Predicting non-uniform indoor air quality distribution by using pulsating air supply and SVM model

Abstract

Mixing ventilation is the most common air distribution strategy, and often the same diffusers provide space cooling and heating. Although aiming to achieve uniform air distributions, there are still non-uniformities at specific indoor locations. This study innovatively predicts the non-uniform indoor air quality (IAQ) distribution with mixing ventilation, directly from air temperature and air velocity. Both heating and cooling cases are conducted with computational fluid dynamics (CFD) techniques, validated by the experimental measurements in a multi-occupant office configuration. The comparison shows that the support vector machine (SVM) model outperforms the back-propagation neural network (BPNN) and genetic algorithm back-propagation neural network (GABPNN) models, with a medium computation time. The innovative method is demonstrated that the pulsating air supply is useful in generating effective inputs, i.e., the dynamic variations of air temperature and velocity during pulsating air supply (ΔT and Δv), for the air age prediction under the corresponding steady state (τsteady). ΔT and Δv can be fast obtained by monitoring the air velocity and temperature for 10–20 min. With a random selection, the data size should be larger than 180 to reach the mean absolute percentage error (MAPE) threshold value of 5%. However, it can also be reduced to 60 when the data points have a similar steady air temperature. The prediction accuracy under heating is slightly higher than that under cooling, as achieving good mixing under heating is more difficult. We aim to provide guidelines on effective and measurable inputs and valid data-driven models for non-uniform IAQ prediction.