Abstract: For essentially nonlinear systems, the Unscented Kalman Filter （UKF） has some advantages such as high estimation precision, fast convergence and easy accomplishment. But the UKF is sensitive to model error of systems. To address this problem, a new UKF method based on model error prediction （MEP） is proposed, which is called Predictive Unscented Kalman Filter （PUKF）. The new filter utilizes the MEP process of Nonlinear Predictive Filter （NPF）, which can adjust the inaccurate model in real time and thus remedy the shortage of the UKF. Theory analysis and simulation results demonstrate that the new filtering method remarkably improves the efficiency of nonlinear filtering. Compared with the UKF, the new filter significantly improves the performance in precision, convergence speed and stability. So PUKF is applicable to uncertain and high nonlinear system.

Key words: state estimation, UKF, Sigma points, predictive filtering, model error, attitude determination