Air Quality Index Prediction Model Integrating Spatiotemporal Information

Aiming at the problem that traditional air quality prediction research ignores spatial information, this paper proposes an air quality index prediction model (SGCN-GRU) that integrates spatiotemporal information based on gated convolution (Gated-Conv), graph convolution (GCN), and gated recurrent unit (GRU). The model first calculates the geodesic distance between each other based on the longitude and latitude coordinates of each monitoring station in the input data and establishes a graph structure. The reciprocal of the geodesic distance is used as the edge weight in the graph structure, and then the data is input into the two-layer Gated-Conv and a layer of Cheb Graph-Conv is used to obtain spatial information in a spatial graph convolution block composed of a “sandwich” structure, and then input the obtained feature map into GRU to capture the long-term dependencies existing in it, and finally obtain it through the output layer final prediction result. Experimental results show that SGCN-GRU achieves an effect of 14.67% on MAPE in one-hour AQI prediction. Compared with other benchmark models, SGCN-GRU performs best in multi-hour predictions.