An Urban Air Quality Prediction Model based on Dynamic Correlation of Influencing Factors

Urban air quality prediction models can predict pollutant values based on its time series. Existing research shows that the correlation between influencing factors is dynamic. In this paper, we propose an Urban Air Quality Prediction Model based on Dynamic Correlation of Influencing Factors (DynamicAir) to address this problem. In the dynamic correlation module, the dynamic correlation of influencing factors is captured by dynamic graph generation and dynamic graph convolution; in the multi-time-step prediction module, the time correlation of each step and the dynamic correlation of influencing factors are mapped by multi-layer non-linear mapping to obtain the future pollutant concentration values at multi-steps. Experimental results on two real datasets(Beijing Capital International Airport and Beijing Olympic Sports Centre) show that the proposed DynamicAir reduces the RMSE by 1.15% and 4.04% respectively compared to the state-of-the-art baseline model (with a statistical interval of three hours).