Stability and optimality of distributed model predictive control

Abstract

This article extends existing concepts in linear model predictive control (MPC) to a unified, theoretical framework for distributed MPC with guaranteed nominal stability and performance properties. Centralized MPC is largely viewed as impractical, inflexible and unsuitable for control of large, networked systems. Incorporation of the proposed distributed regulator provides a means of achieving optimal systemwide control performance (centralized) while essentially operating in a decentralized manner. The distributed regulators work iteratively and cooperatively towards achieving a common, systemwide control objective. An attractive attribute of the proposed MPC algorithm is that all intermediate iterates are feasible and the resulting distributed MPC controllers stabilize the nominal closed-loop system. These two features allow the practitioner to terminate the distributed control algorithm at the end of each sampling interval, even if convergence is not attained. Distributed MPC with output feedback is addressed using the well established Kalman filtering framework for state estimation.