Urban AQI Prediction Using an Inverse Variance Weighted GCN and BiLSTM Combined Model

Forecasting air quality is very important for the development of effective environmental management policies. We propose a model that marries the strengths of a Graph Convolutional Network and a Bidirectional Long Short-Term Memory Network. This model, further refined with inverse variance weight allocation, is designed to accurately delineate the spatiotemporal characteristics inherent in air pollution monitoring networks. This model takes into account factors such as geographical location, multiple pollutant types, and weather to achieve accurate prediction of urban AQI. First, we construct a topological structure graph that reflects the layout of urban air pollutant monitoring stations. This allows GCN to effectively capture the spatial characteristics. Then, considering that the historical data of pollutants is a nonlinear multivariate long time series, we utilize BiLSTM to effectively capture the temporal features. Finally, we leverage inverse variance to determine the fusion weights of GCN and BiLSTM. This results in the generation of spatiotemporal fusion prediction outcomes. Evaluation of the proposed model using air quality data from Chongqing shows that it has a higher level of prediction accuracy compared to other models.