Related Experiment Videos
Forecasting air pollutant concentration using a novel spatiotemporal deep learning model based on clustering, feature
Jusong Kim1, Xiaoli Wang2, Chollyong Kang3
1Tianjin Key Laboratory of Hazardous Waste Safety Disposal and Recycling Technology, School of Environmental Science and Safety Engineering, Tianjin University of Technology, Tianjin 300384, China; Department of Mathematics, University of Science, Pyongyang 999091, DPR Korea.
The Science of the Total Environment
|August 20, 2021
Summary
This study introduces a novel hybrid model for accurate air pollutant concentration forecasting. The model significantly improves predictions for particulate matter (PM2.5), offering a powerful tool for early warning systems.
Area of Science:
- Environmental Science
- Data Science
- Atmospheric Chemistry
Background:
- Accurate air pollutant forecasting is crucial for early warning systems but challenged by limited emission data and process uncertainties.
- Existing methods struggle with the complexities of dynamic atmospheric processes and data limitations.
Purpose of the Study:
- To develop and validate a novel hybrid model for enhancing the accuracy of air pollutant concentration forecasts.
- To address the challenges of limited emission source information and high uncertainties in dynamic processes.
Main Methods:
- A hybrid model combining clustering, feature selection, empirical wavelet transform (EWT) for real-time decomposition, and deep learning (3D CNN-BiLSTM).
- Time series decomposition using EWT, data subset construction, clustering, feature selection, and deep learning prediction.
- Reconstruction of predicted decomposition components to obtain final air pollutant concentration forecasts.
Main Results:
- The proposed hybrid model demonstrated superior performance compared to other models in predicting PM2.5 concentrations.
- Achieved mean absolute percentage errors of 4.03% (1h), 6.87% (6h), and 8.98% (10h) for PM2.5 forecasts.
- Validated using PM2.5 data from Beijing, China, confirming its effectiveness.
Conclusions:
- The developed hybrid model is a powerful and accurate tool for forecasting air pollutant concentrations.
- The integration of EWT, clustering, feature selection, and deep learning effectively handles data complexities and uncertainties.
- The model shows significant potential for improving air quality early warning systems.
