Model Predictive Control, System Identification
Within the last ten years, growing pressure to reduce energy consumption of buildings has led to an increased focus on the development and deployment of advanced control strategies. Heating, Ventilation, and Air-Conditioning (HVAC) constitutes the majority of the energy consumption of buildings. Model predictive control (MPC) has gained significant attention in HVAC control as it computes the control inputs for a given system by iteratively solving an optimal control problem on-line. The problem formulation accounts for the system operating objective and constraints. Several application studies of MPC applied to buildings have been reported in the literature. These studies have demonstrated the benefits of the application of MPC schemes to buildings. However, one theme that appears in some of the literature is the lamentation on the difficulty to apply MPC broadly to buildings. This is a challenging problem because the MPC system design needs to include a robust and broadly applicable system identification methodology to effectively address this problem. Moreover, in many building applications, the desired sensors measuring key variables are not available (e.g., a heat disturbance load and power consumption measurements are usually not available for residential zones controlled by a thermostat). In this work, an economic MPC scheme is developed to manipulate the temperature setpoint of a zone controlled by a thermostat. The MPC scheme is equipped with an economically-oriented objective that includes a system identification procedure, a state estimator, and a control problem formulation. The economically-oriented MPC seeks to minimize the utility bill by manipulating the setpoint to leverage the building mass as thermal energy storage while maintaining the zone temperature setpoint within a comfortable range. Given the lack of a power or HVAC load measurement in a typical thermostat, the HVAC load is approximated by a filtered version of the thermostat stage commands, which provides a normalized time-average version of the HVAC load. All the components of the resulting MPC scheme are designed in a manner to address general applicability of the resulting MPC. Simulation results are presented to demonstrate the effectiveness of the strategy.