Investigating the effects of key drivers on energy consumption of nonresidential buildings: A data-driven approach integrating regularization and quantile regression

Investigation of the key drivers of building energy consumption at the city scale could assist in informing energy efficiency policies. This study proposed a data-driven approach by integrating regularization and quantile regression methods. This approach not only identified the key drivers of energy use by group lasso but also moved beyond the typical model of estimating average effects by quantile regression, quantifying the effects of the key drivers under low, medium and high consumption levels. The effectiveness of the proposed approach was evaluated based on a dataset of more than 2000 hospital and education buildings in Chongqing, China. The key drivers of these two types of buildings were found to be mainly related to building characteristics, including gross floor area, number of occupancies and cooling type. Additionally, the quantile regression results showed that most of the key drivers imposed clear heterogeneous effects on energy consumption throughout the quantiles. Compared with the other variables, gross floor area exerts the greatest positive effect on energy use across all the quantiles. The proposed approach can provide a general solution to understanding the effects of key drivers on energy consumption, and the empirical findings could benefit policy design for energy conservation. •A data-driven approach integrated group lasso and quantile regression was proposed.•This approach identified the key drivers of energy consumption for buildings.•This approach quantified the changing effects of the drivers across the quantiles.•The approach was tested by using a dataset containing more than 2000 buildings.