Several efficient algorithms for solving convex multiparametric quadratic programs (mp-QP) have been developed recently. The growing interest in these problems is due to the fact that explicit solutions to model predictive control problems can be obtained by solving multiparametric programs. The parameter space is partitioned into convex polyhedra, each associated with an optimizer function. In the existing algorithms the optimizer function may be discontinuous even if the problem admits a continuous selection, moreover, the polyhedral partition is generally non-unique and may consist of polyhedra with mutually overlapping interiors. We present a method, which under certain assumptions on the problem data, yields a continuous mapping from parameter to solution space. In addition, the polyhedral partition is always unique and non-overlapping.
Back to Control Seminars Page