On-line calibration of car-following models exploiting input parameter correlation

Abstract

new boost [8-9], in light of new data-collection techniques, Car-following models have been extensively studied in the mostly related to the increasing availability of trajectory last decades. However, the calibration of these models data [10-12] which of course introduce other challenges requires further research, so that drivers’ behavior could be [13]. simulated in high accuracy. The effectiveness of car- The objective of this paper is to motivate, develop and following models is closely related to the potentially best demonstrate with real data a practical and simple approach estimation of their parameters, mainly the calibration for the online calibration of microscopic traffic simulation process. Default parameter values may be inappropriate for models, which considers dynamic parameters for individual capturing driving behavior under various circumstances. In drivers, in time and space. Firstly, a literature review is order to address this problem, an alternative presented in the following section. Then, the overall methodological approach for calibration of car-following methodological framework is presented. A case study setup models is proposed in this research. Different sets of to demonstrate the feasibility and superiority of the parameter values could be set for each time instant of the approach, over previous techniques, is then presented, model, as traffic conditions are modified dynamically over followed by a discussion of the conclusions and future the time. Optimal model parameters are closely related to prospects. input parameters and thus an on-line calibration is proposed. In addition, a static calibration is illustrated and 2. LITERATURE REVIEW is used as a reference benchmark for comparison. Real Reliable representation of driving behavior is a crucial trajectory data was available from an experiment conducted issue for traffic simulation. Appropriate simulation models in Naples and was used for a case study in this research. are chosen according to the requirements of each The results from static and on-line calibration of a well- application; when considering the modeling detail, traffic known car-following model are also compared with the simulation models can be divided into microscopic, results from a previous work in which same data have been mesoscopic and macroscopic. Microscopic models provide used. On-line calibration seems to be a promising solution the highest level of detail for advanced transport for the development of flexible and reliable models. applications [14]. However, the traditional static calibration approach may not allow the incorporation of driving