This thesis addresses the development of efficient nonlinear model predictive algorithms for process control. A Newton-type approach is proposed, generalizing the domain of applicability of previous formulations to new classes of models.; The formulation presented is shown to preserve the global stability properties of previous approaches for nonlinear processes that are asymptotically stable in the large.; Special attention is dedicated to the quantification of the effects of active constraints in the stability of model predictive controllers. Assuming a fixed active set, we show that the optimal solution of the linearized problem can be expressed in a general state-feedback closed form.; To extend the constraint handling capabilities of the method, we introduce the possibility of using soft constraints. Here we compare the use of the {dollar}lsb1{dollar} (exact), {dollar}lsb2{dollar} (quadratic), and {dollar}lsbinfty{dollar}-norm penalty formulations.; A trust region extension, to improve the efficiency of the optimization algorithm in the presence of ill-conditioning in the linearized predictive equations, is also considered in this work.; Finally, the control algorithm is extended to deal with time delays in the dynamic model of the process to be controlled. A separate analysis is presented for constant and variable delay terms. (Abstract shortened by UMI.)... |