Predicting long-term aldehyde concentrations in an occupied residence via multi-factor deep learning model

(Times New Roman Bold 12 pt, UPPERCASE, style: Heading 1) Aldehydes are common and harmful volatile organic compounds (VOCs) in the indoor environment. In this study, three multi-factor deep learning models (LSTM, RNN, GRU) are used to predict the concentration of aldehydes in an occupied residence. RD, MAE, RMSE, and MAPE are selected as evaluation metrics. We firstly introduce the measured time-resolved dataset. Then, we find that the correlation coefficient between temperature (T), relative humidity (RH) and air change rate (ACR) are outside -0.6~0.6, implying no feature redundancy relationship. ACR has the greatest impact on aldehydes concentration. The concentrations of formaldehyde, acetaldehyde, furfural and acrolein are predicted by the three deep learning models over ten days, and results from the evaluation metrics indicate that LSTM model has the best prediction performance and environmental applicability in the occupied residence. Our study illustrates that LSTM model is a promising application in the indoor aldehyde transport prediction.