An on-line optimising control strategy involving a two level extended Kalman filter (EKF) for dynamic model identification and a functional conjugate gradient method for determining optimal operating condition is proposed and applied to a biochemical reactor. The optimiser incorporates the identified model and determines the optimal operating condition while maximising the process performance. This strategy is computationally advantageous as it involves separate estimation of states and process parameters in reduced dimensions. In addition to assisting on-line dynamic optimisation, the estimated time varying uncertain process parameter information can also be useful for continuous monitoring of the process. This strategy ensures that the biochemical reactor is operated at the optimal operation while taking care of the disturbances that are encountered during operation. The simulation results demonstrate the usefulness of the two level EKF assisted dynamic optimizer for on-line optimising control of uncertain nonlinear biochemical systems.
Saccharamyces cerevisia known as baker’s yeast is a product used in various food industries. Worldwide economic competition makes it a necessity that industrial processes be operated in optimum conditions, thus maximisation of biomass in production of saccharamyces cerevisia in fedbatch reactors has gained importance. The facts that the dynamic fermentation model must be considered as a constraint in the optimisation problem, and dynamics involved are complicated, make optimisation of fed-batch processes more difficult. In this work, the amount of biomass in the production of baker’s yeast in fed-batch fermenters was intended to be maximised while minimising unwanted alcohol formation, by regulating substrate and air feed rates. This multiobjective problem has been tackled earlier only from the point of view of finding optimum substrate rate, but no account of air feed rate profiles has been provided. Control vector parameterisation approach was applied the original dynamic optimisation problem which was converted into a NLP problem. Then SQP was used for solving the dynamic optimisation problem. The results demonstrate that optimum substrate and air feeding profiles can be obtained by the proposed optimisation algorithm to achieve the two conflicting goals of maximising biomass and minimising alcohol formation.