Predicting the spatio-temporal distribution of the droplets based on the machine learning algorithm

Aerosol pollutants composed of suspended droplets significantly impact environmental quality and human health. Predicting the spatiotemporal distribution of cough droplets remains a challenge due to their complex multiphase dynamics, involving intricate interactions between droplet motion and turbulent airflow. This study presents a three-dimensional Gaussian parameter model integrating computational fluid dynamics (CFD) with machine learning to efficiently simulate and predict the transport and dispersion of indoor cough droplets. The Gaussian model derived from CFD flow field dynamics and droplet kinematics adheres to conservation principles and hyperbolicity, ensuring physical consistency. An adaptive polynomial feature random forest algorithm predicts model parameters, enabling rapid reconstruction of droplet trajectories and spatial distribution patterns. The approach achieves a 76.4% reduction in computational cost compared to traditional CFD simulations while maintaining high accuracy, with a mean absolute error below 0.07 and a mean squared error below 0.014. This robust and versatile framework advances the understanding of aerosol transport dynamics, offering critical insight and practical tools for indoor air quality management and aerosol pollution control.