Research On Forecast And Influencing Factors Of PM_2.5 Concentration In Chengdu

The rapid development of industrialization and urbanization in China has brought serious air pollution problems to most parts of the country,the haze weather is one of them.The haze weather not only seriously affects people’s daily working life,physical and mental health,but also is not conducive to the sustainable development of society.How to control and prevent haze weather is an urgent problem that needs to be solved.Haze weather is mainly caused by PM_2.5 in the atmosphere.Therefore,it is necessary to analyze and study PM_2.5.Firstly,based on the PM_2.5 concentration data of Chengdu from 2017 to 2019,this paper analyzes the temporal distribution characteristics of PM_2.5 concentration in Chengdu,and finds that the quarterly average of PM_2.5 concentration is the highest in winter and the lowest in summer and other statistical characteristics.Secondly,through correlation analysis and stepwise regression analysis,the relationship between PM_2.5and other atmospheric pollutants and meteorological factors was explored,and it was concluded that PM_2.5 and O₃ show a weak negative correlation,while the positive correlation between PM_2.5 and SO₂,NO₂,CO and PM₁₀ increases in turn.PM_2.5 is positively correlated with atmospheric pressure,and the negative correlations with temperature,wind speed,total precipitation,and relative humidity gradually decrease.Stepwise regression analysis shows that other air pollutants have a greater impact on PM_2.5 than meteorological factors.Finally,this paper makes a one-step prediction of PM_2.5 in three ways:the first,ARIMA model prediction;the second,BP neural network prediction based on ARIMA predicted values of influencing factors and historical values of PM_2.5;the third,BP neural network prediction based on historical values of influencing factors and historical values of PM_2.5.The effect of each prediction method was compared by error analysis indexes such as mean square error root,and finally the analysis shows that the second method has the best prediction performance.