A novel encoder-decoder model based on Autoformer for air quality index prediction.

Area of Science:

• Environmental Science
• Computer Science
• Data Science

Background:

• Economic development exacerbates air quality issues, necessitating accurate prediction for effective pollution control.
• Existing air quality prediction models face challenges in handling complex temporal dependencies and long-term forecasting.

Purpose of the Study:

• To develop a novel encoder-decoder model, Enhanced Autoformer (EnAutoformer), for improved air quality index (AQI) prediction.
• To enhance the extraction of temporal dependencies and computational efficiency in AQI time series analysis.

Main Methods:

• Proposed Enhanced Autoformer (EnAutoformer) model incorporating enhanced cross-correlation (ECC) for temporal dependency extraction.
• Developed a core module CSP_ECC by combining ECC with CSPDenseNet for improved computational efficiency.
• Integrated time series decomposition and dilated causal convolution in the decoder for finer-grained feature extraction and long-term prediction.

Main Results:

• The EnAutoformer model demonstrated significantly improved prediction accuracy compared to baseline models.
• Validation on real-world air quality datasets from Lanzhou confirmed the model's effectiveness.
• The model successfully extracts finer-grained features, enhancing long-term AQI prediction capabilities.

Conclusions:

• EnAutoformer provides a promising and accurate alternative for complex air quality index prediction.
• The model's enhancements in temporal dependency extraction and computational efficiency contribute to its superior performance.
• Accurate AQI prediction using EnAutoformer can support crucial air pollution prevention and treatment strategies.