| Résumé | This article presents a study in which three inverse modeling techniques were applied to hourly heating and cooling load data extracted from 35 office buildings in Ottawa, Canada. These modeling techniques were three-parameter change point models, regression trees, and artificial neural networks. The change point models were trained with outdoor temperature data, whereas the other two models were trained with four regressors: outdoor temperature, wind speed, horizontal solar irradiance, and a binary work hours indicator. The correlations among the change point model parameters of individual buildings were analyzed. The sensitivity of heating and cooling load intensities to the four regressors was examined. The models were used to identify several types of energy use anomalies. The anomalies detected by different modeling techniques were generally in agreement. The results indicate that nearly half of the buildings did not have effective after-hours schedules to save energy. In all but three buildings, the cooling change point temperature was lower than the heating change point temperature—indicating a simultaneous heating and cooling problem. Moreover, a few buildings with anomalies potentially related to high air infiltration or overventilation, high thermal conductance, and high solar heat gains during summer were identified. |
|---|
