Quantifying indoor infection risk based on a metric-driven approach and machine learning

To quantify the risk of infection in a typical office space, this study proposes a new ‘probability of infection’ metric which accounts for the particle number, accumulated viral load, and relevant clinical data. This is achieved by utilising computational fluid dynamics (CFD) and the Eulerian-Lagrangian model. Additionally, the simulations employ models for the exhaled CO2 and the age of air (a function of the ventilation effectiveness) to understand how the flow field influences the transport of both airborne infectious particles and CO2. The distribution of the CO2 concentration in the room shares similarities to that of airborne infectious particles, but there are differences, with ‘smearing’ from high to low concentrations observed. This study demonstrates that the new metric should be used to quantify and assess the risk of cross-infection instead of the widely used CO2 concentration. The crucial parameter in reducing indoor virus transmission and CO2 levels was found to be the ventilation effectiveness, which is dependent on the ventilation system design, also influencing the amount of fresh air required to lower both quantities. Through the optimised random forests regression model, the CO2 concentration and the supply ventilation rate were utilised as inputs to predict the quantifiable infection risk. The model prediction boasted a coefficient of determination of 0.99 and a root mean square a of 0.025. Thus, a computational framework is established for the development of intelligent building systems with CO2 sensors that can automatically counter airborne infection risk by adaptively varying the ventilation rate