Deep learning methods evaluation to predict air quality based on Computational Fluid Dynamics

作者：

Highlights：

• Methods to convert CFD inputs for neural networks are presented.

• Deep Learning was successfully applied to predict pollutant dispersion.

• Several Deep Learning models and loss were tested and compared.

• The custom loss J3D outperformed binary cross entropy and mean squared error.

• MultiResUnet is the best model for pollutant dispersion prediction.

摘要

•Methods to convert CFD inputs for neural networks are presented.•Deep Learning was successfully applied to predict pollutant dispersion.•Several Deep Learning models and loss were tested and compared.•The custom loss J3D outperformed binary cross entropy and mean squared error.•MultiResUnet is the best model for pollutant dispersion prediction.

论文关键词：Deep learning,Convolutional neural network,Computational Fluid Dynamics,Air quality

论文评审过程：Received 3 August 2021, Revised 20 January 2022, Accepted 22 April 2022, Available online 10 May 2022, Version of Record 14 May 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117294