The Analysis And Modeling Of The PM2.5 Concentration Data In Road Construction Lane

A large number of subway road construction projects were carried out in Chengdu which was undergoning a rapid development to build the City of Rail Transit in the World.The road construction may directly or indirectly influence PM2.5 emission by affecting the traffic flow and lead to deterioration of air quality who is a serious threat to people's health and safety.As well,particulate matter pollution consentrate more easily in the road construction area.So the reasearch of PM2.5 change mechanism in work zone had important significance to the concentration warining of PM2.5,the environmental impact eveluation and optimization of the traffic strategy.The data cleaning was carried out with the combination of multiple interpolation and the nearest neighbor mean interpolation on the basis of the control experiment in the work zone.Then,wavelet denoising was proposed by cosidering PM concentration as discrete signal wavelet.The statistics,correlation analysis,auto-correlation analysis and cluster analysis were applied to the contrast of the construction area and the normal traffic section.Aiming at the problem of multicollinearity between independent variables,a modified GA-BP neural network prediction model was built.The wavelet denoising suppressed the noise data effectively and promoted the performance in therrelation analysis and auto-orrelation analysis based on maximizing retention of the original timefrequency information and it proved to be a good application in the air pollutant concentration data denoising.The average PM centration increased by about 20% in working zone and it was demonstrated the increase is mainly due to the vehicle emission mode and the PM pervasion environment change.The weather conditions had a significant influence on the PM concentration data,and the motor vehicle emission is higher in winter.PM concentration data were negatively correlated with temperature,and positively correlated with humidity.The both absolute value of correlation coefficient were more than 0.8.To solve the problem of multicollinearity between variables,compared with principal component analysis,genetic algorithm could overcome the problem of the test sample principal component information change.The GA-BP-PSO neural network prediction model of PM2.5 concentration in work zone had a good application in independent variable dimensionality reduction.