Neighbor-communication Model Predictive Control and HVAC systems

Presented in this paper is a variation of the Distributed Model Predictive Control (DMPC) algorithm that converges to the centralized MPC solution for a steady state target optimization with a minimum of communication between plants. This variation, referred to as neighbor-communication MPC, leverages the network topology frequently found in large building HVAC systems; namely, a series of plants where the output of each is a disturbance to other plants. Each plant communicates to its neighbor its intended control action as well as the effects others' actions will have on itself. By judicious construction of the cost functions, all of the cost information is propagated through the network, allowing a global optimal solution to be reached. The novelty of this approach is that communication between all plants is not necessary to achieve a global optimum, and that changes in one controller do not require changes to all controllers in the network. The approach is demonstrated with a simulated example.