Strong Formulations for Hybrid System Control

Abstract

We study the mixed-integer quadratic programming formulation of an $n$-period hybrid control problem with a convex quadratic cost function and linear dynamics. We first give the convex hull description of the single-period, two-mode problem in the original variable space through two new classes of valid cuts. These cuts are then generalized to the single-period, multi-mode, multi-dimensional case and applied to solve the general $n$-period hybrid control problem. Computational experiments demonstrate the effectiveness of the proposed strong formulations derived through the cut generation process in the original variable space. These formulations yield a substantial reduction in computational effort for synthetic test instances and instances from the energy management problem of a power-split hybrid electric vehicle.