Source Apportionment Of Air Pollutants In Shanghai Based On Machine Learning

Recent years,the problem of air pollution is becoming serious,and its influence is gradually expanding.In order to better control the air pollution and take effective measures,it's necessary to analyze the sources of aerosol in the urban atmosphere.Accurately grasping the sources of pollutants is conducive to controlling pollution from root,so as to improve air quality and provide a good living environment for human survival activities.This study is based on data of pollutant concentration and water soluble ion concentration monitored by hourly in 2015-2016.We studied the temporal and spatial distribution characteristics of air pollutants in Shanghai by statistical analysis.Data mining and machine learning algorithms are increasingly being applied to air pollution research.In this paper,the source analysis model based on GBRT and random forest algorithm is proposed,and the quantitative analysis of the source of particulate matter in Shanghai is made,and some suggestions and measures are put forward for the prevention and control of air pollution in Shanghai.According to this study.The following conclusions are obtained:?1?In Pudong New Area of Shanghai,the average annual concentration of PM_2.5was 51.4?g/m³,and the average annual concentration of PM₁₀ was 71.3?g/m³ in 2015;the average annual concentration of PM_2.5 was 40.9?g/m³,and the average annual concentration of PM₁₀ was 61.12?g/m³ in 2016.Pollutant concentration showed a downward trend overall,but it still did not meet the national standard emission limit.The distribution of pollutant concentration showed a seasonal change.In December to April,there was a high concentration period.The concentration of August and September was lowest in the year.The daily variation characteristics of pollutants was related to the urban traffic condition.There were two peaks,which was coincided with the rush hour.Besides,the concentration distributions of water-soluble ions and some gaseous pollutants were similar to those of particle concentration distribution,showing seasonal variation.?2?It was found that the ions showed positive correlation by analyzing the water soluble ion data.The highest correlation coefficients ware between NO₃^-?SO₄^2-?NH₄⁺which were 0.83,0.69 and 0.68,and the correlation between Na⁺?Ca²⁺whit other ions was weak,and the correlation coefficients were just about 0.1,0.2.The principal component analysis method is used to determine the pollution source to two aerosol products,motor vehicle exhaust and coal combustion,biomass combustion or natural emission,sea salt and dust.?3?A regression model of pollution source and particulate matter is established by machine learning algorithm.The contribution of pollution sources to the measured particulate matter concentration is analyzed.Comparing the fitting results of the model,the results show that the fitting effect of the machine learning method has been improved obviously,in which the goodness of fit of GBRT is increased by 8.5%,the goodness of fit of random forest is increased by 15%,and the fitting result of random forest is the best.According to the analytical results of PCA/random forest model,the contribution rate of four types of pollution sources is:PC1 represents two aerosol products,motor vehicle exhaust and coal combustion,the contribution rate is 63.4%,PC2 represents biomass combustion or natural emission 17.9%,PC3 sea salt contribution rate is 9.8%,PC4 dust 8.9%.?4?The principal component/stochastic forest model was used to analyze the pollutants in Pudong New Area of Shanghai in different seasons.It is considered that the model has strong universality,and the analytic result obtained is ideal.The emission of vehicle exhaust and coal burning are the main sources of pollutants in Shanghai.The sea salt pollutants do not belong to the main source of pollutants.The combustion and decomposition of biomass and the contribution rate of the dust are seasonal changes,which need to be considered in a period of time.