Analysis And Prediction Of PM2.5 Concentration Based On Multivariate Methods And Time Series

Haze has become a prominent problem of air pollution in China.PM2.5 is the main component of haze,which has complex causes,wide sources and serious hazards.In order to predict and control the concentration of PM2.5,it is necessary to carry out in-depth detection and research on PM2.5.Firstly,this thesis analyzed the influencing factors of PM2.5 pollution in Changsha City.Through scatter plot and correlation analysis of PM2.5concentration data and the other five indexes of air quality index,as well as the analysis of synchronous change curve and correlation between PM2.5 concentration data and five indexes of meteorological factors,it was found that PM2.5 had a strong positive correlation with PM10.PM2.5 had a strong positive correlation with CO,NO₂ and SO₂.However,the correlation between PM2.5 and O₃ was negative but not significant.PM2.5was positively correlated with air pressure to a moderate degree,and negatively correlated with temperature,relative humidity,precipitation and wind speed.Further,principal component analysis model and stepwise regression model are constructed according to the analysis,and the relatively better influencing factor analysis model was selected as stepwise regression model through model fitting and prediction effect comparison.Secondly,through the time series analysis of PM2.5 historical data,the ARIMA model is constructed to predict the PM2.5 concentration values for the four days from December 28 to December 31,2017.The average relative error of prediction is 12.35%,and the short-term prediction effect is better.Finally,the PM2.5 pollution in Changsha is analyzed and predicted by building BP neural network,used PM2.5 related data in 2015 and 2016as training samples,and 2017 related data as testing samples.The sum of accurately predicted air quality grades and credible predicted days are 358days,the percentage is as high as 98.08%.The average relative error between the predicted and actual values of the air quality index is 18.21%,and the average relative error between the predicted and actual values of PM2.5 is 22.34%,which verified the possibility of applying BP neural network to the long-term prediction of PM2.5 in Changsha.In this thesis,the principal component regression model,stepwise regression model,ARIMA model and BP neural network model were established for the time series data of PM2.5 concentration in changsha to analyze and predict the PM2.5 concentration in changsha,providing relevant Suggestions for the prevention and control of PM2.5 in changsha.