Residential building energy performance prediction at an urban scale using ensemble machine learning algorithms

Abstract: Data-driven performance assessment techniques have proven to be a viable solution at the urban scale. However, the data-driven performance assessments so far have often been limited in scope, scale and lacked key parameters. This paper proposes a workflow to integrate building archetypes’ simulations, parametric analysis, and ensemble-based machine learning techniques to accurately predict building energy performance at an urban level. The result presented focuses on Irish residential buildings by generating a synthetic dataset using parametric analysis of crucial features of semi-detached building archetypes. The results show that the ensemble method gives higher-quality prediction when compared to traditional machine learning algorithms.