Building hourly thermal load prediction using an indexed ARX model

This paper introduces an easily implementable and computationally efficient, ARX (autoregressive with exogenous, i.e., external, inputs) time and temperature indexed model for 1 h ahead building thermal load prediction. Time and temperature indexing implies that different sets of coefficients are used in the predictive equation depending on the time of the day or the ambient temperature. The indexing and proposed structure of the model follows physically motivated interpretations of the loading conditions and thermal response of the building. One of the main contributions of the proposed model is that it allows determining the dominant factors that affect the thermal load at a given time. A free and widely adopted building energy and thermal load simulation program from the U.S. Department of Energy is used to determine the prediction accuracy of the proposed model on several different benchmark-building types: a small office building, a medium office building, a midrise apartment, and a high-rise apartment.

► An indexed ARX regression model is introduced.
► Two indexing methods using time periods and temperature levels are provided.
► Cooling and heating load prediction results in several building types are supplied.
► The building capacitance effect is considered for better efficiency and accuracy.
► Model comparison for hourly thermal load prediction is provided.