Iterative nonlinear model predictive control. Stability, robustness and applications

The article presents a new methodology applicable to batch processes called iterative nonlinear model predictive control (INMPC). It incorporates the ability of learning from past batches (known as iterative learning control or ILC) to an underlying nonlinear model predictive controller. The main advantage with respect to the existent iterative controllers is a faster convergence to the set points and guaranteed stability. The convergence is proven for a wide class of nonlinear processes when the desired trajectory is reachable. The controller ensures convergence for the nominal model whatever the parameters are. In the presence of uncertainties, a numerical analysis using randomized algorithms is utilized to ensure the desired probability of stability. The controller is tested on a highly nonlinear example with noise and model uncertainty, a pH plant. A numerical robustness analysis has been done in order to verify the sensitivity and performance of this new algorithm. The results obtained in the experiments are good, bettering other existent controllers.