Simulating bicycle traffic by the intelligent-driver model-Reproducing the traffic-wave characteristics observed in a bicycle-following experiment

Bicycle traffic operations become increasingly important and yet are largely ignored in the traffic flow community, until recently. We hypothesize that there is no qualitative difference between vehicular and bicycle traffic flow dynamics in single-file case, so the latter can be described by reparameterized car-following models. To test this proposition, we reproduce German (Andresen et al., 2014) and Chinese (Jiang et al., 2016) bicycle experiments on a ring with the intelligent-driver model (IDM) and compare its fit quality (calibration) and predictive power (validation) with that of the necessary-deceleration-model (NDM), which is specifically designed for bike traffic. We find similar quality metrics for both models, so the above hypothesis of a qualitative equivalence cannot be rejected. Moreover, calibration errors of the IDM turn up to be slightly smaller compared to the NDM ones. The NDM represents significant calibration errors for high flow densities, which correspond to flow states, when stop-and-go wave emerge. According to validation tests, the IDM outperforms the NDM as well. Performing two types of validation techniques we discover that inter-driver variation is much higher than the intra-driver variation for bicycle traffic. It coincides with the results obtained from vehicular traffic experiments (NGSIM trajectory data). In addition, we suggest the measure for quantitative comparison of two microscopic fundamental diagrams, which are derived from experimental data and simulated trajectories. The analysis of speed-density relations provides more or less the same results for both models.