Abstract:

In view of the uncertainty of the system dynamic model may reduce the filtering effect and the system state covariance matrix is negative definiteness, a new unscented particle filter(UPF) based on Gaussian process regression and square-root decomposition(GPSR) is proposed. The importance density function of UPF is gotten by Gaussian process regression. When the system model and observation model are inaccurate, Gaussian process regression is used to learn the training data, the regression models and noise covariance of the dynamic system are gotten; square-root decomposition is used to restrain the negative definiteness of the system state covariance matrix. The proposed algorithm is applied to the integrated navigation system of strapdown inertial navigation system / global positioning system (SINS/GPS). The simulation results show that the proposed algorithm is better than UPF, and also effectively improves the positioning precision of the navigation system.