Model predictive control (MPC) is a form of control in which the current control action is obtained by solving on-line, perhaps approximately, an open-loop optimal control problem. An important advantage of this type of control is its ability to cope with hard constraints on the controls and states. Model predictive control has been widely applied in the petro-chemical and related process industries where economic considerations demand operation near the boundary of the set of admissible states and controls. In this talk we first review some of the basics of MPC and overview the types of problems that have been solved. Next we describe an emerging consensus among researchers on the basic necessary components of model predictive control laws. This discussion focuses on the necessary ingredients to obtain closed-loop stability. Next we present one of the major unresolved problem for this control approach, which is receiving current attention, a practical on-line method for dealing with nonlinear dynamic models. A naive implementation of MPC for nonlinear models requires the on-line global solution of non-convex optimization problems. We present a new method that requires less on-line computation and may ease implementation of MPC with nonlinear models.
The work presented was done in collaboration with Prof. D. Q. Mayne, UC Davis and Imperial College.