Development of car-following model with parameter identification by genetic algorithm

A car-following model has been proposed as a part of a microscopic traffic simulation model for the evaluation of automobile's control systems. The car-following model imitates human drivers' logic in the calculation of longitudinal acceleration rate. It has been designed to be consistent with a macroscopic traffic stream model that fits real world's traffic stream well. The model has parameters that should be different among drivers. We have applied a genetic algorithm to identify the model parameters from the data obtained from experiment on a driving simulator. The model with identified parameters produces speed profiles similar to those of human driver. We have verified by simulations that the model enables stable starting and stopping of traffic stream. The stability in the presence of hard disturbance has been also confirmed. The traffic streams simulated from the model have macroscopic flow parameters that are close to those of real world's traffic.