Calibration and Evaluation of Car Following Models Using Real-World Driving Data

Traffic simulation software are used to evaluate the effect of driving behavior, vehicle technologies and infrastructure on traffic flow and energy demand. To achieve accurate results, the driver behavior in traffic simulation needs to be representative. Since the driver behavior are simulated by car following models, these models need to be calibrated to reproduce traffic flow as well as energy demand similar to drivers in real traffic. In this paper, three car following models commonly used in traffic simulation, i.e., the intelligent driver model (IDM), the Krauss car following model, and the Wiedemann car following model, are calibrated and verified on around 200 recorded trips. The trip data are collected from human drivers driving on the Drive Me route, which is an autonomous drive test route in Gothenburg, Sweden. To find the optimal parameters of the car following models, a Genetic Algorithm minimizes a performance index which is derived to reflect the error in the vehicle's impact on traffic flow as well as its energy demand, compared to a reference trip. The result of the simulation shows that in average, the optimized IDM can replicate driving behaviors best compared to the other two models. It is also shown that adding a time delay improves the performance of all the models.