Adaptive Control of Saccharomyces cerevisiae Yeasts Fed-Batch Cultivations

Fed-batch baker’s yeast cultivation is a complex biotechnological process from the viewpoint of measurement and control. It is a non-linear system with not well-known dynamics. The process is non-stationary due to metabolic changes, modifications in cell physiology and multiple increases in biomass concentration over the cultivation time. Furthermore, there is lack of cheap and reliable online sensors for measurement of important biochemical quantities, e.g. substrate and biomass concentration.1 Because of the complexity and time variant nature of fed-batch cultivations, the use of PID controllers with constant parameters is restricted and modification of these parameters is often necessary during the cultivation.2 This can be accomplished by an adaptive controller, which automatically adjusts its parameters to the actual state of the controlled process, using either a mathematical model of the process – e.g. internal model control principle3 –, or an identification-free algorithm extracting information from the process data in real time as in this work. Reported model-based adaptive controllers for the control of bioprocesses include approaches based on Haldane kinetics,4 adaptive-predictive controllers using a recursive least-square identification method for prediction consisting of an incremental linear model5,6 or a model reference adaptive estimation and control using Lyapunov’s method applied on a pilot-plant fermenter.7 Further there is an application in the lactic fermentation process, in which parameters are estimated on-line and an adaptive-multivariable predictive controller is used.8 Another application to an anaerobic digestion pilot plant introduces non-linearity in the control scheme in order to compensate non-linearity of the system.9 A pole-placement control for time-varying multivariable first order plants10 and a pole-assignment method in conjunction with ARMAX structured process model11 can also be used. Other method is a discrete-time adaptive LQ control law,12 an adaptive controller employing the linearized Kalman filter for the state estimation13,14 or the concept of adaptive regulation based on respiratory quotient (RQ).15,16 Alternatively, neural-network-based adaptive controllers exploiting learning capabilities of the artificial neural nets17,18,19,20 and fuzzy relational predictive controllers with a fuzzy relation model21 are used for the control of bioprocesses. Adaptive linear control strategies can be used for the optimal control of biotechnological processes with a yield–productivity conflict.22,23 Recently, Adaptive Control of Saccharomyces cerevisiae Yeasts Fed-Batch Cultivations