Research On Intelligent Particle Filter Based On Adaptive M-H Hybrid Resampling Strategy

Particle filter algorithm is an effective method to solve the nonlinear,non-Gaussian system state estimation,can track the target better,filtering effect is good.With the introduction of resampling technology,although the degradation of particle weight is solved,the problem of the lack of particle diversity is generated,resulting in the loss of particle diversity.On the basis of the existing M-H resampling,this paper further studies the improved M-H resampling sampling method,using the two strategies of cross and Gaussian variation to improve the diversity of particles.However,in the case of multi-peak distribution,the variance of Gaussian distribution is too large to jump out of the sampling range of particles and particles will be far away from the high probability area,which makes Gaussian variation also lose its original role.Reduce the filtering accuracy.Therefore,it is important to improve the resampling to improve the filtering accuracy.This paper studies the above problems and mainly completes the following work:(1)An improved adaptive M-H resampling strategy is proposed to solve the problem of the lack of particle diversity in the traditional resampling process of particle filter and the problem of jumping out of the sampled particle set when the variance of Gaussian variation is too large.For all low-weight particles,two resampling methods are selected with a certain probability.One is to randomly select high-weight particles to carry out Gaussian variation and establish adaptive variance function by using random walk strategy;the other is to carry out cross operation on the low-weight particles.These two methods improve particle quality,solve the problem that particles jump out of the sampled particle set and alleviate the lack of particle diversity.Finally,through simulation,the minimum root-mean-square error and average absolute error of the proposed method are compared under different noise conditions,which verifies that the proposed method has better filtering effect.(2)In order to solve the problem of uneven particle distribution and low state estimation accuracy when the posterior probability distribution is multi-peak,a particle filtering algorithm based on adaptive clustering M-H resampling is proposed.The particle distribution was optimized by the similarity between particles,and the minimum distance between two particles in the same cluster was used as the variance in Gaussian variation to avoid the particle deviating from the high probability region in the case of multi-peak distribution.Every particle in the particle set with low weight was sampled by clustering M-H resampling,so as to improve the overall particle quality.Simulation results show that the proposed method has better estimation effect.(3)The particle filter algorithm using the above two improved M-H resampler methods is combined with the actual capacity estimation of lithium ion battery data.The experimental results show that the proposed method also has good estimation performance in practical application.