Distributed model predictive control of integrated process networks: Optimal decomposition for varying operating point

This paper is focused on investigating the impact of operating conditions on decompositions of nonlinear integrated process networks for distributed control purposes. The strength of variable interactions is incorporated through weighted graphs, and subsequent community detection-based decomposition leads to distributed controllers with weak inter-controller interactions. Since the operating condition affects the variable interactions (weights of the graph edges), it changes the subsystems' geometry and how the state variables, manipulated inputs, and output variables distribute among them. The resulting system decompositions are then employed for a distributed model predictive control (DMPC) design. A typical reactor-separator process network is considered to implement the proposed distributed control architecture. The computation efficiency and closed-loop performance are evaluated for DMPC using the proposed decomposition method at different operating points and compared to the performance using intuitive decompositions based on physical configurations and material/energy balances.