Predicting PM2.5 concentration with enhanced state-trend awareness and uncertainty analysis using bagging and LSTM...

Area of Science:

• Environmental Science
• Data Science
• Artificial Intelligence

Background:

• Global environmental protection efforts increasingly focus on monitoring fine particulate matter (PM2.5).
• Accurate prediction of PM2.5 concentrations is crucial for environmental management and public health.
• Traditional methods often struggle with the dynamic nature and complexities of air pollution data.

Purpose of the Study:

• To apply the state-trend awareness concept from big data analytics to PM2.5 prediction.
• To develop an advanced model for more accurate and reliable forecasting of PM2.5 concentrations.
• To incorporate interval prediction for quantifying forecasting uncertainty.

Main Methods:

• Utilized state-trend awareness for holistic data insights and decision support.
• Developed a hybrid model combining Long Short-Term Memory (LSTM) neural networks with a bagging ensemble algorithm.
• Implemented interval prediction to forecast probability ranges of PM2.5 variations.

Main Results:

• The proposed LSTM-bagging model significantly outperformed traditional LSTM and Support Vector Machine (SVM) methods in accuracy and generalization.
• Achieved a 12% reduction in errors compared to SVM-LSTM models.
• Demonstrated improved prediction accuracy, enhanced model generalization, and reduced overfitting.

Conclusions:

• The developed state-trend aware LSTM-bagging model with interval prediction offers a robust tool for environmental monitoring.
• This approach enhances decision-making capabilities for public health and environmental protection.
• The method effectively addresses forecasting uncertainties by providing probability ranges for PM2.5 variations.