| dc.description.abstract | Model Predictive Control (MPC) is a well-established technology for advanced control of many industrial processes. One of its main benefits is the ability to handle process constraints. However, accuracy of the process model is critical to the output of standard/nominal MPC, and typically many sources of uncertainty exist. Even though it has some degree of inherent robustness—under quite strict assumptions—performance is usually insufficient for nonlinear systems.
The main goal of this thesis is to study existing robust Nonlinear Model Predictive Control (NMPC) approaches, specifically multi-stage NMPC and min-max NMPC, and design a robust controller for a real industrial process.
To this end, a literature review was conducted, covering the most prominent robust MPC techniques available today as well as the challenges involved. Then ways of addressing the issues of computational complexity and conservativeness were explored briefly. Finally, based on the acquired knowledge, robust NMPC was implemented on an oil production optimization case study.
Simulation results showed that, in contrast to standard NMPC, both multi-stage NMPC and its min-max variation can ensure constraint satisfaction for all possible values of the uncertainties—if properly designed. In terms of conservativeness, both performed equally well, but it was clear that multi-stage NMPC is the more computationally attractive choice.
It was also seen that there exist methods for successfully dealing with the inevitable loss of performance resulting from robust NMPC; results suggest that if an efficient implementation is carried out, for certain cases, even real-time implementation may be possible.
Overall, the thesis findings indicate that robust NMPC is an essential tool for advanced control of industrial processes and that the multi-stage NMPC approach in particular is rather promising. | |
