Multivariable Long-Range Predictive Control Algorithm Applied to a Continuous Flow Fermentation Process

Abstract

The dynamics of biotechnological processes are highly complex and usually unknown. These processes are described by a set of non linear and non stationary equations derived from mass balance considerations. The complexity of these systems limits the application of classical control techniques and indicates that adaptive techniques may be of interest. This paper presents a simulation study of the multivariable generalized predictive control applied to a continuous flow fermentation process. The control design is based on predicting the process outputs over several steps and assumptions on future control actions. The control design is based on a linear discrete-time model with unknown and possibly varying parameters. The simulation is performed using a nonlinear and time-varying model of the process . These parameters are estimated by a constant trace adaptation algorithm with necessary features (data filtering and normalisation, UD factorization, dead zone .. ) that enhance the robustness of the resulting self-tuning algorithm. In the parametric model, the dilution rate and the influent substrate concentration are the control variables (Inputs) and the biomass and substrate concentrations stand for the controlled variables (Outputs). Numerical results illustrate the good performance and the excellent properties of the presented control scheme, particularly its ability to cope with varying dynamics and high level disturbances.