Extension Particle Filtering algorithm for state and parameter estimation in dynamic control process

Aiming to solve the problem of unknown parameters estimation in nonlinear and/or non-Gaussian dynamic system, Extension Particle Filtering (EPF) algorithm was proposed. EPF algorithm modeled unknown parameters by Gaussian random walk process, regarded unknown parameters in dynamic system as a part of state variations, and then estimated the state variations in extension nonlinear dynamic system by particle filtering algorithm. In order to improve the estimate precision of unknown parameters by utilizing observable information effectively, a new important density was purposed to instead of Bootstrap filter, further more, it avoided transformation about covariance. In order to solve the problem that covariance augmented infinitely with time in Gaussian random walk model, and kernel smooth factor restrained covariance excessively so as to the values of estimate parameters could not access the values of true parameters sufficiently, the gradually reduce(GR) factor was purposed to instead of Kernel factor. At the end of this paper, the effectiveness and availability of purposed algorithm was validated by an exemplum.