Semi-explicit MPC for classes of linear and nonlinear systems

In this thesis, a novel fast and efficient model predictive control (MPC) scheme termed semi-explicit MPC is presented. Existing MPC schemes typically belong to one of two main types: Either the state-dependent numerical optimization problem inherent to MPC is solved repeatedly online during runtime or a solution of the optimization is precomputed offline in advance for all states of interest so that online the solution only has to be evaluated for the current state. The scheme proposed here joins both approaches in an innovative fashion and thereby combines their individual advantages. At the core of the proposed MPC scheme is a particular type of state-dependent parametrization which is computed data-based in advance offline so that online it can be employed to simplify the numerical solution of the optimization problem. In the thesis, the general approach is introduced, required algorithms are presented and an extensive theoretical foundation is provided. Results for linear as well as for nonlinear dynamical systems are included. Several numerical examples illustrate the approach and highlight its benefits over existing methods.