Bp 391 en
Quadratic Programming Algorithms for Fast Model-Based Predictive Control
Author: Ondřej Mikuláš
This thesis deals with quadratic programming (QP) algorithms for the use in fast model based predictive control applications. First, general overview of model based predictive control and quadratic programming is given. Then, several QP algorithms - active set method, fast gradient method and interior point method - are described. Then, these algorithms are tested on model predictive control example and on randomly generated QPs. We treat especially computation time required to obtain a solution, the factors that influence the run-time (condition number, correction horizon etc.) and also the typical properties related to the computation time of the above mentioned algorithms. Finally, the thesis deals with robust control performance of model predictive controller without constraints. With selected plants, the effect of plant model mismatch on the robust quality of control is studied in conjunction with correction horizon length.