An Improved Metropolis-Hastings Algorithm Based on Particle Filter

Abstract

Particle filter is a kind of Monte Carlo simulation method under the framework of Bayesian theory, and it can deal with state estimation problem of nonlinear models with non-Gaussian noise. But the use of resampling scheme to decreases the degeneracy phenomenon also introduces the sample impoverishment. M-H moves step is presented to increase the diversity of the samples and improve the performance of filtering. Aim at the acceptance ratio in standard M-H algorithm and RWM are too low, this paper proposes an improved M-H based particle filter. By improving the candidate proposal distribution, the algorithm reduces the estimate errors and increases the rate of accepted candidates. The simulation shows that the method achieves better performance of filtering compared to general particle filters and several other M-H based particle filters.