Comparative Analysis And Optimization Of Car-following Models For Emission Estimation

The rapid increase of vehicle emissions inurban traffic is deteriorating the air quality, which has attracted increased attentions to the problem of traffic emissions. As a result, extensive relevant research has been conducted. The efforts to estimate traffic emissions by integrating a traffic simulation model with an emission model have been becoming an evolved research area. However, recent studiesindicated thatthe use of existing traffic simulation models may lead to significant errors in emission estimation due to theproblems indescribing vehicle accelerations in these models.In this context, this thesis attempts to conduct a comparative study of car following models on their performances in emission estimation based on the best explanatory parameter of emissions, Vehicle Specific Power (VSP) distribution, and develop an optimization to the car-following models.First, the thesis provides an overview of the state-of-the-art on car following models, emission estimation methods, and the model integration between traffic simulation and emission models. It recognizesthatthe average-speed-specific VSPdistributionis an effective approach to describe the characteristics of vehicle emissions of traffic flow. Thereafter, typical car following models including Collision Avoidance models (CA models, including OVM, GFM,FVDM and IFVDM) and Psycho-physical models (including Wiedemann and Fritzsche model) are selected as the targetof study, and the emission estimation models based on the average-speed-specific VSPdistribution are chosen for the purpose of calculating emissions.Second, massive car-following driving behavior data of light duty vehicleson Beijing expressways are collected by using GPS. Then, it designs a numerical-simulation-basedexperimental method to generate the following car’s activity data by using the real-world vehicle activity data as the input of leading car in car-following models. Indicators of the relative mean square error of VSP distribution and the relative error of emission factorsare used to evaluate errors of numerical simulation results of different car-following models. It is found that theVSP distribution resulted from CA models (FVDM and IFVDM) is similar to the field VSP distribution in low speed range, but produces high errors in high speed conditions (ASI>25). VSP distribution provided by Wiedemann model hasapparent differencein comparison to the field data, and its output of vehicle’s acceleration is proved to be much more aggressive than the field activity. VSP distribution provided by Fritzsche model is the most consistent one with the minimum relative error in the estimation of emission factors.Finally, the thesis conducts a sensitivity analysis for key parameters of the car following models. The VSP distribution and emission estimation results based on the default parameter values, and the adjustment of30%up and down are compared. It is found that the emission-sensitive parameter in Wiedemann model is the maximum acceleration (bmax). However, an adjustment of bmax fails to improve the accuracy of VSP distribution for Wiedemann model. Fritzsche model has several emission-sensitive parameters including While Desired Time (TD), Risk Time (Tr), Safe Time (Ts), and Acceleration (an). The values of these parameters are then optimized forFritzsche model based on the indicators of the mean square error of VSP distribution, the relative error of emission factors and the error of emission estimation. After the emission calculation and comparison, the reduction of the average of Error of Emission Estimation is54.9%in the optimized model.