Comparison of unconstrained nonlinear state estimation techniques on a MMA polymer reactor

Estimation theory finds a wide variety of applications in process engineering. Most chemical processes exhibit highly nonlinear dynamics, and the extended Kalman filter (EKF) has been widely used to solve the estimation problem in chemical processes. However, it is claimed that the EKF performs poorly when the noise sequences are non-Gaussian (NG). Owing to high nonlinearity of chemical process dynamics, it is likely that innovation sequences are non-Gaussian. Nonlinear estimators such as the unscented Kalman filter (UKF) and particle filters (PF) have been developed to address the theoretical limitations of the EKF. In this paper, we study the effect of filter assumptions on their practical performance. Different estimation algorithms are applied onto a methyl methacrylate (MMA) continuous stirred tank reactor (CSTR) under different scenarios of state and measurement noise and plant-model mismatch.