Title:Adaptive Optimal Control of Unknown Nonlinear Systems via System IdentificationAuthor(s):Patrick SchranglAbstract:Today’s technical systems are often required to be operated in an optimal way, for example in the sense of minimizing energy, maximizing performance or a trade-off between both. For control systems this requirement is usually achieved by optimal control. Optimal control typically requires a precise model of the system to be controlled and as most systems are nonlinear and system complexity is continuously increasing, systematic modeling approaches have been developed. One way is to use blackboxsystem identification to model systems,which are available for measurements but only limited first principle knowledge about their dynamic behavior is available. This thesis is a contribution for extending a systematic method for nonlinear system identification towards online identification and merging it with a method of model predictive control (MPC) to get an adaptive controller that approximately achieves the goal of optimal control. The main part of the thesis is devoted to advancing the scope of the previous work by making it possible to adapt to a changing behavior of the system over time, e.g. due to aging. Therefore, an online algorithm for model parameter identification based on a multi-step prediction is developed. A multi-step prediction is chosen in view of the desired application of the model: nonlinear model predictive control. This algorithm is developed with standard exponential data forgetting and extended to a directional forgetting strategy to cope with the possible mismatch of data quality and number of model parameters in a controlled operation. Finally, the well-known C/GMRES algorithm for model predictive control is implemented to work with the generic polynomial model class used in the identification method and modified towards a combination with the online identification algorithm. It is shown on the practical example of selective catalytic reduction(SCR)control that the findings and approaches developed in this thesis can be used to improve the control performance of the system by using an MPC based on the identified model in a non-nominal case: when catalyst aging occurs.Page Reference:195 page(s)Publishing:1/2020