Multi-stage regression linear parametric models of room temperature in office buildings

Mathematical models of the heating, ventilation, and air conditioning (HVAC) components play an important role in control design and fault detection of the system. The work in this paper incorporates architectural parameters in linear parametric models of room temperature in office buildings. Specifically, we allow the physics-based autoregression moving average (pbARMAX) model to have a multi-stage structure in order to explicitly include the architectural parameters of the room in the model. Extensive measurements of the room temperature are used to develop and validate the multi-stage model. The resulting model can predict the temperature in different rooms accurately in both short-term and long-term. Over a period of four weeks, the predictions have a root mean squared error less than 0.10 with a coefficient of determination larger than 0.99.

► We study autoregressive moving average models of room temperature.
► A two-stage regression structure is developed.
► The architectural parameters are explicitly included in the models.
► The models demonstrate improved prediction compared to existing models.
► The models provide analytical foundations for parametric study and control design.