Expiratory aerosols' spread, removing and infection risk investigation in public toilet: CFD and machine learning

Abstract

Public toilets are locations to shelter and spread viruses because of the high traffic flow of people. Exposure to virus-carrying aerosol particles is a major way of spreading respiratory disease. This paper focuses on the distribution of human-exhaled aerosols in the toilet. The distribution and temporal variations of aerosol particles within toilet cubicles were investigated through numerical simulation. The percentage of suspended, deposited, and escaped particles was 28.43 %, 21.06 %, and 50.51 % under top evacuation. Higher volume ventilation and different ventilation modes were conducted to explore the efficient removal method. An infection risk assessment based on the Wells-Riley model was conducted and the back-side ventilation had the best performance. Random Forest, Light Gradient Boosting Machine (LightGBM), and Multilayer Perceptron (MLP) models are employed to reveal the connection between particle size, density, time, and suspension rate. Residual analysis is utilized to determine the best machine learning model. The Random Forest model has the best performance, with the residual obeying normal distribution. Its Mean Absolute Error (MAE) and Mean Squared Error (MSE) are the lowest, at 0.0249 and 0.0011. Finally, the SHapley Additive exPlanations (SHAP) value explores the effects of particle size and density on the spread of aerosols. With the study's results, more efficient ventilation or disinfection techniques can be adopted to lower the risk of infection and prevent the spread of respiratory disorders.