Research on air quality prediction based on improved long short-term memory network algorithm.

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Air quality is significantly impacted by industrial, agricultural, and human activities.
• Traditional machine learning models struggle with time-series data and long-range dependencies, limiting air quality prediction accuracy.
• Existing methods often overlook crucial temporal information, leading to suboptimal forecasting.

Purpose of the Study:

• To develop a superior air quality prediction model that addresses limitations of traditional methods.
• To enhance prediction accuracy by incorporating time-series analysis and attention mechanisms.
• To validate the effectiveness of the proposed integrated model against existing approaches.

Main Methods:

• An attention mechanism was integrated into a Long Short-Term Memory (LSTM) network to manage information weighting.
• A hybrid model combining the Light Gradient Boosting Machine (lightGBM) with the LSTM-attention model was constructed.
• The performance of the integrated lightGBM+LSTM-attention model was evaluated against 11 other models.

Main Results:

• The integrated lightGBM+LSTM-attention model demonstrated superior performance in air quality prediction.
• The model achieved a coefficient of determination (R2) of 0.969 for prediction accuracy.
• Root Mean Square Error (RMSE) was improved by 5.09, 4.94, 4.85, and 4.0 compared to other benchmark models.

Conclusions:

• The developed lightGBM+LSTM-attention model significantly enhances air quality forecasting accuracy.
• The integration of attention mechanisms and hybrid modeling effectively captures complex temporal patterns.
• The proposed model offers a more reliable solution for monitoring and predicting air quality changes.