The Critic Technologies Research Of Particle Filter

The state estimation problem in the nonlinear and non-Gaussian system has attracted theattention of many domestic and international scholars, which has become the focus ofresearch question. A particle filter technology could solve the above problem effectivelywhich has been developed in recent years. Particle filter is also known as sequential MonteCarlo method. The basic idea of it is to obtain the value of the state estimation, which use theweight value of a series of random sample and the posterior probability density of therequired. It has been widely used in many area for its unique advantages in dealing withnonlinear, non-Gaussian system state filtering problems, such as communications, targettracking, fault diagnosis and satellite navigation, etc.At present, the particle filter algorithm has made gratifying progress both at home andabroad. However, it has many problems which is urgent to be solved because it is still in theearly stages of development, such as the problem in selection of importance function, theproblem in sample dilution and convergence of the algorithm, etc. Therefore, It would besignificant to improve the particle filter algorithm. This paper proposed an improved weightsof optimized combination of particle filter algorithm and the improved adaptive geneticalgorithm of particle filter for particle filtering algorithm resampling caused by samplesimpoverishment and the defects of large amount of calculation. The former is able toguarantee the accuracy and reduce the amount of calculation in the process of resamplingeffectively, which is through setting threshold Thershold of the weights of the particles andeliminating the particles whose weight is less than the Thershold particle, then the particles ofweights is less than particle swarm average weights will be optimized combination. The latterimprove the filtering accuracy which rely on a good global search ability of genetic algorithmto improve the particle degradation and sample impoverishment problem.It puts forward a differential evolutionary particle filter algorithm based on particleswarm optimization in this paper for view of the importance density function selectionproblem. The convergence factor and mutation particle individual extreme value is introducedto update the particle’s speed and modified particle’s position, which can drive the particle setto move to high likelihood area and improve the global search ability of the algorithm, andthen it can also prevent algorithm falls into local optimum and alleviate the degradation of aweight problem. In resampling phase, it suppressed the particle diversity loss throughresampling combination of differential evolution optimization which is according to the sizeof the threshold in traditional heavy sampling.The simulation results indicated that thealgorithm has good estimation precision.