Calibrating dynamic train running time models against track occupation data using simulation-based optimization

In the last decades advanced simulation models have been more and more used by railway timetable designers and dispatchers to support both the off-line planning and the real-time management of traffic. Fundamental requirements for these models are the accuracy and reliability of describing real train dynamics. To this aim it is necessary to calibrate train running time models against real data collected from the field. In this paper a simulation-based calibration approach is proposed to fine-tune the parameters of the different phases of train motion (acceleration, deceleration, coasting and cruising) against track occupation data. A customized genetic algorithm is developed to minimize the error between observed and simulated data. The model has been calibrated for different classes of trains against a significant number of observed trains running on the Dutch corridor Rotterdam-Delft. A probability distribution is then estimated for each parameter to understand how driver behavior affects their variations and to identify the most probable value for each of the parameters. The results show the ability of the proposed model to calibrate train parameters robustly and reproduce observed train trajectories accurately. It is observed that the coasting phase is not applied frequently on the case corridor. Also, drivers adopt a braking rate that is significantly smoother than the default value used by the railway undertaking.