Calibrating Car-Following Models on Scenarioized Data using GA-NM Algorithm

As the core content of microscopic traffic flow theory, the car-following model links human microscopic driving behavior with macroscopic traffic phenomena, which is widely used in traffic simulation and traffic safety evaluation. In our opinion, calibrating these models to real-world data is crucial. Previous works mainly focused on the calibration of car-following models on special scenarios or coarse-grained scenarios. In this paper, we employ K-means for the fine-grained mining of vehicle-following scenarios, and we apply a novel hybrid optimization algorithm GA-NM to calibrate the parameters of the car following model in four car following scenarios. The experiment results show that the GA-NM algorithm consistently outperforms the original GA algorithm, and FVD model has the lowest and most stable calibration errors, which are suited for Chinese highway simulations.