Optimizing Urban Energy Efficiency Through a Machine Learning-Driven Framework: A Case Study in Reykjavik

This paper explores the potential of digital tools to guide early-stage design towards energy efficiency at the neighborhood scale, addressing escalating environmental concerns and energy costs. This paper tests a novel, data-driven framework that employs machine learning to optimize urban energy consumption by analyzing building typology, morphology, and energy usage patterns. The framework operates through a multi-phase urban design strategy, beginning with evaluating building forms and climate data to provide early insights into energy usage. Subsequent phases predict and optimize energy use intensity (EUI) across individual buildings, incorporating technical assumptions and seasonal variations to refine predictions and reduce overall energy consumption. Finally, the methodology extends to a comprehensive urban scale, focusing on clusters of buildings. This integrated approach is tested through a case study in the Ártúnshöfði area of Reykjavik, Iceland. The framework’s applicability is evaluated through two phases, replicating a standard workflow. Initially, multiple architectural options and building typologies are evaluated based on building coverage ratio, green coverage ratio, and shape factor. Following the selection of the most efficient solution, a machine learning algorithm using ensemble techniques like CatBoost Regressor assesses the energy use intensity of nine different solutions. The results show high predictive accuracy, with R2 scores of 0.88 for both cross-validation and test datasets. The model demonstrated reliability, with discrepancies between the test model and a business-as-usual energy model ranging from -9 to +12%. These findings underscore the framework’s potential to integrate energy efficiency into urban planning from the earliest stages, providing accurate and reliable predictions that can inform sustainable design and policymaking. Moreover, the machine learning-based framework avoids the need for numerous detailed point simulations, streamlining the overall design process.