Abstract:

The filtering performance of unscented Kalman filter(UKF) would be degraded or even divergent due to unknown or inaccurate system noise statistics. An adaptive UKF based on maximum likelihood principle and receding horizon estimation is presented to address this problem. An estimation model of system noise statistics is constructed according to the maximum likelihood principle. Then, the receding horizon strategy is employed to optimize the above model. Eventually, the sequential quadratic programming is applied to calculate the estimation of noise statistics and the adaptive UKF with a noise statistics estimator can be obtained. It can realize  online estimation of system noise statistics and overcome the defect of standard UKF. The performance of the proposed adaptive UKF is verified through the application examples in inertial navigation system/global positioning system integrated navigation system.