摘要：In the presence of model-plant mismatch, a standard "two-step" approach, involving repeated identification and optimization steps, cannot guarantee convergence to the process optimum. Model parameter adaptation can be used for handling model error by correcting for mismatch between predicted and measured cost and constraint gradients while simultaneously satisfying both identification and optimization objectives. However, updating all model parameters at once is often impractical due to estimability and increased sensitivity to noise. This work presents a procedure for selecting, after each run, a particular subset of parameters based on parametric sensitivity of the model output and of cost and constraint gradients. The resulting improvements with respect to previous run-to-run studies are illustrated using a simulated case study of a penicillin fed-batch process.
