Air Quality Prediction Based on Graph Attention Network

Air quality is related to air pollution, weather, and exhaust emissions. Several previous studies have used temporal data for air quality prediction. However, the air is diffuse, and the air between sites will affect each other. Therefore, we found that more than temporal data is needed to predict air quality accurately. Spatial data between sites should also be considered. An air quality prediction model based on graph attention network is proposed in this paper. The spatial data are used for the graph attention network to obtain the influence of adjacent sites. The spatial data between non-adjacent sites is obtained by diffusion convolution. The prediction is then completed by the gate recurrent unit. Experiments show that the model fully uses spatial data and accurately predicts air quality.