Fission Bootstrap Particle Filtering
Zhang Jian-yun · Dianzi xuebao · 2008
Among the various particle filtering(PF) evolutions,the bootstrap PF(BPF) is the most classical and popular algorithm.However,it is subject to severe sample impoverishment after resampling.To overcome the above problem,the fission BPF(FBPF) algorithm is proposed,in which the preprocess including weights sorting,particle reproducing by fission,and weights normalizing is inserted before the original resampling step as soon as the importance weights degenerate severely.The results of Monte Carlo simulations about a typical severely nonlinear filtering problem with bimodal posterior density have demonstrated that compared to the BPF,the more robust FBPF can escape successfully from the sample impoverishment problem as well as preserve the estimation accuracy and computation burden.