A Data-driven Approach For Mesoscopic Traffic Emissions Estimation And Applications

With the rapid economic growth,urban cities attract a huge number of activities and generate significant amount of travel demand.Transportation systems,as the core to the operation of cities,oftentimes produce emissions and air pollutants that adversely affect public health,local ecology,and even the global climate.To tackle these severe environmental problems,there is an increasing need in understanding the physic between dynamic transportaion systems and emissions.To this end,this work focuses on the correlation between urban traffic and emission,utilizing the field data collected from Beijing which are typically unavailable.1.We set up 3 sensors and obtain the concentrations of the major vehicle related pollutants(_2.5,₁₀)from 3 sites located in the northwest,southwest and southeast of Beijing.Simultaneously,we record Hourly Air Quality Index(,,_2.5)from an open source data center,located at point D,and Beijing Environment Protection Agency,who has established 35 air quality monitoring stations.2.To fill in this gap,we propose to combine the Vehicle Specific Power?VSP?model with the Fundamental Diagram model which captures the microscopic and mesoscopic traffic dynamics respectively.SAE?speed-acceleration-emission?model and SFE?speed-flow-emission?model are thus developed to reveal the relationship between traffic and emission at vehicle and link level.Finally,a mesoscopic vehicular emission concentration model has been established considering Gaussian Dispersion Model.3.Given these data,firstly,we estimate the Pearson Coefficient to reflect the correlation magnitude between air quality and its influential factors,including 5meteorologic factors?temperature,humidity,pressure,etc.?,and 5 traffic factors?such as speed,density and acceleration?.Our result indicates that traffic indeed can have impacts on air quality,an phenomenon which is insufficiently investigate in the current exisiting research works.However,both high positive and negative correlation exists.4.Based on the captured air quality data,we build the prediction model to estimate air quality using machine learing?e.g.ANN,SVR?.Adding traffic can improve the prediction accuracy with a small extent,while the concentration 1 hour first can enhance the model greatly.The proposed model addresses the correlation between traffic and emission,an this modeling tool can be further used to support decision-making on environment-oriented traffic management strategies,providing scientific and quantitative basis to governments.Howener,the results reveal that traffic indeed affects air quality,but infinitely.