Air quality forecasting using a spatiotemporal hybrid deep learning model based on VMD-GAT-BiLSTM

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Precise air quality forecasting is crucial for early warning systems but faces challenges due to limited emission data and dynamic process uncertainties.
• Existing methods struggle with the complexity of spatiotemporal air quality data.

Purpose of the Study:

• To develop an advanced spatiotemporal hybrid deep learning model for improved air quality forecasting.
• To enhance the accuracy and reliability of air quality prediction models.

Main Methods:

• A novel VMD-GAT-BiLSTM model was proposed, integrating Variational Mode Decomposition (VMD), Graph Attention Networks (GAT), and Bi-directional Long Short-Term Memory (BiLSTM).
• VMD decomposes PM2.5 data into stable sub-sequences to mitigate uncertainties.
• GAT captures spatial dependencies between monitoring stations, while BiLSTM learns temporal features within sub-sequences.

Main Results:

• The VMD-GAT-BiLSTM model demonstrated superior performance compared to other methods in air quality forecasting tasks.
• The model achieved high accuracy in both short-term and long-term air quality predictions on a Beijing dataset.

Conclusions:

• The proposed VMD-GAT-BiLSTM model effectively addresses the challenges in air quality forecasting by integrating VMD, GAT, and BiLSTM.
• This hybrid deep learning approach offers a promising solution for accurate and reliable air quality prediction.