Investigating structural and occupant drivers of annual residential electricity consumption using regularization in regression models

Highlights: • Reviews literature on modeling energy consumption and determinants of residential electricity consumption. • Reviews regularization methods in regression analysis to improve prediction and model interpretability. • Assembles dataset of 58 predictive factors and annual electricity usage for one thousand households California. • Applies five regularization techniques to these data to compare model prediction and variable selection. • Highlights: influence of occupant socio-demographics, physical dwelling characteristics, and occupant behaviors. -- Abstract: Achieving further reductions in building electricity usage requires a detailed characterization of electricity consumption in homes. Understanding drivers of consumption can inform strategies for promoting conservation and efficiency. While there exist numerous approaches for modeling building energy demand, the use of regularization methods in statistical models can address challenges inherent to building energy modeling while also enabling more accurate predictions and better identification of variables that influence consumption. This paper applies five regularization techniques to regression models of original survey and electricity consumption data for more than one thousand households in California. It finds that of these, elastic net and two extensions of the lasso—group lasso and adaptive lasso—outperform other approaches in terms of prediction accuracy and model interpretability. These findings contribute to methodological approaches for modeling energy consumption in buildings as well as to our understanding of key drivers of consumption. The paper shows that while structural factors predominate in explaining annual electricity consumption patterns, habitual actions taken to save energy in the home are important for reducing consumption while pro-environmental attitudes and energy literacy are not. Implications for improving building energy modeling and for informing demand reduction strategies are discussed in the context of the low-carbon transition.