Effect of Control Structure on the Performance of Distributed Model Predictive Control

Distributed model predictive control (DMPC) has emerged as a powerful approach for large scale integrated systems due to its ability to provide a balance between performance, computational time, and real-world implementability. The decomposition of the overall control problem into the distributed subsystems is an important aspect of the design of a DMPC system. This paper aims at establishing a connection between structural properties of the distributed architecture with the corresponding closed-loop performance. Specifically, the goodness of decomposition is quantified using a well-known concept of modularity from graph theory. For an octuple tank system, four different distributed control structures varying in the number of subsystems and decoupling objectives are proposed. The closed-loop performance of these controllers are compared and subsequently correlated with the corresponding modularity of the structure. Thus the proposed work provides a tool to screen candidate distributed architectures.