Nonlinear state estimating using Adaptive Particle Filter

It is well known the standard particle filter has a good effect when the observation accuracy is low. However, if the observation accuracy is high, the likelihood distribution may become aiguilles-like and locate at the tail of the prior distribution curve; this will make the filter diverge. To solve the problem, a kind of adaptive particle filter is proposed in this paper. The adaptive particle filter has a higher filtering stability by changing the likelihood distribution according to the Statistic characteristic of the observation noise and enlarging the overlap of the prior distribution and the likelihood distribution. A simulation is developed in nonlinear and non-Gaussian integrated navigation system in this paper. The simulation has been done in the condition that the observation accuracy went from low to high. The simulation result indicates that the adaptive particle filter has a high filtering precision and stability even if the observation accuracy is high.