of ERP Prediction.The results and outcomes of the thesis can be used as good reference for improving the accuracy and reliability of the ERP prediction.1.When polar motion data contains gross error,the prediction accuracy and reliablity of most existing models is greatly influenced.This paper proposes a new model of combining Adaptive Robust Least Squares(ARLS)and autoregressive(AR)model for PM prediction to control the influence of outliers.The results prove that ARLS+AR model in PM prediction has significant advantages than LS+AR model.When there is no gross error,the prediction accuracy of ARLS+AR model will be almost the same as that of LS+AR model,and can be significantly improved when the gross error is included.2.Two modified algorithms are proposed to improve the PM prediction accuracy based on combination of Least Square and Autoregressive Model(LS+AR).One is to combine Kalman Filtering(KF)to improve AR model accuracy,namely LS+AR+KF algorithm.The comparison results show that the prediction accuracy of LS+AR+ KF is improved to a certain extent,and the accuracy of 360-day long-term prediction can be improved by 7.24%.The other is to use Least Mean Square Adaptive Filtering(LMSAF)to improve the accuracies of LS fitting terms and predicting extrapolations,namely LS+AR+AF algorithm.The results show that LS+AR+AF algorithms can significantly enhance the prediction accuracy of PM especially for long-term perdition,and LS+AR+AF is obviously superior to LS+AR and LS+AR+KF for PM prediction.The accuracy improvement of PM X component,PM Y component and PM can reach 26.21%,23.02% and 24.82% respectively,when using LS+AR+AF algorithm.3.The adaptive filtering with variable forgetting factor can improve the convergence speed,anti-jamming ability and tracking speed of general adaptive filtering algorithm,so LS+AR+IAF algorithm is proposed to to improve the PM prediction accuracy of LS+AR model.The numerical results show that the LS+AR+AF algorithms can significantly enhance the prediction accuracy of PM,especially for the long-term perdition,the accuracy improvement of PM X component,PM Y component and total PM can reach 30.66%,28.19% and 29.59% respectively when using LS+AR+AF algorithm for 360-day prediction compared to those of LS+AR model.All these prove that the adaptive filtering with variable forgetting factor can improve the PM prediction accuracy of classical LS+AR model greatly.4.GM(1,1)can be used to extract valuable information and forecast from a ?small sample‘ data.The advantages of Gray theory are that the prediction model is simple and the accuracy of linear term prediction is high.Considering the time-varying of periodic and linear term in the UT1-UTC,the combination of Gray Model and Autoregressive Integrated Moving Average(GM(1,1)+ARIMA)is proposed to predict UT1-UTC in this paper.The analysis and comparison related to the effect of different data length on GM(1,1)modeling and different basic sequence length on the UT1-UTC prediction are conducted.The results show that the accuracy of short-term UT1-UTC prediction with combination of GM(1,1)and ARIMA model is slightly lower than those of LS+AR,LS+MAR and WLS+MAR,but the accuracies of middle and long-term prediction are obviously higher than those of other three methods.All these prove that the proposed method can improve the UT1-UTC accuracy of middle and long-term prediction greatly.