Barrier function based model predictive control (Q705163)

A re-centred model predictive control (r-MPC) formulation is developed. A self-concordant barrier function is added into the quadratic penalty for state and input deviations to ensure the strict satisfaction of constraints. A typical interior-point algorithm forces the barrier weighting towards zero. The current r-MPC approach uses a gradient re-centred barrier function in which the barrier weighting is allowed to converge to some positive value. The optimization problem to be solved at each step is effectively unconstrained in the sense that the solution always lies in the interior of the constraint set. An ellipsoidal invariant set is constructed (by exploiting the geometry of self-concordant functions), and asymptotic stability of r-MPC under full state feedback is established. The proposed r-MPC has been successfully applied to control an edible oil refining line.