Optimizing the Communication Topology in a Coordinated Model Predictive Control Architecture

The design of control architectures for large-scale systems formed by the interconnection of several interacting subsystems is a challenging task. The interactions between the subsystems have a significant influence on the local control decisions as well as the overall system optimality and need to be accounted for explicitly. An increasingly popular technique to control such systems is the Coordination-based Model Predictive Control (C-MPC), where a decentralized control architecture is maintained, but the performance of the system is driven towards that of a centralized control architecture. The C-MPC control strategy facilitates communication between the local controllers and ensures cooperation in order to drive the performance towards plant-wide optimality. However, the amount of communication required increases exponentially with the number of subsystems, making the C-MPC architecture computationally inefficient. Hence, it is desirable to reduce the computational load of the C-MPC architecture without significantly compromising on the overall performance of the system. In order to achieve this, a Genetic Algorithm (GA) based optimizer is utilized to identify the trade-off between communication topologies of varying complexities and the associated performance of the C-MPC strategy. The effectiveness of the optimally designed C-MPC framework with partial communication between the controllers is evaluated on a popular benchmark chemical engineering problem and the performance is compared to that of the traditional centralized and decentralized control architectures.