A weights particle filter for state estimation

The particle filter (PF) is an effective technique for state estimation in the nonlinear and non-Gaussian systems. However, one serious problem of PF is resampling. In this study, a weights PF (WPF) is proposed to mitigate the particle impoverishment problem common in PF. The WPF is inspired by the selection operator of genetic algorithm (GA). The weights of WPF are divided into two parts, and the large weights are used to replace the small ones. In this strategy, more particles are able to participate in the approximation of the posterior distribution after the procedure of resampling. Consequently, the results of state estimation can be more accurate. Meanwhile, the WPF has fairly low computational burden. Finally, the effectiveness of the proposed technique is verified by two simulation examples.