Hierarchical cooperative distributed model predictive control

Cooperative distributed model predictive control has recently been shown to provide stabilizing feedback for plants composed of any finite number of dynamically coupled subsystems. No assumption is made about the strength of the coupling between the subsystems. This control scheme provides optimal feedback in the limit of infinite optimization iterates, can be terminated before convergence of the iterates, and guarantees stability and feasibility. These properties require communication between all subsystems in the plant at every iterate, however. In this paper, we weaken this requirement and propose an extension in which the subsystems are grouped in a hierarchy. The subsystems communicate with their neighbors at every iterate, but communicate with subsystems outside their neighborhood on a slower and asynchronous time schedule. We show that this extension is plantwide stabilizing. We introduce a method to modify the information sent between neighborhoods in order to reduce the volume of communication and to hide input trajectories between neighborhoods. To achieve these properties, the leader of each subsystem performs a minor additional calculation at each plantwide information exchange. We conclude with an example demonstrating control performance in a chemical plant.