System Identification for Model Predictive Control of Building Region Temperature

A system identification procedure is presented for identifying the models used in a cascaded model predictive control (MPC) architecture for airside demand response. In the cascaded MPC approach, the application considered is temperature control of a region, which is a collection of zones served by the same air handling unit. The outer loop MPC leverages a dynamic model to predict the region temperature and the corresponding power consumption required to produce the cooling duty for the region. Subsequently, the outer loop MPC determines a power profile that optimizes the energy costs by appropriately shifting the cooling load while maintaining occupant comfort constraints. The power profile is sent to the inner loop MPC, which is also formulated with a dynamic model. The inner loop MPC adjusts the region temperature setpoint to force the actual power consumption to track the desired power profile. In the developed system identification approach, two grey-box models are developed that capture the relevant dynamics on the time-scale of interest for the outer and inner loop MPCs. The grey-box models are parameterized, and the resulting model parameters are fit to input-output data for a particular region application so that the resulting model accurately predicts the temperature and power consumption of the region. State and disturbance estimation, which is required by the MPCs, is performed via a Kalman filter with a steady-state Kalman gain.