Control-Oriented Hybrid Modeling Framework for Building Thermal Modeling

Abstract

Model predictive control has a high potential of delivering superior operations of buildings compared to conventional approaches. However, the unavailability of a scalable building modeling approach has hampered the broad adoption of MPC in buildings. Building spaces have unmeasured time-varying heat gains that evolve on similar timescales as the relevant building dynamics, making building model identification/parameter estimation challenging. In this chapter, a hybrid modeling framework for building thermal modeling is developed. The modeling framework consists of a control-oriented thermal resistance-capacitance (RC) model for the building thermal dynamics, a data-driven model for forecasting the unmeasured time-varying heat gains from external sources, and a data-driven model for modeling the nonlinear dynamics of the cooling or heating rate of the heating, ventilation, and air conditioning equipment with respect to the temperature setpoint. Specialized training approaches are developed for parameter estimation of the thermal RC model parameters to minimize the parameter bias caused by the correlation between the known and unknown inputs. The modeling framework is applied to an illustrative building space to demonstrate the prediction accuracy of the trained hybrid model.