STGATN: A novel spatiotemporal graph attention network for predicting pollutant concentrations at multiple stations

Area of Science:

• Environmental Science
• Data Science
• Computer Science

Background:

• Accurate air pollutant concentration prediction is crucial for public health and environmental governance.
• Existing models face challenges in capturing dynamic spatiotemporal dependencies and avoiding error propagation.

Purpose of the Study:

• To propose a novel spatiotemporal graph attention network (STGATN) for enhanced air pollutant prediction.
• To address the limitations of existing methods in handling dynamic dependencies and long-term prediction errors.

Main Methods:

• Developed STGATN, an encoder-decoder architecture incorporating spatiotemporal embedding, graph attention, and gated temporal convolutional networks.
• Integrated a fusion gate for adaptive merging of spatiotemporal and temporal features.
• Introduced transformer attention to prevent prediction error accumulation during decoding.

Main Results:

• The proposed STGATN model demonstrated superior performance compared to state-of-the-art baseline methods.
• Experimental results validated the model's effectiveness on an air pollution dataset from Beijing.

Conclusions:

• STGATN offers a significant advancement in air pollutant concentration prediction.
• The model's ability to capture complex spatiotemporal dynamics and mitigate errors provides a valuable tool for environmental monitoring.