Research On The Correlation Between Air Quality And Traffic By Deep Learning

Air pollution is a growing worldwide problem,which concerns people’s lives and health.Recently,with the increasingly strict control of industrial emissions and the rapid growth of total urban vehicle number,traffic pollution has become a large drag to the sustainable developments.Therefore,a comprehensive understanding and analysis of how urban traffics impact air quality is urgent and very important to more effective and sustainable environmental governance.However,the traditional models cannot ef-fectively quantify the correlation between traffic emissions and air quality due to the large number of parameter assumptions and the failure to establish a clear relationship between traffic and PM2.5 concentration.In this thesis,based on the multi-source het-erogeneous data,we propose a novel source apportionment model DeepSA.Based on the accurate prediction of PM2.5 concentration,DeepSA achieves the spatiotemporal fine-grained positioning of traffic emissions,and quantifies the contribution of anthro-pogenic sources including traffic sources on PM2.5 concentration,finally analyzes and evaluates the impact of the future traffic emissions on PM2.5 concentration under dif-ferent policies.Based on the prior knowledge of the diffusion process of air pollution,this thesis first proposed a deep hierarchical multi-information fusion PM2.5 concentration pre-diction model,DeepAir,which uses the Traffic Fusion Module(TFM)to extract the temporal and spatial emission information of traffic,and uses the carefully designed feature interaction module(FIM)and temporal interaction module(TIM)to simulate pollution chemical reaction process and temporal accumulated process respectively.Compared to the best comparison model DA-RNN,mean absolute error(MAE)and rooted mean squared error(RMSE)were improved by 20.1%and 14.4%in 24h re-spectively.Next,we embed the latest Layer-wise relevance propagation(LRP)algo-rithm into the DeepAir model to realize a traceable and transparent model iDeepAir.The model takes traffic spatiotemporal information as input,and reveals spatiotemporal emission patterns of urban traffic while transparenting the pollution process.Further,we established a bottom-up integrated source traceability model DeepSA from emis-sion source to PM2.5 formation by coupling the emission inventories with the iDeepAir model,which quantified the contribution of anthropogenic sources to PM2.5 develop-ment for each year from 2010 to 2017.It is found that the contribution of Shanghai’s transportation sector to PM2.5 con-centration increased from 21.62%to 35.67%in 2010-2017,indicating that traffic emis-sions will increasingly become the issue of air pollution control.Finally,we analyzed and evaluated the future impact of traffic emissions on PM2.5 concentration under dif-ferent policies.We infer that the promotion of new energy vehicles can bring a 4.18%reduction of PM2.5 formation by 2030.These insights are of great significance to pro-vide the decision-making basis for accurate and high-efficient traffic management and urban pollution control,and eventually benefit people’s lives and high-quality sustain-able development of cities.