Exploring building energy performance prediction using graph neural networks

Dalach, Agata and Wang, Zijian and Nousias, Stavros and Borrmann, André; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Exploring building energy performance prediction using graph neural networks. In: EG-ICE 2025. University of Strathclyde Publishing, GBR. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093246)

Preview

Text. Filename: Dalach-etal-EG-ICE-2025-Exploring-building-energy-performance-prediction-using-graph.pdf Final Published Version License: Download (1MB)| Preview

Abstract

Building energy performance prediction is critical for optimizing energy efficiency, and recent studies have increasingly focused on replacing computationally intensive physics-based simulations with faster, data-driven methods. However, the wide variability in building typologies poses challenges when converting building energy models into effective machine learning training samples. The graph data format offers a promising way to simplify the representation of building energy models and introduce relationships among their components. This paper explores a data-driven approach of representing energy models as graphs and adopting graph neural networks (GNN) to predict their energy performance. To achieve that, we first select the total end-use building energy intensity (EUI) as the key building performance criterion to be predicted. We construct a synthetic component-level building energy graph dataset where nodes are building elements, and edges are their spatial and hierarchical relationships. We formulate the energy prediction as a graph regression task and evaluate the performance of several GNN architectures. The results demonstrate that all evaluated GNN models show promising predictive accuracy, highlighting their potential for delivering rapid building performance predictions, especially suitable for the early design stages.

• Share and Export
  

  RDF+N-TriplesRDF+XMLMultiline CSVRIS (EndNote Online, Zotero, Ref manager, etc)BibTeXJSONDublin CoreOpenURL ContextObjectHTML CitationMODSEP3 XMLMPEG-21 DIDLReferOpenURL ContextObject in SpanMETSRefWorksSimple MetadataRDF+N3Data Cite XMLEndNoteAtomASCII CitationRIOXX2 XML
• Item metadata

  Item type:	Book Section
  ID code:	93246
  Dates:	Date Event 1 July 2025 Published 9 June 2025 Accepted
  Subjects:	Fine Arts > Architecture
  Department:	Faculty of Engineering > Architecture
  Depositing user:	Pure Administrator
  Date deposited:	27 Jun 2025 10:28
  Last modified:	03 Jul 2025 09:06
  URI:	https://strathprints.strath.ac.uk/id/eprint/93246