A novel deep learning model integrating CNN and GRU to predict particulate matter concentrations

Abstract

PM2.5 is a significant environmental pollutant that damages the environment and endangers human health. Precise forecast of PM2.5 concentrations is very important to control air pollution and improve people’s life quality. In the subway indoor air quality (IAQ) system, the data collected by telemonitoring systems is frequently lost due to many reasons. A deep learning model called RF-CNN-GRU, which combines random forest (RF), convolutional neural network (CNN) and gated recurrent unit (GRU), is proposed to predict atmospheric PM2.5 concentrations with incomplete original data. The RF-CNN-GRU model employs the RF to fill in missing values in the data and subsequently applies the CNN to extract features from the imputed data. The data is finally sent to the GRU network to train and predict PM2.5 concentrations. Comparing with single CNN, GRU and long short-term memory (LSTM) models, the predictive accuracy of the RF-CNN-GRU model is significantly improved. The RF-CNN-GRU model shows a slight improvement in prediction results when compared to models such as CNN-GRU, RF-CNN, RF-GRU, and RF-LSTM. The findings demonstrate that the RF-CNN-GRU model has excellent accuracy in PM2.5 concentration prediction when the original data is incomplete.