摘要：AbstractIn order to achieve a complete insight in the state-of-the-art of traffic modelling, several typical car-following models are evaluated by using trajectory data from real traffic conditions and genetic-algorithm-based calibration method in this study. The models with calibrated parameters are validated not only under uncongested traffic conditions but also under congested traffic conditions. Unlike the results in previous study based on experimental data, there are obvious differences in the performance of these models. Models with more parameters produce relatively lower error rate in calibration process but over-fitting problem appears in validation process. The model very popular in the physical community is found not suitable for real traffic simulation, although it can represent some traffic phenomena under certain condition. Even with simple rules and discrete variables, cellular automata model achieves satisfactory simulation results both in calibration and validation process. Besides, it is also noteworthy that all of the models perform rather worse in validation process than in calibration process. Using different parameters or even different models under different traffic conditions seems to be feasible for depicting real traffic more accurately.
