Fast flow field prediction of pollutant leakage diffusion based on deep learning

Area of Science:

• Environmental Science
• Computational Fluid Dynamics
• Artificial Intelligence

Background:

• Accurate prediction of pollutant leakage and diffusion is vital for public safety.
• Deep learning (DL) offers efficiency but lacks application in pollutant diffusion flow field prediction.
• Traditional computational fluid dynamics (CFD) methods are computationally intensive and grid-dependent.

Purpose of the Study:

• To develop and evaluate a deep learning method for modeling pollutant diffusion flow characteristics.
• To improve the accuracy and computational efficiency of pollutant diffusion prediction.
• To explore the potential of DL as a supplementary tool for CFD.

Main Methods:

• A novel two-level deep learning network architecture was developed.
• The network models the complex flow characteristics of pollutant diffusion.
• Performance was benchmarked against traditional DL methods and CFD.

Main Results:

• The proposed DL method significantly improved flow field prediction accuracy.
• Computational efficiency was enhanced by over 800 times compared to traditional CFD.
• The DL method is not limited by grid expansion requirements inherent in CFD.

Conclusions:

• Deep learning methods show great promise for pollutant leakage and diffusion prediction.
• This approach can serve as a valuable supplement to existing CFD methods.
• The findings provide a foundation for integrating DL into environmental safety applications.