Towards Benchmarking Microscopic Traffic Flow Models

For the simulation of traffic flows various macroscopic and microscopic models exist. Developing these models it is important to check the models against reality, whereas there are well-established means how to do this for macroscopic fluid-dynamical models. For microscopic models this work has yet to be done. This article is intended as a first step towards a common benchmarking of those models. A method is suggested to test microscopic models against single-car data sets and applied to two public available single-car data sets recorded by C. Daganzo on a one lane road in the USA in 1997. The calibration of the used models is done by analysing the deviations of the travel times on the observed street segments. As a result one gets a single number for each model giving the mean percentage error it produces in comparison to reality. This way the models can be compared directly to each other quantitatively. Furthermore, the obtained results point towards some deficiencies of the models and give ideas how to correct them.