Research On Particle Filter Algorithm And Its Application

With the increasing complexity of modern systems and the improved requirements offiltering accuracy to nonlinear system, the conventional nonlinear state estimation algorithmshave been difficult to meet the demands of some applications. Particle filter (PF) is a newnonlinear filtering algorithm, it uses finite weighted particles to approximate posterioriprobability distribution of the system state. When the number of particles is large enough, PFcan infinitely approach the optimal state estimation in theory. PF can not be restricted bysystem model and conditions of noise distribution, so it has a better performance to satisfy therequirements in real filtering task, which has extensive application prospects in faultdiagnosis、target tracking、positioning and navigation、communications biostatistics、statisticalsignal processing and other fields.In this thesis, the theory of PF and its application methods are the main research contents.In order to advance the estimation accuracy robustness, some improved PF algorithms areproposed by aiming at the main problems of conventional PF. In addition, PF theory is used tosolve the mobile robot simultaneous localization and mapping problem for expanding itsrange of applications. Main results and contributions of this dissertation are as follows:1．To solve the problem of particle degradation and sample dilution in PF algorithm, anew kind of particle filter algorithm named EM-PF is proposed. Electromagnetism-likemechanism (EM) is introduced in resampling procedure and each sampling particle is lookedas a charged particle. Electromagnetism-like attractions drive the particles approach to highlikelihood region, so particle degradation is reduced. Similarly, electromagnetic-likerepulsions drive the particles separate from each other and maintain a certain distance inmoving process to assure diversity. Through simulation experiments in different models, theimproved algorithm is verified better estimation performance and adaptability than theconventional PF.2．PF theory is used to deal with mobile robot simultaneous localization and mappingproblem, a new FastSLAM2.0algorithm based on electromagnetism-like mechanism(EM-FastSLAM2.0) is proposed. Firstly, in order to reduce model linearization error andmake sampling particles closer to the true state, UKF is used to replace EKF for posterioriproposal distribution of robot pose. Then, the EM is introduced in resampling procedure toimprove particles distribution, and effectively avoids the particle degradation and particlediversity scarcity. Through comparing simulation experiments show the superiority of thisproposed algorithm. 3．The multi-robot robot simultaneous localization and mapping is in-depth studied toexpand the scope of application of PF, and a multi-robot FastSLAM algorithm based onlandmark consistency correction (MBLCC-FastSLAM) is proposed. It is based on singlerobot EM-FastSLAM, and when multiple robots observe the same landmarks, every robot isregarded as one node and Kalman-Consensus Filter is used to update landmark information,which integrate the priori estimated to landmark from different directions by various robotsand further improves the accuracy of localization and mapping.