Deep Spatio-Temporal Graph Network with Self-Optimization for Air Quality Prediction.

Area of Science:

• Environmental Science
• Data Science
• Computer Science

Background:

• Environmental pollution poses significant risks, necessitating accurate pollutant prediction.
• Existing time series models often neglect spatial transmission effects, limiting prediction accuracy.
• Developing advanced models is crucial for effective environmental protection strategies.

Purpose of the Study:

• To propose a self-optimizing spatio-temporal graph neural network (BGGRU) for enhanced air pollutant prediction.
• To integrate spatial and temporal data analysis for improved accuracy.
• To address the challenge of suboptimal model performance due to hyperparameter selection.

Main Methods:

• Utilized a graph sampling and aggregation network (GraphSAGE) for spatial information extraction.
• Employed a Bayesian graph gated recurrent unit (BGraphGRU) to model temporal dynamics.
• Implemented Bayesian optimization for automatic hyperparameter tuning.

Main Results:

• The proposed BGGRU network effectively captures both temporal evolution and spatial propagation of pollutants.
• The model demonstrated high accuracy in predicting PM2.5 concentrations using real-world data from Beijing.
• Bayesian optimization successfully mitigated model inaccuracy caused by suboptimal hyperparameters.

Conclusions:

• The BGGRU model offers an effective and accurate method for air pollutant concentration prediction.
• Integrating spatio-temporal analysis significantly enhances prediction performance compared to traditional methods.
• This approach provides a valuable tool for environmental monitoring and pollution control efforts.