Predicting the concentrations of VOCs in a controlled chamber and an occupied classroom via a deep learning approach

The ability to predict indoor pollutant concentrations is an indispensable function for smart homes. In this study, we present a Long Short-Term Memory network (LSTM) model in the deep learning field, to predict the concentrations of volatile organic compounds (VOCs) in different indoor settings, and a mean absolute percentage error (MAPE) is used as a metric to evaluate the performance of the LSTM model. The selection of some key parameters on the LSTM model prediction is firstly discussed. We then analyze the concentrations of different VOCs emitted from three kinds of furniture in a controlled chamber, and concentrations of 6-methyl-5-hepten-2-one (6-MHO) and 4-oxopentanal (4-OPA) due to ozone/squalene reactions, in an occupied classroom. The model's predictions for the VOCs in the chamber tests, have all MAPE within 10%; for ozone and 6-MHO in the classroom tests, 85% of the MAPE is within 15%; and for 4-OPA, 82% of the MAPE is within 15%. The small MAPE indicates good performance. Comparison analysis reveals that the LSTM model is superior to the widely used artificial neural network (ANN) model. The LSTM approach doesn't require building complex physical or chemical models and measuring various key parameters, instead a learning network is established and the learning parameters are adjusted according to practical situations. This study demonstrates that the LSTM model is promising for the prediction of pollutant transport in various indoor environments.