An attention temporal convolutional network-based hybrid approach to simulating indoor air pollutants and their determinants in classroom and office spaces

Accurately assessing and source tracing indoor air quality (IAQ) is crucial for implementing targeted interventions to improve IAQ. This study aimed to investigate the concentrations and underlying determinants of multiple indoor air pollutants in classrooms and office spaces of educational buildings and develop a robust model for precise IAQ assessment and simulation. A walkthrough-based inspection with continuous monitoring of indoor and outdoor PM2.5, PM10, NO2, and O3 was conducted in ten educational buildings in Gainesville, Florida. IAQ dynamics and determinants were systematically compared across spaces to discern setting-specific patterns. An Evolutionary Polynomial Regression method-assisted Attention Temporal Convolution Network (EPR-ATCN) model was developed to assess and predict IAQ. The results revealed that the influencing factors of indoor pollutants in classrooms and offices were similar, but the contribution proportion and weight of each factor differed. Street distance, wall damage, indoor relative humidity, and air conditioning vent size were identified as key determinants for most indoor pollutants. The EPR-ATCN models significantly outperformed Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) ANN-based models in simulating indoor environments of both classrooms (RMSE: 1.6, 2.4 and 4.9) and offices (RMSE: 1.3, 2.9 and 5.1). Follow-up studies exploring IAQ in similar architectural and environmental settings may accordingly reference these findings. •Factors influence the same pollutant differently in classrooms and offices.•Gaseous pollutants vary more than particulate pollutants in both spaces.•The EPR-ATCN model can efficiently process complex indoor air quality data.•The EPR-ATCN outperforms standard ANNs in nonlinearity and collinearity issues.