Optimisation en Temps Réel: Optimiser les Performances des Procédés Chimiques malgré l’Incertitude et les Erreurs de Modélisation

This document summarizes the author’s research activities since 1999. They mainly focus on the modelling, the control and the optimization of processes and contributed to the development of real-time optimization methods (RTO), that is, process optimization methods that are capable of driving a real process to optimal performances despite uncertainty, modeling errors and process disturbances. The main philosophy of these methods is to use available process measurement to correct the nominal model-based inputs. Several methodological contributions have been recently obtained in the fields of RTO via modifier adaptation (RTO-MA) but also in the establishment of sufficient conditions for feasibility and optimality that are applicable to any RTO algorithm. The need for the availabilty of an adequate model has been circumvented by the use of convex model approximations and the applicability of RTO-MA was extended to the optimization of discontinuous processes and to the use use transient information for the optimization of steady-state performances of continuous processes. These techniques have been successfully applied to industrial polymerization reactors, to experimental and industrial fuel cells stacks, to the iterative controller tuning problem and to the simulated production of energy using tethered kites. Research has been also performed in the field of modelling and control of blood sugar concentration for patients with Type I diabetes. Attention was paid to the construction of model-based prediction tools that are consistent with standard therapy.