Efficient algorithms for model predictive control of input constrained linear systems

Abstract

Model Predictive Control (MPC) of both-side input constrained linear systems is formulated as quadratic programing problems (QP), which are solved time and time again. In general, a QP to be solved at each iterate is a slight modification of that in previous iterate, and hence, iterative algorithms, which exploit previous information, are promising. In this paper, we apply semismooth Newton method and projected Gauss-Seidel method to MPCs and show that the methods with appropriate initial solutions are practically very efficient by computational experiences.