Study on predictions of wind velocity distribution around buildings using 3D-CNN

Abstract

Predictions of pedestrian wind environment and planning of countermeasures against strong winds are required in an environmental impact assessment for urban developments. It has been difficult to conduct assessment of the wind environment for each design modification because of the high cost of wind tunnel experiments or numerical simulations for predicting wind environments in complex urban areas. In this study, the author has developed a wind environment simulator using three-dimensional convolutional neural network (3D-CNN) that enables predictions of wind environment in complex urban areas to be made with speed and ease. This paper describes the methodology for prediction of the wind environment by 3D-CNN, the learning method, and the validation results.