Model based estimation and optimal control of fed-batch fermentation processes for the production of antibiotics

This contribution addresses the application and comparison of model based estimation, optimization, and control methods for fed-batch bioprocesses. For the application of model based control, appropriate knowledge of the system’s state is required. The estimation quality of two constrained optimization based state estimation algorithms, namely the Bayesian maximum a posteriori based Constrained Extended Kalman-Filter (CEKF) and the Moving-Horizon-State-Estimation (MHE) is compared to the classical unconstrained Extended Kalman-Filter (EKF). The comparison is based on Monte Carlo simulations of a small mechanical and a high order grey-box model of a biological system. Moreover, EKF, CEKF, and MHE usually introduced separately, are described in a coherent setting starting from the same maximum a posteriori estimation problem. Finally the well-known nonlinear model predictive control (NMPC) is compared with a control strategy that uses online optimization of the complete future trajectory. Extensive Monte Carlo simulation studies and a real world application are considered.