| id |
ecaade2024_105 |
| authors |
Wu, Zhaoji; Li, Mingkai; Liu, Wenli; Wang, Zhe; Cheng, Jack C.P.; Kwok, Helen H.L. |
| year |
2024 |
| title |
A Data-Driven Model for Sustainable Performance Prediction of Residential Block Layout Design Using Graph Neural Network |
| source |
Kontovourkis, O, Phocas, MC and Wurzer, G (eds.), Data-Driven Intelligence - Proceedings of the 42nd Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2024), Nicosia, 11-13 September 2024, Volume 1, pp. 575–584 |
| doi |
https://doi.org/10.52842/conf.ecaade.2024.1.575
|
| summary |
Performance evaluation plays a pivotal role in sustainable architectural design, guiding the design direction towards sustainable objectives. Building simulations based on physical modeling are commonly adopted for performance prediction, but the high computational cost hinders their applications in early design stages that require prompt feedback. Surrogate models have been proposed to emulate the expensive high-fidelity building simulation models using data-driven models. Several studies have been conducted to develop surrogate models for sustainable performance prediction of residential block layout design, but the features proposed by these studies were based on specific cases and cannot represent general residential block layout design. To overcome this gap, this study proposes a novel surrogate model for multi-objective sustainable performance prediction based on graph neural network (GNN), which can be adopted in practical early design stages of residential block layout design. First, a graph schema is proposed to represent the general topological relations among components in residential block layout design. Second, an architecture using graph attention network (GAT) is proposed for multiple sustainable performance predictions. Third, a dataset is established based on parametric design models of residential blocks and simulations of sustainable performance, including energy consumption, daylighting, and thermal comfort. Fourth, the proposed surrogate model using the proposed architecture are trained and fine-tuned to learn the relationship between the residential block design and sustainable performance. Finally, the proposed model is evaluated in terms of accuracy, comparing with benchmark models using graph convolutional network (GCN) and artificial neural network (ANN). The results show that the proposed model (GAT) outperforms the benchmark models (GCN and ANN). The proposed model can achieve a satisfactory accuracy with small CV(RMSE)s of 11.97%, 7.88% and 10.11% in terms of energy use intensity (EUI), annual comfort hour (ACH) and useful daylight illuminance (UDI) in the test dataset. |
| keywords |
Surrogate model, Graph neural network, Building performance prediction, Sustainable building design, Residential block |
| series |
eCAADe |
| email |
|
| full text |
 file.pdf (729,805 bytes) |
| references |
Content-type: text/plain
|
| last changed |
2024/11/17 22:05 |
|
