Research On Improved Algorithm And Application Of Particle Filter

In the face of increasingly complicated modern system,the traditional nonlinear state estimation algorithm is difficulty to meet some requirements in practical application,so more and more scholars are gradually focus on how to improve the filtering precision of nonlinear state system.Particle filter is a new algorithm emerging in recent years and can be widely used in nonlinear systems,it can fulfill the filter task better not affect by the distribution of the noises of the system model and the constrained conditions,so it has unique advantages in some areas,such as fault diagnosis and analysis,visual tracking,communications,statistical signal processing,automatic control and so on.At present,the particle filter algorithm remains to be improved,this paper based on the theory of particle filter algorithm,putting forward some improvement aim at the article degradation and impoverishment phenomenon which often appears in conventional particle filter,hence,the filtering precision and convergence speed of particle filter can be improved,next applying the improved particle filter algorithm into SLAM to achieve the purpose of expanding the scope of its application.The main contributions of this dissertation are as following.Firstly,we introduces the basic knowledge of recursive Bayesian estimation theory and Monte Carlo simulation theory briefly,and then,the whole process of particle filter algorithm are given on this basis.Finally the particle filter algorithm is compared with the classical Kalman filter algorithm and Unscented Kalman Filter algorithm,the simulation results of MATLAB shows that,the performance of the standard particle filter algorithm is superior to other traditional filtering algorithms,it does have the potential.Secondly,according to the problem of degradation and dilution which often occur in standard particle filter algorithm,a particle filter algorithm based on the gravitational field(GFA-PF)is proposed.The gravitational field optimization ideas is introduced into GFA-PF which aims at improving the resampling process of particle filter,the mobile factors of GFA-PF can make the sample particle concentrate near the real value rapidly,and the rotation factors can make the particles which around true state away from the true state randomly,that can avoid particles excessive concentrating,at the same time,increasing the diversity and robustness.Finally,through the non-Gaussian model and UNGM typical nonlinear model simulation,proved that the estimation precision and convergence time of GFA-PF were superior to APO-PF and EM-PF.Thirdly,the improved particle filter algorithm is used to the mobile robot SLAM study under the framework of particle filter,we put forward a FastSLAM2.0 algorithm based onoptimization of the gravitational field.Replacing the general particle filter algorithm that estimates the path part of FastSLAM2.0 algorithm with GFA-PF,thus to optimize the sample particle distribution of mobile robot pose,enables the particles to toward the robot real pose state more rapidly,at the same time improve particle degradation and dilution in FastSLAM2.0 Algorithm.By SLAM simulation experiments which under the same noise condition show that,the accuracy of robot pose and road signs characteristics estimation in GFA-FastSLAM2.0 algorithm are better than EKF-SLAM and FastSLAM2.0 algorithm.Finally,the conclusions and the future research work are discussed.