Effect of Sampling Rate on the Divergence of the Extended Kalman Filter for a Continuous Polymerization Reactor in Comparison with Particle Filtering

Abstract

The Extended Kalman Filter (EKF) is the most widely used state estimation technique for non-linear systems in the field of process engineering. In this contribution, we investigate the performance of the EKF for continuous polymerization of acrylic acid in a tubular reactor with multiple side feeds of monomer. The EKF usually yields satisfactory estimations if the nonlinearities of the underlying system are not too severe. We observed that the EKF for this process diverges regardless of its tuning unless it is iterated at very fast sampling rates. In order to verify this, we have tested the EKF for different tunings and different sampling rates in 100 independent Monte Carlo simulations for each setup. In contrast Particle Filters use the nonlinear model of the process directly and do not suffer from the problems caused by linearization. On the other hand the computation times are significantly higher.