A general approach for mixed-integer predictive control of HVAC systems using MILP

Abstract

Abstract The topic of the paper is optimal control of heating, ventilation, and air conditioning (HVAC) systems. Using mixed-integer linear programming (MILP) the main contribution is a flexible and modular MILP model of HVACs. It is the centerpiece of the proposed hierarchical multi-input multi-output (MIMO) model predictive control (MPC) framework for energy, comfort, and wear optimization. On the upper-level a plant MPC takes care of the slow dynamics of a plant. On the lower-level a MILP-MPC based on a MILP optimization model of an HVAC system optimizes the HVAC operation. The MILP-MPC considers the first few samples of the control trajectory of the plant MPC as its reference trajectory. It minimizes power consumption and switching of the HVAC. Thereby it has to obey constraints, consider component characteristics by nonlinearity approximation, and solve a unit commitment problem. The fact that the plant MPC can be based on an almost arbitrary plant model and the MILP model of the MILP-MPC can represent a variety of different HVAC systems unifies HVAC control. Features and results are presented in case studies: The first case study shows that state of the art HVAC operation is outperformed even if the predictive capability is not used. The second case study demonstrates that temperature, CO2 level, and humidity can be controlled simultaneously in a decoupled fashion. The third case study reveals that if also the lower-level MILP-MPC oversees the latency periods of HVAC components optimal switching is achieved.