SIGNATURE WORK
CONFERENCE & EXHIBITION 2022

The outbreak and spread of COVID-19 pandemics resulted in a stringent lockdown policy in many areas in China in 2020, which reduced transportation, as well as commercial and industrial activities. In this study, machine learning techniques (Gradient Boosting) are utilized to evaluate and quantify the impacts of lockdown on the concentration of NO2 in three selected regions: Jing-Jin-Ji Region (JJJ), Yangtze River Delta Region (YRD), and Pearl River Delta Region (PRD). By simulating NO2 concentration under the business-as-usual scenario, it is estimated that the lockdown reduced NO2 by about 37-39% in the selected regions from January to March 2020. Besides, to assist the public to prepare for potential NO2 pollution in advance, the study also provides short-term forecasting for NO2 pollution in the next hour and next day with deep learning models such as Long Short-Term Memory network (LSTM) and Convolutional Neural Network-LSTM (CNN-LSTM). Moreover, Discrete Wavelet Transformation (DWT) is also adopted in data preprocessing to improve data mining. The performance of LSTM, CNN-LSTM, and DWT-LSTM models are compared. Among them, DWT-LSTM performed the best, with an accuracy of 81.3%, root mean square error (RMSE) of 6.83 (μg/m3), and mean absolute value (MAE) of 5.02 (μg/m3) for day-ahead prediction in JJJ. The results reveal that the combination of DWT and LSTM can effectively improve the performance of modeling in air pollutant prediction.