摘要：AbstractDiscrepancies between model and process responses typically result from modelling simplifications and assumptions. In the presence of structural model-plant mismatch, a run-to-run optimization procedure that follows a standard repetitive “two-step” approach, involving identification followed by optimization, may not converge to the process optimum. Parameter adaptation is a methodology that can be used to correct for model error at each batch run to drive the process to an optimum. Model parameters are updated during each iteration (batch) to account for changes in operating points as well as to correct for errors in the predicted gradients of the cost function and constraints. However, parameter values which results in a good model fit in terms of fitting the predicted outputs to data do not necessarily give an accurate prediction of the gradients of the cost function, which are necessary for reaching the optimum. We propose a parameter estimation objective which not only minimizes the differences between measurements and predictions but also maximizes the sensitivities of the cost function and constraint gradients with respect to the parameters. Towards these goals, we make use of set-based constraints which enforce uncertainty bounds over the model predictions. The resulting improvements in convergence to the optimum as compared to earlier studies are illustrated using a simulated case study of a penicillin fed-batch process.
