Bioprocess State Estimation: Some Classical and Less Classical Approaches

Abstract

This paper reviews some classical state estimation techniques for bioprocess applications, i.e., the extended Kalman filter and the asymptotic observer, as well as a more recent technique based on particle filtering. In this application context, all these techniques are based on a continuous-time nonlinear prediction model and discrete-time (low sampling rate) measurements. A hybrid asymptoticparticle filter is then developed, which blends the advantages of both techniques, i.e., robustness to model uncertainties (through a linear state transformation eliminating the reaction kinetics) and a rigorous consideration of the process and measurement noises. A simulation case-study is used throughout this paper to illustrate the performance of these state estimation techniques.

Keywords: Nonlinear estimation, particle filter, asymptotic observer, Kalman filter, biotechnology.