基于Metropolis-Hastings变异的粒子群优化粒子滤波器

Particle Swarm Optimized Particle Filter Based on Metropolis-Hastings Mutation

为了解决粒子滤波在粒子数量较少时估计精度不高的问题,提出了一种基于Metropolis-Hastings(MH)变异的粒子群优化粒子滤波算法。该算法将Metropolis-Hastings(MH)移动作为粒子群优化的变异算子,通过将MH变异规则与粒子群的速度-位置搜索过程相结合,使得重采样后的粒子群更接近真实的后验概率密度分布,有效解决了一般的变异粒子群算法容易发散的问题,加快了粒子滤波在序贯估计过程中的收敛速度,提高了其估计精度。仿真试验证明,基于MH变异的粒子群优化粒子滤波算法可以有效地克服粒子贫化现象,改善对非线性系统的跟踪估计效果。

A particle swarm optimized resampling method for particle filter based on Metropolis-Hastings（MH） mutation was proposed for improving estimation performance and particle impoverishment problem in the particle filter. The new algorithm chooses the MH moving as a mutation operator of particle swarm optimized, combines the mutation operator with velocity-position searching progress, and generates the particles so that their stationary distribution is a target posterior density. The new algorithm solves the problem of particle divergence effectively, speeds up the convergence rate,and improves the estimation precision. The simulation results show that the PSO resampling based on MH muta- tion can remove the degeneracy phenomenon and improve the tracking estimating effects in non-line system.