A fuzzy logic-based multi-agent car-following model

• A completely artificial intelligence CF (CAICF) model is built to imitate a driver.
• An extensive five-layer structure, and a fuzzy logic mechanism are used in CAICF.
• A genetic algorithm is designed to calibrate the parameters of CAICF model.
• The veracity and stability of CAICF model are validated based on NGSIM data.

Most existing analytical car-following models can simulate traffic flow realistically from some aspects, such as stop-and-go, congestion, and nonlinear characteristics, but cannot predict the driving behavior and psychological process as a driver is interacting with preceding vehicles. The reason for that is the difficulty that is generated by the impact of human factor. In this paper, a completely artificial intelligence car-following model, which has no analytical model incorporated, is developed to accurately imitate a human driver. This model comprises the classic stimulus–response framework, an extensive five-layer structure, Perception–Anticipation–Inference–Strategy–Action, and a fuzzy logic-based inference mechanism. A genetic algorithm is employed to calibrate the parameters of this model. The results of experiments, which were conducted by using Next Generation Simulation (NGSIM) dataset to validate the proposed model, indicate that the vehicle trajectories simulated by this model coincide with the actual vehicle trajectories in terms of mean value and deviation. In addition, they show that the proposed model has very good stability.