With the development of the intelligent transportation system (ITS), Car-following behavior at signalized intersections has become a hot topic in traffic flow theory. Many scholars propose various Artificial Neural Network (ANN) car-following models, however to our best knowledge, few people pay attention that the trajectory prediction performance of the different ANN car-following models at the signalized intersection. In this research, we first take the traditional Intelligent Driver Model (IDM) as an example, analyze the car-following behavior at the signalized intersection from the theory perspective, then choose the Lankershim-Boulevard dataset of the Next Generation Simulation (NGSIM) as the data source. In order to make the experiment more reality, stop-and-go behavior and glide behavior, two different test scenarios were set. In the experiment, five state-of-the-art car-following models were implemented, they include one physical car-following model which is called IDM, four different ANN car-following models which are called BP, Elman, General Regression Neural Network (GRNN), Nonlinear autoregressive exogenous model (NARX) respectively. The experiment indicates that when the vehicle takes stop-and-go behavior, the accuracy of the GRNN, NARX ANN car-following model with the ability to predict the time-series dataset is best, BP, Elman is medium, IDM is lowest. When the vehicle glides across the signalized intersection, four ANN car-following models perform better than the IDM, furthermore, there is no critical difference between four ANN car-following models.

CICTP