The shortcomings of an adaptive Sage filter are analyzed in this paper.An improved adaptive Sage filter is developed by using a weighted average quadratic form of the historical residuals of observations and predicted states to evaluate the covariance matrices of observations and dynamic model errors at the present epoch.The weight function is constructed based on the variances of observational residuals or predicted state residuals and the space distance between the previous and the present epoch.In order to balance the contributions of the measurements and the dynamic model information,an adaptive factor is applied by using a two-segment function and predicted state discrepancy statistics.Two applications,orbit determination of a maneuvered GEO satellite and GPS kinematic positioning,are conducted to verify the performance of the proposed method.
A two-way adaptive Kalman filter is proposed by combining a two-way filter with an adaptive filter for orbit determination of a maneuvered GEO satellite.A method of using Newton's high-resolution differential formula and polynomial fitting for modeling the thrust force of a maneuvered GEO satellite is developed.The adaptive factor,which balances the contributions of the measurements and the dynamic model information,is determined by using a two-segment function and predicted residual statistics.Simulations with a maneuvered GEO satellite tracked by the Chinese ground tracking network were conducted to verify the performance of the proposed orbit determination technique and the method of thrust force modeling.The results show that refining the thrust force model is beneficial for the orbit determination of a maneuvered GEO satellite;the two-way adaptive Kalman filter can efficiently control the influence of the dynamic model errors on the orbit state estimate.
