Data-driven Prediction of Residential Building Energy Performance at an Urban Scale through End-use Demand Segregation

Urban planners face significant challenges when identifying energy performance opportunities in the context of strategic planning for efficient and sustainable urban environments. Over the past few decades, predictive modeling using sparse data has aided in forecasting building energy use. However, end-use energy prediction studies can focus on individual buildings. This paper presents an integrated use of building simulation, parametric analysis, and machine learning prediction models as a solution to predict building energy performance at an urban level. The experimentation procedure focuses on Irish semi-detached building archetypes, using parametric analysis to develop a synthetic building dataset and comparing different machine learning algorithms. Results show that the Gradient Boosted Trees algorithm provides a better prediction score, and the implementation of the segregation method improves the overall accuracy for electricity, heating, and domestic hot water demand predictions. The study allows stakeholders such as energy policymakers and urban planners to make informed decisions when planning retrofit measures on a large scale.