An optimized hybrid deep learning model for PM 2.5 and O 3 concentration prediction

As people focus more on environmental protection, air quality prediction plays an increasingly important role in reducing pollution hazards. Both fine particulate matter (PM 2.5 ) and ozone (O 3 ) pollutants can cause serious damage to human health and property, so it is necessary to accurately predict the concentration of these pollutants. In this study, a hybrid deep air quality prediction model consisting of a one-dimensional convolutional neural network (CNN), bidirectional long-term and short-term memory (BiLSTM), and a gated recurrent unit (GRU) is proposed to predict air quality pollutant concentrations. This model overcomes the limitations of a single model while taking advantages of its benefits. The BiLSTM neural network has more parameters and poor convergence performance, and the GRU has a poor ability to capture long-distance dependencies between features. Compared with the other three deep learning models, the CNN-BiLSTM-GRU model achieves better prediction results. The model proposed in this paper with both meteorological factors and pollutant factors shows the best prediction results with an R 2 of 0.956 and RMSE of 17.2 μg/m 3 for PM 2.5 and an R 2 of 0.958 and RMSE of 13.43 μg/m 3 for O 3 . The original data set from the Aotizhongxin Observator of Beijing with 35,064 samples is selected as the experimental data. The experimental results show that the CNN-BiLSTM-GRU model proposed in this paper achieves the best prediction results. The results show that the proposed model can predict PM 2.5 and O 3 more accurately and more robustly, which indicates that it is a promising method for air and particulate pollutants’ performance prediction.