Demand Response in Smart Districts: Model Predictive Control of Building Cooling.

The present paper develops and compares three different Model Predictive Control (MPC) strategies for Demand Response (DR) of building cooling in a "smart district": Decentralized MPC (DeMPC), Centralized MPC (CeMPC), and scheduled Distributed MPC (DiMPC). A standard DeMPC approach leads to peaks in aggregated cooling power demand, as local controlling agents optimize each building decentrally. Because these peaks can stress the local power grid severely, we present and compare two additional MPC approaches, CeMPC and DiMPC, to enforce a constraint on the aggregated power use of the district. Our simulation results show that CeMPC and scheduled DiMPC reduce the peak load of the total cooling power in the district by 50 % while still providing DR and cooling the buildings adequately.