Rapid Prediction of Local Mean Age of Air for Energy-Efficient Ventilation Systems Using Permutation Feature Importance

Abstract

Prediction of local mean age of air (MAA) is a key technology that can enhance the comfort, health, and productivity of indoor residents by adjusting and optimizing the indoor environmental conditions. In this study, we developed a deep neural network (DNN)-based regression model to predict indoor air quality (IAQ) and proposed a permutation feature importance (PFI)-based explainable artificial intelligence (XAI) model to implement efficient ventilation systems in a hospital ward utilizing this regression model. The rapid prediction of the MAA in the space near each patient in the ward, depending on the location of the heating, ventilation, and air conditioning (HVAC) inlets and fluid velocity, were successfully measured through data-driven deep learning model training. Consequently, the proposed MAA prediction model achieved average R-squared values of 0.9506 and 0.9220 for MAA(1) and MAA(2), respectively. In addition, the DNN model demonstrated rapid predictive performance (similar to 0.4 ms/prediction), highlighting the possibility of real-time monitoring compared to conventional methods. Furthermore, the contribution of the location and fluid velocity of the HVAC system to the MAA in the space near the patient was analyzed using PFI. These results support the rapid virtual sensing and recommendation method that has the potential to be applied in future IAQ management, human healthcare, and energy management systems.