Research On Farst Particle Filter

Particle filter algorithm is a very effective way to solve the nonlinear system model.The essence of the particle filter algorithm is using a set of sample values in the space, withsample mean instead of integral operation, so as to obtain of the minimum variancedistribution. But particle filtering is to sacrifice the effectiveness and diversity of sampleand huge calculation for cost. Because this algorithm requires a large number of samplescan only approximate the posterior probability density of system very goodly, and causingthe loss of the sample in the resempling stage which will result in the sample dilution. According to the particle degradation and poor real-time phenomenon in the standardparticle filter, this paper studied the following particle filter algorithm, to improve thereal-time of the filtering algorithm.First, the U particle filter algorithm and systematic resempling algorithm are analyzedand introduced in the paper, aiming at the degeneracy phenomenon exist, the systematicresempling algorithm is used into the UPF algorithm, in order to improve the real time. Thealgorithm is improved from samples and the use of systematic resempling, for the purposeof reducing the cost and real increasing speed of computation speed. Simulation resultsshow that the improved algorithm can reduce computational complexity, and improve therunning efficiency.Secondly, combined with the likelihood distribution adaptive particle filter and sampleadaptive particle filtering, combining the likelihood distribution adaptation and sampleadaptive, first in each step of variance estimation of the lower limit of the number ofsamples, but also consider the state variance is too large or too small, in the resamplingphase embedded likelihood sampling, according to the precision factor reflecting the realthe statistical properties of observed noise from the size to adjust the likelihood distributionstate, so that the tail is more flat, can improve filtering accuracy and speed. Simulationresults show that improved algorithm can effectively solve the problem of particledegeneracy, increase particle diversity, and real-time performance of the algorithm.Finally, the study and analysis of the structure and algorithm of parallel distributedparticle filter, and then the clustering idea into distributed particle filter algorithm to thestructure of the clustering algorithm, and make improvement to the traditional. Finally,combined with the improved clustering and adaptive thoughts, forming a distributedadaptive particle filter to improve the real-time performance of the original algorithm. Simulation results show that the improved algorithm can reduce computational complexity,and improve the running efficiency.