Related Experiment Videos
Air quality prediction at new stations using spatially transferred bi-directional long short-term memory network
Jun Ma1, Zheng Li2, Jack C P Cheng1
1Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong Kong, China.
The Science of the Total Environment
|January 25, 2020
Summary
This study introduces a transfer learning-based stacked bidirectional long short term memory (TLS-BLSTM) network to improve air quality forecasting for new monitoring stations with limited data. The method effectively transfers knowledge from established stations, significantly reducing prediction errors.
Area of Science:
- Environmental Science
- Data Science
- Computer Science
Background:
- Air pollution is a critical environmental issue, particularly in developing nations.
- Accurate air quality forecasting is vital for effective pollution control and mitigation strategies.
- Existing deep learning models for air quality prediction require substantial data, limiting their application in data-scarce environments like new monitoring stations.
Purpose of the Study:
- To develop a novel methodology addressing the data shortage challenge for air quality prediction at new monitoring stations.
- To leverage transfer learning and deep learning to enhance forecasting accuracy for stations with limited historical data.
Main Methods:
- Proposed a transfer learning-based stacked bidirectional long short term memory (TLS-BLSTM) network.
- Integrated advanced deep learning techniques with transfer learning strategies.
- Applied the TLS-BLSTM model to a case study in Anhui, China, for air quality forecasting.
Main Results:
- The TLS-BLSTM network demonstrated effectiveness in transferring knowledge from existing stations to new ones.
- Achieved an average reduction of 35.21% in Root Mean Square Error (RMSE) for three key pollutants in new stations.
- Successfully addressed the data shortage problem for air quality prediction in under-resourced monitoring locations.
Conclusions:
- The proposed TLS-BLSTM method offers a robust solution for improving air quality forecasting accuracy in data-limited scenarios.
- Transfer learning combined with deep learning networks is a promising approach for overcoming data scarcity in environmental monitoring.
- The findings provide valuable support for developing effective air quality management strategies in regions with newly established monitoring infrastructure.