Deep learning for 3D reconstruction and trajectory prediction of dust and polluted aerosols in educational environments

Introduction The accurate reconstruction and prediction of dust and polluted aerosol trajectories in educational environments are critical for assessing air quality and mitigating health risks. Traditional numerical models for aerosol transport rely on Eulerian or Lagrangian approaches, which often suffer from trade-offs between computational efficiency and physical accuracy. Eulerian models struggle with resolving small-scale turbulence, while Lagrangian tracking methods face challenges in capturing multiscale interactions effectively. Methods To address these limitations, we propose a deep learning-driven approach that integrates a hybrid Eulerian-Lagrangian computational model with machine learning-enhanced optimization. Our method employs a high-fidelity aerosol transport model incorporating stochastic corrections for sub-grid scale effects and adaptive meshing for efficient resolution of dynamic aerosol distributions. We introduce a data-driven optimization framework that leverages physics-informed neural networks to enhance predictive accuracy while reducing computational overhead. Results and Discussion Experimental validation demonstrates that our approach significantly outperforms conventional numerical methods in both accuracy and efficiency, making it highly suitable for real-time applications in educational environments. This study provides an innovative and scalable solution for understanding and mitigating aerosol dispersion in indoor spaces, contributing to improved air quality management and public health protection.