Time Series Analysis for Air Pollution Data of Beijing

Abstract

For a long time, air pollution has been a serious problem encountered Beijing due to the city’s rapid economic development, high level social production and living standard, and unfavorable climate. This paper will use hour-by-hour data for six air pollutants and eight weather variables collected from January 1, 2014, to December 31, 2015, at an air pollution monitoring site in Beijing. We mainly analyze the data on SO2 pollutants with the ARIMA model which is used to make predictions on SO2 data. Since the periodicity of hour-by-hour data is too long and fluctuates drastically, using the ARIMA model will face problems such as lacking valid forecasting time and significant forecast bias. Besides, we have data on many climate indicators, including temperature, moderation, and concentration of many pollutants, which may be correlated. Therefore, with proper treatment, variable correlation analysis can assist us in predicting specific indicators. We use two different time series decomposition methods in the SO2 variable. The relationships between the periodic terms and air temperature and humidity were compared separately to verify the conclusions of some previous papers. Moreover, SCINet was used to predict and get excellent results.