A predictive model of indoor PM2.5 considering occupancy level in a hospital outpatient hall

The hospital outpatient hall is more complex and sensitive than other indoor places because of its high density, flow of patients, and risk of infection. The prediction of indoor pollutants, such as PM2.5, is a critical health risk factor and an important topic in the study of indoor air quality. Numerous black-box models have been built to predict PM2.5, which are prone to overfitting and low precision in long sequence time prediction due to their limited weighting calculation and factors considered In this study, subject-object weighting incorporates a long sequence time-series model that considers occupancy (SO-LSTS) to predict PM2.5 concentrations in a hospital outpatient hall. First, the occupancy level was obtained using image recognition technology. Second, both the subjective (improved AHP) and objective (entropy weight) information were coupled by a distance function and then integrated into the LSTS model. Finally, the model performance was compared to six traditional models and the impact on the output length and hyper-parameter confirmation was assessed. The results demonstrate that the occupancy factor can improve the model performance by 54 %, and the model accuracy is improved by 89 % compared to the traditional Informer method. Our study considers real-time environmental and occupancy levels, which can compensate for the difficulty of interpreting the black-box model and identifying an accurate and resource-efficient proactive control model for hospital environmental management compared to conventional approaches.