Hierarchical graph-based method for static daylight prediction of 3D irregular office buildings
Open Access
- 21 January 2026
- conference paper
- Published by EDP Sciences in E3S Web of Conferences
- Vol. 689, 04001
Abstract
Efficient daylight prediction in geometrically complex office buildings remains challenging due to computational constraints and oversimplified representations in existing methods. While data-driven approaches accelerate simulations tenfold, they often neglect architectural form sensitivity and window placement effects. This study overcomes these limitations through a novel graph-based framework integrating physical daylight principles with machine learning. We develop a hierarchical semantic grammar for building topology representation and introduce dihedral angle-based encoding to capture critical glazing-to-space geometric relationships as graph edge features. An adaptive graph convolutional network subsequently aggregates multi-scale neighbor information for light transport modeling. Validated against Radiance benchmarks, the framework demonstrates an 80% accuracy improvement in daylight factor prediction while achieving 30% greater computational efficiency than comparable CNN methods. This approach effectively resolves the persistent accuracy-speed trade-off in performance simulation, enabling robust daylight optimization for sustainable design.Keywords
This publication has 10 references indexed in Scilit:
- A cross-scale normative encoding representation method for 3D building models suitable for graph neural networks2025
- Research on Daylight Inside Buildings Based on Machine Learning2024
- A predictive model for daylight performance based on multimodal generative adversarial networks at the early design stageEnergy and Buildings, 2024
- Research on the generation method of missing hourly solar radiation data based on multiple neural network algorithmEnergy, 2023
- Ensemble machine learning framework for daylight modelling of various building layoutsBuilding Simulation, 2023
- A multilayer perceptron-based fast sunlight assessment for the conceptual design of residential neighborhoods under Chinese policyBuilding and Environment, 2023
- Predictive models for daylight performance of general floorplans based on CNN and GAN: A proof-of-concept studyBuilding and Environment, 2021
- Graph neural networks: A review of methods and applicationsAI Open, 2020
- Relational inductive biases, deep learning, and graph networksarXiv, 2018
- Representation Learning on Graphs: Methods and ApplicationsarXiv, 2018