A comparison of implementation strategies for MPC

Abstract

Four quadratic programming (QP) formulations of model predictive control (MPC) are compared with regards to ease of formulation, memory requirement, and numerical properties. The comparison is based on two example processes: a paper machine model, and a model of the Tennessee Eastman challenge process; the number of free variables range from 150-1400. Five commercial QP solvers are compared. Preliminary results indicate that dense solvers still are the most efficient, but sparse solvers hold great promise.