Efficient Robust Multi-stage Nonlinear Model Predictive Control Strategies to Handle Plant-model Mismatch

The focus of this thesis is to reduce the conservatism introduced by robust NMPC approaches in the presence of plant model mismatch. First multi-stage NMPC is improved for parametric uncertainties. The standard approach overapproximates the uncertainty set by a box and generates the scenario tree from the vertices of the box. If the uncertainty set is not a box, this enlarges the uncertainty set and results in a significant reduction of the performance. This is mitigated by using sigma points to generate the scenario tree and by computing the future plant evolutions using the unscented transformation. The sigma points capture the true mean and covariance of the uncertainty set more precisely and results in a better performance. Second, adaptive and dual approaches are introduced to improve the performance of multi-stage NMPC in the presence of unknown but time-invariant parameters. The adaptive approach uses plant measurements to estimate the unknown parameters. The dual approach explicitly deals with the trade-off between the excitation of the controlled system by probing actions which lead to a more accurate estimation of the unknown parameters and optimizing the control inputs in order to improve the overall performance. Third, the existing robust techniques do not address structural plant-model mismatch. The concept of Model-error models (MEM), as used in linear control theory to achieve robustness against structural plant-model mismatch, is extended to the robust NMPC framework. The MEM dynamically bounds the uncertainty region around the nominal model of the plant and a scenario tree is constructed using the nominal and the MEM to address structural plant-model mismatch. The extensions are evaluated using examples from the chemical and biochemical field.