Search

In the presence of model-plant mismatch,the set of parameter estimatesobtained using standard model identification procedures cannot accuratelypredict the gradients of the optimization problem. Hence, a “two-step”approach involving an identification followed by an optimization step,performed repeatedly, often fails to converge to the true processoptimum. In our recent work,1 we proposeda new identification procedure that progressively corrects the modelfor structural error such that the updated set of parameter estimatessimultaneously satisfies both identification and optimization objectiveswith a guaranteed convergence to the true optimum. In this paper,we expanded our previous methodology in two directions: first, a newmodel correction based on a higher order approximation of model isproposed, and second, the effect of model uncertainties is consideredexplicitly by formulating a robust optimization problem at each iteration.The resulting improvements are then illustrated using a simulatedstudy of fed-batch penicillin production process. [ABSTRACT FROM AUTHOR]