Exploring spatial and environmental heterogeneity affecting energy consumption in commercial buildings using machine learning

Building energy consumption (BEC) is a critical indicator for promoting sustainable building development. However, previous studies have neglected the spatial heterogeneity of the urban environment and its landscape configuration on BEC impacts. This paper aims to identify the factors influencing commercial building energy consumption in Singapore, including building basic information, Green Mark certification, usage behaviours, and urban environment. We employed geographically weighted regression models to analyse influencing factors, K-means to classify clusters, and artificial neural networks for prediction. Our approach yielded a significant improvement in the fitting effect (by 38%) compared to traditional regression algorithms. Our findings showed a strong correlation between BEC and geographical information and economic development. Building basic in-formation was the most influential aspect, with building coverage area being the most dominant driving force with positive impacts, reaching a mean regression coefficient of 0.423. Land surface temperature had the most potent negative effect on BEC among urban factors, while water area ratio and gross plot ratio had a strong non-stationary influence on BEC. Our study provides insights into the evolving and heterogeneous nature of the urban environment, supporting decision-making for more sustainable metropolitan development that benefits both people and nature.