A case-centered behavior analysis and operation prediction of AC use in residential buildings

This study analyzed behaviors of occupants in residential buildings with a focus on physical environmental elements. To this end, collection and analysis of Air Conditioner (AC) use behavior data, as well as the prediction and validation of AC on–off state were conducted. For the collection of behavioral data collection, IR signals of AC remote were collected, based on which various AC adjustment behaviors (on–off, set-point temperature) that determines energy use, use rate, and set-point temperature were analyzed. Using the analysis, the AC use behavior was defined as on–off state prediction, and the difference between indoor and outdoor temperature was suggested as the standard for expressing AC use rate. Further, analyzing the duration of AC operation for each set-point temperature revealed that there are differences in AC operation propensities of respective units. Next, the collected data was used to predict and test AC on–off state. Three prediction algorithms (Logistic Regression, Random Forest, and Support Vector Machine) were applied, and the results of the cross-validation was evaluated using three criteria: F-measure, Cohen's Kappa, on–off frequency. Results showed superior prediction performance of Random Forest. Despite the importance of resident behavior studies, there has been only a limited number of studies conducted on residential AC use behavior. The research process and analysis method in each step of this study is expected to be helpful for researchers hoping to conduct behavioral studies in the future.