| This dissertation deals with the potential role of parameter adaptation as a means of performance improvement within the context of model predictive control. The first such role addressed here involves the adaptation of parameters that define the input sequence applied in the model predictions. In contrast to standard approaches involving iterative nonlinear programs and discrete sampling behaviour, the adaptive perspective on the parameter search developed here allows for near-continuous closing of the feedback loop. An additional benefit of this approach is that it clarifies opportunities for reducing the dimensionality of the necessary calculations by facilitating the use of efficient representations of the input trajectory. This approach is particularly well-suited for systems whose dynamics are relatively fast compared to the computation-time requirements of standard MPC controllers, since it provides a feasibility-preserving, stabilizing dynamic feedback even when the process and the parameter search evolve within comparable tirnescales.; The second role of adaptation addressed here involves its ability to improve upon the performance of robust control designs for process models exhibiting significant static uncertainties. In addition to reducing the conservatism of the robust controller over time, many additional benefits are made possible by imbedding an internal model of the adaptation meclianisin within the model used by the robust controller. The resulting adaptive controller, while admittedly numerically complex, has the potential to address a very general class of systems involving both state and input constraints. Furthermore, it is demonstrated that, at the cost of increased conservatism, the numerical burden of the adaptive controller can be reduced to within any arbitrary margin of the underlying robust control calculation. |
