Research on PM2.5 concentration prediction algorithm based on graph convolutional neural network model

Air quality prediction is a crucial environmental science issue, with PM2.5 posing a serious threat to human health. Accurate forecasting of PM2.5 is essential for implementing effective pollution control measures and assisting the public in planning their activities sensibly. However, many current PM2.5 prediction models often focus solely on forecasting for individual monitoring stations, neglecting the impact of pollutants around these stations. Moreover, traffic emissions constitute a significant source of air pollutants, and excluding traffic information in PM2.5 prediction models can lead to unrealistic predictions. Therefore, this paper proposes a novel deep learning model (GCN-GT-ATT) to predict future PM2.5 concentrations in Zhengzhou city. The model incorporates traffic emission factors directly affecting PM2.5 and utilizes graph convolutional neural networks to consider the mutual influences among adjacent monitoring stations due to factors such as pollutants, meteorological conditions, and traffic emissions. After forming the time-series characteristics for the target station, the data is fed into GRU (Gated Recurrent Unit) and TCN (Temporal Convolutional Network) models. Attention mechanisms are then applied separately to GRU and TCN networks, enabling them to capture more valuable information. Experimental results demonstrate that our proposed model outperforms other comparative models in terms of prediction accuracy for the selected three stations.