Distributed Model Predictive Control for networks with limited control communication

Heating, ventilation, and air conditioning (HVAC) systems in large buildings frequently feature physical interactions where the outputs of each dynamic subsystem act as disturbances to other subsystems in its neighborhood. Centralized control of such interconnected systems is usually not practical due to the large communication burden, while decentralized control may achieve the non-optimal performance as they don´t account for system interactions. In this paper, a Neighbor-Communication based Distributed Model Predictive Control (NC-DMPC) framework is described that can handle such systems. Along some prediction horizons, the optimizer of each subsystem communicates its predicted optimum setpoints to its neighbors in addition to the costs imposed by the neighbor´s predicted setpoints. Since there are different interconnected subsystems with different prediction and control horizons, each subsystem´s optimizer considers the effects from its upstream neighbors only along its own prediction horizons (state and control horizons). In the proposed NC-DMPC framework, communication between all plants is not necessary to achieve a global optimum. Convergence to Pareto optimal trajectories for the proposed NC-DMPC is proved and a numerical example is used to demonstrate this aspect.