DISTRIBUTED MODEL PREDICTIVE CONTROL OF CONSTRAINED LINEAR SYSTEMS

Abstract

This thesis studies Distributed Model Predictive Control (DMPC) of a group of discrete-time linear systems with and without coupled dynamics or constraints. Several problems are studied: systems with coupled dynamics and systems with independent dynamics but coupled constraints. Under these situations, centralized MPC may be computationally inefficient and more efficient approaches are desired. For a network of dynamically-coupled linear systems, a decoupling strategy is proposed with the use of time-varying terminal sets, which result in a less conservative formulation compared with most DMPC approaches. For a network of linear systems with coupled constraints, this thesis proposes a DMPC approach based on the Alternating Direction Multiplier Method. Under mild assumptions, the convergence is guaranteed. To further accelerate the online computations of the DMPC problem, this thesis also proposes an accelerated distributed dual gradient algorithm, which has faster convergence. Simulation results are provided to show the performances of the proposed DMPC approaches.