This thesis presents different novel concepts for computationally efficient multistep model predictive electric drive control.
The drive consists of an induction machine and a three level-neutral point clamped (NPC) inverter.
In order to control this inverter it is necessary to keep the DC-link voltage balanced.
Therefore a nonlinear mixed integer optimization problem has to be solved.
After developping a multistep prediction model for the neutral point voltage this thesis presents three different computationally efficient solutions to the coresponding optimization problem:
1) A priori state selection which uses a reduction of the inverter states taken into account.
2) Linearization of the nonlinear prediction model and
3) Design of a nonlinear branch-and-bound (called nonlinear sphere decoding) algorithm.
Afterwards
the methods are compared by means of MATLAB/ SIMULINK simulations.
«