Recent Developments in Constrained Predictive Control
Dr B. Kouvaritakis (University of Oxford)
Abstract
Approaches such as GPC (Generalized Predictive
Control) have made Model Based Predictive Control (MBPC) popular. Yet
in its original formulation, GPC lacked a general stability theory.
Several developments since, involving the introdcution of terminal
constraints or reverting to the use of infinite horizons, have
provided the missing stability guarantee. However the main reason for
using receding horizon MBPC instead of classical LQR control is to do
with the problem of handling constraints. In this context terminal
constraints are not always sensible: they can lead to infeasibility
and instability. The use of infinite horizons on the other hand lead
to infinite dimensional quadratic programming problems. The seminar
will consider these issues and will describe algorithms which provide
effective predictive constrained control.