Semi-automated modular modeling of buildings for model predictive control.

ABSTRACTA promising alternative to standard control strategies for heating, ventilation, air conditioning and blinds positioning of buildings is Model Predictive Control (MPC). Key to MPC is having a sufficiently simple (preferably linear) model of the building's thermal dynamics. In this paper we propose and test a general approach to derive MPC compatible models consisting of the following steps: First, we use standard geometry and construction data to derive in an automated way a physical first-principles based linear model of the building's thermal dynamics. This describes the evolution of room, wall, floor and ceiling temperatures on a per zone level as a function of external heat fluxes (e.g., solar gains, heating/cooling system heat fluxes etc.). Second, we model the external heat fluxes as linear functions of control inputs and predictable disturbances. Third, we tune a limited number of physically meaningful parameters. Finally, we use model reduction to derive a low-order model that is suitable for MPC. The full-scale and low-order models were tuned with and compared to a validated EnergyPlus building simulation software model. The approach was successfully applied to the modeling of a representative Swiss office building. The proposed modular approach flexibly supports stepwise model refinements and integration of models for the building's technical subsystems.