Comparative Analysis of Car-Following Models for Emissions Estimation

Abstract

Recent studies have indicated that the accuracy of the emissions estimation in a traffic simulation model can be little improved by using the traditional model calibration approaches. Instead, the model's internal mechanism in depicting the second-by-second vehicle activities needs to be investigated. Since the car-following model is the core component of a traffic simulation model, this paper attempts to conduct a comparative study of car-following models concerning their effects on the explanatory parameter of vehicle emissions, namely, the vehicle specific power (VSP) distribution. The car-following models selected for the analysis are the optimal velocity model (OVM), generalized force model (GFM), full velocity difference model (FVDM), Wiedemann model, and the Fritzsche model. Massive field car-following trajectories are collected, and a numerical simulation method is designed for each car-following model to generate its vehicle trajectories and the speed-specific VSP distributions. By a comparison of VSP distributions collected from the field and generated by car-following models, it was found that OVMs and GFMs generate unrealistic VSP distributions, which will lead to significant emissions estimation errors. By adding the variable of positive velocity difference, the FVDM can effectively improve the accuracy of the VSP distribution and emissions estimation. The VSP distribution of the Wiedemann model differs largely from the field data, which overestimate the peak VSP fraction and the fractions in aggressive driving modes. The Fritzsche model produces VSP distributions consistent with the field distributions. It is also found that the speed-specific VSP distribution is highly correlated with the acceleration distribution. Therefore, improving the accuracy of the speed-specific acceleration distribution is an effective measure to improve the accuracy of the VSP distribution and thus the emissions estimation of the car-following models.