Calibration of Car-Following Models Using Floating Car Data
We study the car-following behavior of individual drivers in real city traffic on the basis of publicly available floating car datasets. By means of a nonlinear optimization procedure based on a genetic algorithm, we calibrate the “Intelligent Driver Mode
- PDF / 2,544,328 Bytes
- 11 Pages / 467.717 x 685.984 pts Page_size
- 59 Downloads / 343 Views
Summary. We study the car-following behavior of individual drivers in real city traffic on the basis of publicly available floating car datasets. By means of a nonlinear optimization procedure based on a genetic algorithm, we calibrate the “Intelligent Driver Model” and the “Optimal Velocity Model” by minimizing the deviations between the observed driving dynamics and the simulated trajectory when following the same leading vehicle. The reliability and robustness of the nonlinear fits can be assessed by applying different optimization criteria, i.e., different measures for the deviations between two trajectories. We also investigate the sensitivity of the model parameters. Furthermore, the parameter sets calibrated to a certain trajectory are applied to the other trajectories allowing for model validation. We found that the calibration errors of the Intelligent Driver Model are between 11% and 28%, while the validation errors are between 22% and 30%. The calibration of the Optimal Velocity Model led to larger calibration and validation errors, and stronger parameter variations regarding different objective measures. The results indicate that “intradriver variability” rather than “inter-driver variability” accounts for a large part of the fit errors.
1 Introduction As microscopic traffic flow models are mainly used to describe collective phenomena such as traffic breakdowns, traffic instabilities, and the propagation of stop-and-go waves, these models are traditionally calibrated with respect to macroscopic traffic data. Nowadays, as microscopic traffic data have become more and more available, the problem of analyzing and comparing microscopic traffic flow models with real microscopic data has raised some interest in the literature [1–3]. We will consider three empirical trajectories of different drivers that are publicly available and that have been provided by the Robert Bosch GmbH [4]. The datasets have been recorded in 1995 during an afternoon peak hour on a fairly straight one-lane road in Stuttgart, Germany. A car equipped with a radar sensor in front provides the relative speed and distance to the car ahead.
118
Arne Kesting and Martin Treiber
The duration of the measurements are 250 s, 400 s and 300 s, respectively. All datasets show complex situations of daily city traffic with several acceleration and deceleration periods including standstills due to traffic lights. Here, we will apply the “Intelligent Driver Model” [5] and the “Optimal Velocity Model” [6] to these empirical trajectories. By means of a nonlinear optimization, we will determine the “optimal” model parameters which fit the given data best. Since the fit errors alone do not provide a good basis for an understanding or benchmarking of the applied car-following models, we will address further investigations on the role of the objective functions, the structure of the parameter space and the robustness concerning validation. In the following section, we introduce the car-following models under investigation, the simulation set-up, the objective functions and the nonline
Data Loading...
