Air quality index forecast in Beijing based on CNN-LSTM multi-model

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Accurate air quality trend prediction is crucial for environmental protection and decision-making.
• Existing time series models have limitations in capturing complex air quality dynamics.

Purpose of the Study:

• To propose and evaluate a novel Convolutional Neural Networks-Long Short-Term Memory (CNN-LSTM) model for enhanced air quality prediction.
• To compare the performance of the CNN-LSTM model against several established prediction models.

Main Methods:

• Utilized Convolutional Neural Networks (CNN) for efficient feature extraction from air quality data.
• Employed Long Short-Term Memory (LSTM) networks to learn temporal dependencies and predict future air quality.
• Compared the CNN-LSTM model with ARMA, SARIMA, RNN, LSTM, and GRU models using Beijing's air quality index data.

Main Results:

• The CNN-LSTM model demonstrated superior prediction accuracy compared to all single prediction models evaluated.
• Specifically, CNN-LSTM showed significant improvements over the SARIMA model, a representative time series model.
• Mean Absolute Error (MAE) decreased by 3.17%, Root Mean Square Error (RMSE) by 5.46%, and R-squared (R²) improved by 8.45%.

Conclusions:

• The proposed CNN-LSTM model offers a significant advancement in air quality prediction accuracy.
• This hybrid approach effectively combines feature extraction and temporal learning for robust forecasting.
• The findings support the application of CNN-LSTM for more effective environmental monitoring and management.