Game approach to distributed model predictive control

Abstract

This study introduces a framework for distributed model predictive control (MPC) based on dynamic games, where centralised and decentralised control algorithms can be viewed as dynamical games with coupled control sets. The original optimisation problem is decomposed into smaller coupled optimisation problems in a distributed structure, which is solved iteratively. Then, the resulting dynamic game is analysed using the theory of potential games to derive the properties of the resulting algorithms. This sheds new light on the properties of existing MPC algorithms and allows us to establish a unified framework to analyse them. The control problem of a heat-exchanger network (HEN) is used to illustrate the effectiveness, practicality and limitations of the proposed framework.