Data-driven approach to estimate urban heat island impacts on building energy consumption

Abstract

Urban heat island effects can significantly increase building energy consumption. Assessing the impact of the urban heat island on building energy use is challenging due to temperature variations. Scalable building models and cooling energy consumption data are essential for accurate load demand predictions. This paper presents a data-driven method to predict cooling load intensity (heat removed per unit floor area to maintain the setpoint in conditioned spaces), under urban heat island conditions in residential buildings in Phoenix, Arizona, using a synthetic dataset of 27,681 buildings. The approach incorporates physics-based parametric modeling using building geometrical features, urban heat island intensity simulation through the Urban Weather Generator, and cooling load intensity estimation via EnergyPlus and OpenStudio. Machine learning models, including Extreme Gradient Boosting, Gaussian Process Regression, Random Forest, Support Vector Regression, and Deep Neural Networks, are employed for cooling load predictions. Urban building energy performance analysis indicates that the Deep Neural Network model performs best in estimating cooling load intensity, achieving a high coefficient of determination of 0.98. These findings support informed decision-making to enhance building energy efficiency, reduce consumption, and facilitate large-scale smart city planning.