A scalable QP solver for optimal control of cascades with constraints

Abstract

A finite-horizon linear-quadratic control problem is studied. The structure of this problem is such that the input constraints, state constraints, and performance index, all separate across the underlying cascade of dynamical sub-systems. An equivalent quadratic program is formulated, for which a custom interior-point method is devised that exploits the special spatial structure of the problem. The computational burden of this method scales linearly with the number of sub-systems. By contrast, the computation cost scales cubically with the time horizon. Therefore, the custom method is advantageous in cases where the number of sub-systems is large relative to the time horizon. A numerical exa..