PM 2.5 has been identified as a major pollutant which is harmful to human health and causes destruction of ecosystems, and many investigations and studies have illustrated that air pollutants containing PM 2.5 cause severe diseases such as respiratory and cardiovascular diseases.
Since the correlation between different air pollutants and their own inherent characteristics is complicated, there have been many attempts to improve the forecasting accuracy by using deep neural network (DNN) for air quality forecasting. The results of these studies demonstrate that deep learning combined with spatiotemporal correlation analysis is of great significance in improving the performance of a model.
Pak Un Jin, a researcher at the Faculty of Automation Engineering, has proposed a new PM predictor to predict the daily average PM 2.5 concentration of the next day in Beijing City with regard to the seasonal pattern of air pollution.
He has demonstrated that the performance of the proposed PM predictor is excellent in comparison with MLP and LSTM models, and found clear evidences that the PM predictor is appropriate for overall forecasting and LSTM is more suitable than other models for seasonal forecasting.
If more information is needed, you can refer to his paper “Novel particulate matter (PM2.5) forecasting method based on deep learning with suitable spatiotemporal correlation analysis” in “Journal of Atmospheric and Solar-Terrestrial Physics” (SCI).
© 2021 Kim Chaek University of Technology