Dwell time is crucial for railway operations corresponding to 20% of the total travel time in a mass transit context. With this, it is also a source of delays due to its stochastic nature. One way to ensure the robustness of a timetable is to add margins and much work has been done on run time margins. On the other hand, dwell time margins received little attention, except for a few heuristics. This paper aims to provide a novel method to define dwell time margins. To do so, we introduce the notion of tight dwell time (dwell without margins) in this work and compute it from high-resolution passenger flow data. Then, we propose two novel methods (the cluster method and the quantile method) to estimate it.

Given the access to this data, the method enables the estimation of the tight dwell time for all the stops while existing heuristics are limited to late trains and/or few passengers. Our developed method highlights the propensity of existing heuristics to overestimate what they measure. The estimation, a posteriori, of dwell time margins thanks to the computed tight dwell time would help design future timetables.