Prediction Of Air Pollutant Concentration Based On XGBoost Algorithm And Potential Source Area Analysis

When a certain substance or a certain kind of substance in the atmosphere accumulates to a certain concentration and lasts for enough time,it will cause damage to the ecosystem and human living environment,and harm human beings,animals and plants.As the main pollutants,PM2.5,PM10 and O₃ have attracted much attention at present.The scientific and accurate prediction results of air pollutants can guide public activities and reduce the exposure risk of residents’air pollution.It can also provide some reference for environmental protection departments in formulating prevention and control measures.In this paper,Hefei city is selected as the research object to collect the concentration data of major air pollutants and hourly meteorological data in the corresponding time period.The monitoring stations cover10 automatic monitoring stations controlled by the state for ambient air quality.The monitoring scope is comprehensive and can basically represent the change characteristics of air quality in Hefei city.The contents to be completed in this article include:（1）The results of statistical analysis show that the time distribution of PM2.5 and PM10 is similar,and the concentration value decreases year by year,and both reach the highest value in winter.The concentration of O₃ increased year by year,and the highest value appeared from May to September.While the concentrations of PM2.5and PM10 decreased,the concentration of O₃ increased,and there was no linear correlation.The correlation degree of PM2.5 and PM10 with meteorological factors is as follows:Rainfall>Air Pressure>Wind Speed>Air Temperature>Relative Humidity;The correlation degree between O₃ and meteorological factors is as follows:Air Temperature>Wind Speed>Air Pressure>Relative Humidity>Rainfall.（2）The grey correlation analysis results are used for the preliminary selection of features.Firstly,the benchmark XGB model is established and the super parameter is optimized,and then the EEMD is used to decompose the data,concentrate all high-frequency noise into one sub sequence,model all sub sequences,and summarize and establish the EEMD-XGB coupling model.Finally,the root mean square error RMSE and goodness of fit R²of benchmark XGB model,superparametric optimization XGB model and EEMD-XGB coupling model are calculated respectively.The results show that the prediction performance of EEMD-XGB coupling model is RMSE=8.56 and R²=0.8526,which is significantly higher than that of benchmark XGB model（RMSE=16.84 and R²=0.3618）and superparametric optimization XGB（RMSE=15.05 and R²=0.4675）,and can meet the requirements of O₃ concentration prediction.（3）The meteorological data of the global data assimilation system（GDAS）are analyzed by using the meteoinfo software.Combined with the mass concentration data of air pollutants,the air pollution transmission process in Hefei from 2014 to2020 is simulated and analyzed.Through cluster analysis and transportation trajectory classification,the transmission path and main pollution source areas of large air pollution are determined.The results show that the air mass trajectory and transportation route are mainly concentrated in the north in January.In April,the air flow trajectory is no longer dominated by the northwest,but there are still more air mass routes in the north.The subtropical high has an impact on the air mass transport in July,making the air mass trajectory from the South longer.In October,as the cold air from the northwest gradually decreased to the south,the air mass was mainly imported from the north.Combined with potential source contribution factor（PSCF）analysis and concentration weighted trajectory（CWT）analysis,WPSCF and WCWT high value areas do not show the distribution trend centered on Hefei.