Effective Features to Predict Residential Energy Consumption Using Machine Learning: Fid-Bau Portal

Effective Features to Predict Residential Energy Consumption Using Machine Learning

Mo, Yunjeong / Zhao, Dong / Syal, Matt

Humans have a greater influence on energy consumption in residential buildings than other types of buildings. Although existing studies focus on how energy consumption is affected by building technologies and occupant demographics, few studies have incorporated the impact of occupant energy use patterns. The goal of this study is to identify the features that affect energy consumption in residential buildings and to measure their predictive performance. The researchers examined the impact of occupants’ energy use behaviors and the energy use patterns of home appliances on home energy consumption. The patterns reflect on a combination of appliances, their use times and frequencies, and the configurations set by users. Data from the Residential Energy Consumption Survey (RECS) are analyzed to select features for prediction, using multiple machine learning algorithms including support vector machine (SVM) and random forest. The results provide a list of features that efficiently predict energy consumption in residential buildings. The selected 32 features achieve 98% of the prediction performance of that from the entire 271 features. This list of effective features can be used to improve the effectiveness of energy saving programs and to educate occupants about their energy use patterns. The relationship between occupants’ behavior patterns and energy use patterns revealed from this study provides the groundwork for researchers to further explore the prediction of occupant behavior from energy consumption.