Transfer function modeling of linear dynamic networks for distributed MPC

Abstract

Distributed model predictive control (DMPC) is a new technology for controlling large, geographically distributed systems, such as traffic networks and petrochemical plants, that advocates the distribution of sensing and decision making. However, some challenges should be overcome before DMPC can be deployed in large scale. This paper presents a distributed optimization framework based on the approach of generalized predictive control (GPC), where the prediction model is based on transfer functions, unlike other approaches that are based on state-space models. The motivation for this work is the ability of transfer-function models to represent systems with transportation delay in a reduced form, leading to faster algorithms for predictive control. This is of fundamental importance for the use of DMPC in on-line control loops, particularly so in high-order systems. The paper presents some theoretical results and a comparison between transfer-function and state-space models used in a DMPC application to a distillation column.