A simulation-based optimization approach for the calibration of dynamic train speed profiles

Predictions of railway traffic are needed by planners and dispatchers for the design of robust timetables and real-time traffic management of perturbed conditions. These tasks can be effectively performed only when using train running time models which reliably describe actual speed profiles. To this purpose calibration of model parameters against field data is a necessity. In this paper a simulation-based optimization approach is introduced to calibrate the parameters of the train dynamics equations against field data collected at the level of track sections. A genetic algorithm is used to minimize the error between simulated and observed speed profiles. Furthermore, a procedure for the estimation of train lengths has been developed. This method has been applied to trains with different rolling stock running on the Rotterdam-Delft corridor in the Netherlands. The model parameters were calibrated for a significant number of trains of different compositions. We also derived probability distributions for each parameter which can be usefully employed for simulations. The results show that the train length estimation model obtained good computation accuracy. The effectiveness of the calibration method in giving a reliable estimation of the real train path trajectories is shown. It has been observed that some of the parameters of tractive effort and resistance do not affect the train behaviour significantly. Also, the braking rate is significantly smoother than the default value used by the railway undertaking while calibrated resistance parameters tend to have lower mean than defaults. Finally, the computational efficiency of the approach is suitable for real-time applications.