Air Pollution Forecast and The Air Quality Changes in The Beijing-Tianjin-Hebei Region

Air pollution control and its health impacts are critical concerns. Traditional air pollution prediction often focuses on the Air Quality Index (AQI), which represents the most harmful pollutant but neglects the cumulative effects of others, limiting its accuracy in assessing hazards. To address this, a neural network model is proposed to simultaneously predict three indices: AQI, the Aggregate Air Quality Index (AAQI), and the Health Risk-Based Air Quality Index (HAQI). The model integrates Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP), and Transformer, selecting the best prediction among them to enhance performance and stability. Its effectiveness was validated using data from 12 monitoring stations in Beijing.
In addition, a GCN+LSTM model was developed for AQI prediction, leveraging Graph Convolutional Networks (GCN) to extract spatial correlations from historical data and integrating them with LSTM’s temporal modelling capabilities. This approach simplifies spatial data collection, improves prediction accuracy, and outperforms other models, including Long Short-Term Memory, LSTM+Attention, and Gated Recurrent Unit (GRU), in experimental evaluations.
Furthermore, a comprehensive analysis of air quality trends in the Beijing-Tianjin-Hebei region from 2014 to 2023 was conducted. This analysis examined the effects of strict pollution controls, COVID-19, seasonal variations, and associated health risks, providing insights into regional and temporal patterns.
This study contributes by proposing a multi-index prediction framework for air quality assessment, introducing an innovative spatial-temporal model, and offering a decade-long analysis of air pollution trends and their health impacts in the Beijing-Tianjin-Hebei region.