Air Quality Prediction Based on Dynamic Graph Convolutional Network with Spatiotemporal Attention Mechanism

Air pollution is one of the serious problems facing the world today, which not only leads to a significant decrease in atmospheric visibility in some areas, but also poses a great threat to human health. In recent years, although some existing studies have modeled the correlation and dependence of data from each monitoring station in both spatial and temporal dimensions, most of the studies have only focused on the static geographic locations of the stations without fully considering the dynamic spatial correlation between the stations when extracting the spatial information between neighboring stations. In this paper, we propose a dynamic graph convolutional neural network model (STA-DGN) based on the spatio-temporal attention mechanism, which encodes the spatial location information of the monitoring stations by combining the static topology between the stations and the pollutant concentration data, and by dynamically taking into account the interactions between the stations. In addition, in the time dimension, the historical information of pollutant data will be mined by combining the gated recurrent unit (GRU) with the temporal attention mechanism, which fully considers the influence weight of historical time information on the current time. Experimental results on real-world air quality datasets show that our method can effectively capture the dynamic spatio-temporal correlation of pollutant data and accurately predict air quality.