An Intelligent Evolving Car-Following Model

The existing data-driven car-following models do not take structural and range variations of velocity and acceleration into account. This paper develops a new approach to address both structural and range variations in the car-following process through time. The proposed approach relies upon an intelligent evolving time-variant local model (ETLM), capable of changing its structure and adapting its parameters. The evolving model includes a network of temporal local linear models, each covering a range of car-following behavior in a microscopic traffic flow. Furthermore, a decision-making procedure is designed to determine if model should evolve to a new structure is sole adaptation of its parameter is sufficient to describe the new behavior of the car-following process. The decision-making is carried out based on the comparison between the current temporal linear behavior of the process and existing temporal local linear models. Results of implementation of the ETLM on several benchmark case studies as well as real traffic data demonstrate the efficacy of the proposed approach. Comparisons to other methods show the superior performance of the ETLM model.