Files

1-s2.0-S0378778826011059-main.pdf
  • Embargo Access until 2027-04-15
  • Adobe PDF
  • 3.88 MB
Request a copy

Details

Authors
Abstract
(en) Urban building energy predictions are essential for assessing decarbonization strategies and informing evidence-based urban planning under climate change scenarios with rapid urbanisation. While physics-based building energy modelling is theoretically robust, their application at the urban scale is constrained by high computational demands and limited data on building stock characteristics. Machine learning (ML) offers scalable, data-driven alternatives and has gained prominence in Urban Building Energy Modelling (UBEM). However, existing ML applications remain fragmented across domains and methodological families, and previous reviews have largely focused on individual model types, or limited subset of approaches. Effectively, a systematic comparison and clear taxonomy of ML methods for UBEM is lacking. This study addresses this gap by developing a comprehensive, PRISMA-guided taxonomy of ML applications in UBEM published over the past five years. The taxonomy is structured around major ML model families, providing a conceptual framework to classify modelling approaches and their applications within UBEM workflows. For each model family, the reported performance and contextual suitability/fit are synthesized to clarify strengths and limitations of each method. Additionally, strategies for enhancing applicability of ML in UBEM workflows are reviewed, including input pre-processing, validation strategies, performance metrics, and interpretability techniques. Recurring challenges are effectively highlighted-including limited real-world validation, inconsistent benchmarking standards, and limited generalizability across urban contexts. By consolidating recent advances into a structured framework, this review provides both: a primer on ML methods for UBEM, and guidelines for aligning model selection and workflow design with specific UBEM tasks. This contributes to more consistent, scalable, and reliable integration of ML in UBEM, for supporting research, practice, and policy aimed at enabling and accelerating sustainable urban energy transitions.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Liang, S., Maskarenj, M., Llaguno, M., & Altomonte, S. (2026). A systematic review of machine learning based methods for applications in urban building energy modelling. Energy and Buildings, 369, 118045. https://doi.org/10.1016/j.enbuild.2026.118045 (Original work published 2026)