Investigation of performance of an adaptive kalman filter for state estimation of a gas phase reaction in a CSTR

State estimation is useful in estimating immeasurable or difficult-to-measure variables in a system from the measured inputs and outputs. For example, estimating chemical composition of material in a reactor from feed flowrate, temperature and pressure measurements. On-line state estimation plays an important role in effective process control and optimisation of nonlinear processes. The Kalman filter (KF) is an online platform for fusing propagated previous state estimates with incoming measurements to provide improved current state estimates. A number of KF variants exist, the most common being the Extended Kalman filter (EKF). Adaptive Kalman filters (AKF) have been developed but have not been investigated in detail in terms of their effectiveness in state estimation of chemical processes. The AKF is designed to adapt to changing conditions by allowing for the covariances of the process model and measurements to be determined online along with the state estimate.

In this work, a highly non-linear system, a reversible gas-phase reaction in a continuous stirred tank reactor (CSTR), is considered. AKF and EKF for estimating the concentration of the species based on online measurements of pressure and temperature were implemented. The performance and robustness of the AKF and EKF are compared under conditions such as errors in the initial conditions, covariance of the process model and measurements, and errors due to model mismatch. Possible improvements to the EKF and AKF will be considered.