Going Beyond the Mean: Distributional Degree-Day Base Temperatures for Building Energy Analytics Using Change Point Quantile Regression

Building energy consumption patterns are primarily affected by building function, operation, occupancy and thermal characteristics. A robust method of energy use pattern recognition is, therefore, essential. Heating degree-days (HDD) are routinely used for heating energy consumption prediction and analytics, the accuracy of which depends on how well the base temperature corresponds with the patterns of energy use. A change-point quantile regression (CPQR) technique is proposed for better identification of the base temperature, which is then applied in three buildings with distinct operational energy use patterns: weekday only, weekday plus occasional weekend, and all-year operation. Compared with the conventional regression and change-point least square (CPLS) methods, our CPQR approach determines a range of base temperatures of corresponding energy use patterns across quantiles from 0.05 to 0.95, at an interval of 0.05. Consequently, daily HDDs computed using the range of base temperatures of corresponding quantiles result in more accurate predictions of heating energy consumption. CPQR improves estimation accuracy and is more robust than CPLS because (a) it considers the whole distribution of energy consumption not just the mean, (b) pre-processing of raw data other than the removal of anomalies is not needed, and (c) it can better characterize the data with abnormal energy distribution. Also, CPQR-based method can better characterize the weather dependence of energy consumption than the conventional CPLS regression.