Improved State Estimation for Stochastic Nonlinear Chemical Reactor using Particle Filter based on Unscented transformation

Abstract

The state estimation problem in a stochastic nonlinear and non-Gaussian system is solved by the particle filter. Particle filters require an importance proposal distribution from which samples are drawn which would be equivalent to drawing samples from the posterior distribution. The choice of a suitable importance distribution to represent true posterior density is a crucial step in the design of particle filter. The unscented Kalman filter (UKF) using unscented transformation technique provides better state estimates and also has the capability of generating heavier tailed distributions than the widely used extended Kalman filter (EKF). Hence, this paper utilizes the UKF based on statistical linearization method in the particle filter for generating an importance proposal to achieve improved state estimation than the extended Kalman particle filter (EKPF). The effectiveness of the particle filter based on unscented transformation over EKPF is illustrated by conducting simulation studies on the stochastic nonlinear continuous stirred tank reactor (CSTR) and the divergence problem in the EKPF is also discussed.