Quantitative impact analysis of driving factors on annual residential building energy end-use combining machine learning and stochastic methods

Abstract

Residential buildings consume a large amount of energy in operation, which can be reduced by planning energy-efficient design and operation strategies. Understanding and quantifying the impact of key driving factors to the residential building energy end-use is essential for promoting more energy-efficient building design and operation schemes. Although the climate, building, and occupant-related features have been proven as the key driving factors to residential building energy end-use, their impacts are rarely compared and quantified simultaneously. This study conducts the first attempt in combining the machine learning method with Monte Carlo method to quantify the impacts of these key driving factors simultaneously. Data collected from the Residential Energy Consumption Survey 2015 of the U.S. was investigated in this study. The results indicate that the total energy end-use has positive correlations with the total building area, rooms’ numbers, windows’ numbers, indoor heating temperature setpoint and the occupants’ age; and has a negative correlation with the cooling temperature setpoint; the impacts of heating degree days and cooling degree days are nonlinear and complicated; the impact of the level of insulation is nuanced and offset by the harsh climate. The results of this study contribute good references to policymakers and architects for the synthesis of energy-efficient residential building design and operation; guidelines can be developed for the future survey on residential building energy for relevant and precise data collection to improve the building energy modelling.