Convergence Analysis of Iterative Identification and Optimization Schemes

The use of measurements to compensate for model uncertainty has received increasing attention in the context of process optimization. The idea consists of iteratively using the measurements for identifying model parameters and the updated model for optimization. This paper investigates the convergence of various iterative identification and optimization schemes in the presence of model mismatch. The optimization can be model-based, data-based or of mixed nature. Based on the advantages and drawbacks of the various approaches, a novel scheme is proposed, by which the optimization starts model-based so as to ensure fast improvement and finishes as a data-based approach so as to converge towards the true optimum. The performance improvement obtained with the proposed methodology is illustrated via the simulation of a semi-batch reaction system.