Lane selection behavior modeling in an agent-based traffic simulation

A dynamic-cognitive lane selection model is proposed under an agent-based traffic simulation. For the realistic lane selection modeling, this study employs two algorithms: (1) the extended decision field theory (EDFT), a psychological decision-making algorithm to represent complex real-time deliberation of drivers, and (2) the Bayesian network (BN) to mimic limited perception capabilities of drivers on dynamic road conditions. To calibrate the proposed lane selection model, a single next-generation simulation (NGSIM) traffic dataset on Peachtree Street in Atlanta, Georgia has been used. During the calibration process, its modeling accuracy is compared with the NGSIM discretionary lane changing model via a cross-validation approach. The calibrated lane selection model is then implemented into the AnyLogic® agent-based simulation platform in conjunction with a NGSIM car-following model in order to evaluate the performance of the proposed lane selection model in regard to the physical movements of vehicles on the roadway. Computational execution time and the lane changing behavior of drivers are investigated in the proposed lane selection model and compared with conventional lane selection models. The results have demonstrated the modeling flexibility and accuracy of the proposed lane selection model.