Data-driven prediction model of indoor air quality by the preprocessed recurrent neural networks

In this study, data-driven prediction methods based on recurrent neural networks (RNN) for indoor air quality in a subway station are developed. The RNN can predict the air pollutant concentration of PM10 and PM_2.5 at a platform of a subway station by using the previous information of NO, NO₂, NO_X, CO, CO₂, temperature, humidity, and PM₁₀ and PM_2.5 on yesterday. For comparison, the other prediction models such as neural networks (NN) and multiple regression model are used. To optimize the prediction model, the variable importance in the projection (VIP) of the PLS is used to select key input variables as a preprocessing step. Experimental result shows that the selected key variables have positive influence on the prediction performance. The predicted result of RNN model gives better modeling performance and higher interpretability than other data-driven prediction modeling methods.