(687g) Distributed Model Predictive Control of Integrated Process Networks Using Community Detection

The applicability of model predictive control (MPC) as a ubiquitous control method in today’s chemical processes, hinges on the solvability of the underlying constrained dynamic optimization problem in real-time [1]. Such a concern makes the centralized approach to MPC challenging, especially for large-scale and highly integrated process networks [2]. An alternative solution is distributed model predictive control (DMPC), where we decompose the large-scale optimization-based control problem into multiple weakly interacting sub-problems, which can be solved in-parallel to reduce the complexity [3].

The system decomposition lies at the heart of DMPC design. It identifies the number of controllers and describes how the manipulated inputs and controlled output variables must be distributed across the control system. The closed-loop performance and the computational efficiency of the DMPC are impacted directly by a selected decomposition method that alters the geometry of subsystems and the communication between them [2].

Community detection on the representing graphs of integrated process systems has been shown a promising approach to synthesize distributed estimation and control architectures [4, 5]. In previous work, a community detection-based DMPC architecture was developed by maximizing the modularity of the representing directed graphs [6, 7]. Such a decomposition method results in a minimum inter-subsystem communication without considering the strength of interactions between variables belonging to different subsystems [8].

In this work, we explore the impact of variables’ interactions on the community detection-based DMPC of complex process networks. Specifically, decomposition of the control problem into distributed sub-controllers is pursued to maximize the strength of input-output impact. To this end, the control problem is recast as a weighted directed graph and subsequently decomposed using community detection algorithm. As the strength of variable interaction is dependent on the operating state (e.g. recycle ratio), the corresponding optimal decomposition changes based on the operating state. We synthesize an iterative DMPC for output regulation of based on such distributed architecture and we evaluate the closed-loop performance and the computation time of the proposed DMPC for a benchmark reactor-separator process network over a wide operating range. It is illustrated that, for each operating point, the optimal decomposition maximizing the input-output impact performs better compared to ad-hoc architectures like topology-based or balance variable-based decompositions.

1. Rawlings, J.B.; Mayne, D.Q.; Diehl, M. Model Predictive Control: Theory, Computation, And Design; Nob Hill Publishing: Madison, WI, 2017.
2. Pourkargar, D.B.; Almansoori, A.; Daoutidis, P. The impact of decomposition on distributed model predictive control: A process network case study. Ind. Eng. Chem. Res. 2017, 56, 9606−9616.
3. Christofides, P.D.; Scattolini, R.; Munoz de la Pena, D.; Liu, J. Distributed model predictive control: A tutorial review and future research directions. Comput. Chem. Eng. 2013, 51, 21−41.
4. Yin, X.; Liu, J. Subsystem decomposition of process networks for simultaneous distributed state estimation and control. AIChE J. 2019, 65, 904−914.
5. Pourkargar, D.B.; Moharir, M.; Almansoori, A.; Daoutidis, P. Distributed estimation and nonlinear model predictive control using community detection. Ind. Eng. Chem. Res. 2019, 58, 13495−13507.
6. Jogwar, S.S.; Daoutidis, P. Community-based synthesis of distributed control architectures for integrated networks. Chem. Eng. Sci. 2017, 172, 434−443.
7. Pourkargar, D. B.; Almansoori, A.; Daoutidis, P. Comprehensive study of decomposition effects on distributed output tracking of an integrated process over a wide operating range. Chem. Eng. Res. Des. 2018, 134, 553−563.
8. Jogwar, S.S. Distributed control architecture synthesis for integrated process networks through maximization of strength of input-output impact. J. Process Control 2019, 83, 77−87.