Noncooperative Distributed Model Predictive Control: A Multiparametric Programming Approach

Abstract

The distributed control system architecture strikes a balance between the decentralized control system architecture, where subsystem interactions are unaccounted for, and the computationally expensive centralized control system architecture. Subsystem interactions can be significant in chemical process systems, especially when energy or material recycle loops are present. A drawback of the distributed control system is that it is computationally expensive as it requires intermediate iterations involving the solution of multiple optimization problems to be performed. To address this drawback, we develop a noncooperative, iterative, multiparametric distributed model predictive control (mpDiMPC) technique with an aim to decrease the computational costs of conventional, online distributed controllers by avoiding the need to solve an optimization problem at each intermediate iteration. We apply the developed control algorithm on an interacting reactor–separator process and study its control and computational performance. For the case study presented in this paper, mpDiMPC resulted in a reduction in computational costs by approximately 95% compared to its online counterpart.