Powering the future: Unraveling residential building characteristics for accurate prediction of total electricity consumption during summer heat

Elevated electricity consumption during summer heat poses significant challenges for urban energy management. This study employs a novel data-driven bottom-up machine learning approach to predict electricity consumption and identify influential building-related characteristics in Beijing, China. Through mobile survey campaigns, we collected comprehensive electricity consumption data (24,439 records) and detailed building information for 2,087 buildings in 209 neighborhoods. Our models achieved high accuracy, with R2 of 0.80 (RMSE 11.77 kWh, MAE 8.70 kWh) at the household level and R2 of 0.95 (RMSE 4.56 kWh, MAE 3.13 kWh) at the building level. We identified specific building characteristics associated with higher electricity demand, including housing sizes of 86–221 m2, floors 10–25, >3 households per floor per unit, buildings over 19 years old, and higher housing prices. At the neighborhood level, a building density of 21.7%–22.3% and low road network density were linked to higher electricity demand. Notably, summer electricity consumption was 20.08% higher on workweeks and 21.29% higher on weekends compared to autumn. This comprehensive approach provides valuable insights for targeted energy efficiency strategies and urban planning.