Adaptive predictive control of dissolved oxygen concentration in a laboratory-scale bioreactor

We present an algorithm for the adaptive control of dissolved oxygen concentration in a bioreactor, based on the agitation rate. The dynamics are represented by an incremental first-order model with variable dead-time and parameters. These are estimated on-line by a recursive least-squares identification method with a forgetting factor and a constant sensitivity. The model is employed to predict the behaviour of the dissolved oxygen concentration over a finite horizon, using an original method which requires little computation. Then, a Generalized Predictive Control optimisation strategy computes the agitation rate from the predictions and the desired set point, while gradually updating the controller smoothness. This algorithm, which requires little preliminary knowledge, has been implemented on a laboratory-scale fed-batch bioreactor for which the use of conventional controllers showed limited performance, due to the unpredictable and evolutive nature of the dynamics. The new controller proved to be robust and effective over a wide range of operating conditions, while requiring no operator adjustments.