A study on air quality prediction with multiple features based on GCN-LSTM

Abstract Air pollution is closely intertwined with public health. However, air quality forecasting is a very challenging project, and the influencing factors are intertwined. This study focuses on analyzing and forecasting urban air quality from a spatiotemporal perspective, based on various factors that affect air quality. In this paper, a diverse dataset impacting air quality was collected, including pollutant concentrations, meteorological conditions, and traffic flow data. A GCN-LSTM model was developed to thoroughly explore the spatial relationships between different sites within the study area. This paper also conducted forecasts over different time lengths. The experiments revealed that weather data and traffic data can help improve air quality prediction results. The forecasting accuracy for RMSE and MAE were improved by an average of 8.7% and 8.5% in PM 2.5 , and 12.8% and 11.9% in NO 2 , respectively. Within the model, the GCN-LSTM model, which considers spatiotemporal features, outperformed single models and proved to be more advantageous in long-term forecasting. The findings of this study contribute to the improvement of air quality prediction effectiveness.