Impact of Decomposition on Distributed Model Predictive Control: A Process Network Case Study

Abstract

This paper addresses the impact of decomposition on the closedloop performance and computational efficiency of model predictive control (MPC) of nonlinear process networks. Distributed MPC structures with different communication strategies are designed for regulation of an integrated reactor−separator process. Different system decompositions are also considered, including decompositions into local controllers with minimum interactions obtained via community detection methods. The closed-loop performance and computational effort of the different MPC designs are analyzed. Through such a comprehensive comparison, tradeoffs between performance and computation effort, and the importance of systematic choice of the system decomposition, are documented.

A data-driven robust optimization approach to scenario-based stochastic model predictive control

“…are the constant coefficient vector and the lower triangular matrix, respectively, which serve as decision variables of the optimal control problem. Unfortunately, if the disturbance {w t } has an unbounded support, system inputs will be unbounded, thereby inevitably resulting in violations of hard constraints (7). To address this issue, Ref.…”

“…Model predictive control (MPC) has become a commonly accepted technology in various industrial scenarios thanks to its wide applicability and effectiveness in addressing complex optimal control problems subject to input and state constraints [1][2][3][4]. In the context of MPC, process behaviors are characterized by various dynamic models, such as step response models, transfer function models, and state-space models, based on which the dependence of system evolutions in the future on control actions can be explicitly predicted [5][6][7]. Based on the idea of setting the initial state to be the current state of the process, the optimal control sequence can be derived by solving an optimization problem at every control instant, and the first control action is implemented only.…”

Subsystem decomposition of process networks for simultaneous distributed state estimation and control

“…It has been demonstrated by Pourkargar et al that subsystem decomposition may affect the performance of a distributed predictive control system significantly. In the literature, some results on subsystem decomposition for decentralized or distributed control have been obtained.…”

“…The community detection concept originating from network theory provides a very promising way to address the subsystem decomposition problem . By means of the measure of modularity, community‐based approaches have been proposed to find distributed control structures where the subsystems involving state, input, and controlled output variables are made well‐decoupled .…”

“…By means of the measure of modularity, community‐based approaches have been proposed to find distributed control structures where the subsystems involving state, input, and controlled output variables are made well‐decoupled . In the study by Pourkargar et al, the community‐based method was applied to a benchmark chemical process to recommend different distributed control structures, and DMPC was designed based on each structure for extensive performance comparison. In the study by Pourkargar et al, the impact of an optimal decomposition on the control performance of distributed set‐point tracking control was systematically studied.…”