Actuator selection for cyber-physical systems

Abstract

In cyber-physical systems (CPS), the problem of controlling resources can be depicted as an actuator selection problem. Given a large library of actuators and a control objective, what is the least number of actuators to be selected, and what is the corresponding optimal control law? These dynamic design questions are inherently coupled. In this paper, we show that a breadth of actuator selection and optimal control problems (stabilizability, robust and LQR control routines, control of uncertain, nonlinear systems) that do not satisfy the submodularity property lead to the formulation of two classes of combinatorial optimization routines for unstable CPSs: mixed-integer semidefinite programs and mixed-integer bilinear matrix inequalities. Branch-and-bound and greedy algorithms are proposed to address the computational complexity, and numerical results are given to illustrate the proposed formulations.