Decentralized model predictive control of constrained linear systems

This paper proposes a novel decentralized model predictive control (MPC) design approach for open-loop asymptotically stable processes whose dynamics are not necessarily decoupled. A set of partially decoupled approximate prediction models are defined and used by different MPC controllers. Rather than looking for a-priori conditions for asymptotic stability of the overall closed-loop system, we present a sufficient criterion for analyzing a posteriori the asymptotic stability of the process model in closed loop with the set of decentralized MPC controllers. The degree of decoupling among submodels represents a tuning knob of the approach: the less coupled are the submodels, the lighter the computational burden and the load for transmission of information among the decentralized MPC controllers, but the less performing is the control system and the less likely the proposed stability test succeeds. The designer can therefore trade off between simplicity of computations/limited transmitted information and performance/stability.