Kalman Particle Filter Algorithm Research

Kalman Particle Filter algorithm is a kind of filtering method which completes a bayesian recursive process through the Monte Carlo simulation method, and now it has been widely used. As a kind of nonlinear filtering algorithm based on bayesian estimation, it has a unique advantage in the treatment of the parameter estimation and the state filtering problems in non-Gaussian nonlinear time-varying systems.Based on the research status and development trend of the Kalman Particle Filter algorithm, the existing problems of this algorithm are studied and a new developed method is presented in this paper. The concrete research content is divided into three parts:Firstly, to resolve the particle degeneration in Kalman Particle Filter algorithm, the Iterative Unscented Kalman Filter(IUKF) combined with RTS smoothing algorithm is used to generate a new importance density function, in which the posterior probability estimation of the system status is conducted by IUKF and the filtering result is revised by RTS smoothing algorithm.Considering the latest observations of the system, this method can inhibit the particle degradation effectively and the precision of the state estimation is greatly improved.Secondly, to deal with the problem of particle impoverishment in Kalman Particle Filter algorithm, a variable step-size adaptive Artificial Fish Swarm Algorithm(AFSA) is introduced into the resampling process. This improved algorithm prompt the particle set moving to the posteriori probability distribution of the real value, so the problem of accuracy reducing in particle filtering has been improved. And the algorithm is also used for the robot localization in flicker noise conditions and improves the precision of localization significantly.Thirdly, the Central Difference Kalman Filter algorithm is combined with Particle Filter algorithm and introduced the degradation factor to alleviate theparticle degeneration problem, by which the operation speed of particle filter algorithm is increased and its operation time is reduced while remains the algorithm accuracy.Finally, the simulation research is conducted on the improved algorithms proposed above and the results show the correctness and validity of them.