Deep learning, Real-time wind speed prediction, Wind-driven design, Landscape form finding
| Abstract |
Wind environment simulation is important in quantitatively analyzing urban microclimate studies, but its time-consuming, costly and complex setup is widely regarded as a challenge. Many studies apply deep learning (DL) models to predict wind speed for rapid and accurate prediction. However, the process of feature engineering for AI is still time-consuming. In this study, we evaluate a new data sampling method to train AI for ambient wind speed simulation that doesn’t require paired data feature labeling. This paper leverages this convenient method of training a custom wind environment prediction DL model to impose a wind-driven landscape form-finding workflow that satisfies wind environment performance and enhances architects’ decision-making. This study can provide a framework that connects wind simulation and landscape form finding, offering real-time feedback and continuous design guidance for wind-driven landscape form-finding by real-time parameters feedback and architectural perception inputs. The results show that the CycleGAN method that without labelling can also simulate the wind speed accurately. Real-time wind speed simulation enables rapid landscape layout findings based on the architect’s perception that provides economical solutions for improving the wind environment. This method has the potential to improve the quality of the urban microclimate that can help designers and policymakers to make informed decisions for early-stage design exploration.