SNCF is a large-sized railway transportation company that needs to be operated 365 days a year and 24 hours a day. In order to schedule a certain category of workers in train stations and selling points, rosters are designed to cover a cyclical demand. However, the highly combinatorial nature of the rostering problem makes it very difficult to solve manually, and experts spend a huge amount of time to derive implementable solutions that improve a number of preference criteria.

This thesis presents two formulations based on mixed-integer programming to adress the cyclical rostering problem. The first one uses variables to express the nature of each day of the roster, whereas the second one uses patterns corresponding to feasible blocks of seven days and assigns them to each week of the roster. Different strategies relative to the management of some preference criteria are compared, some of them leading to significant reductions in computational times. Cuts are finally introduced to improve the bounds obtained by the linear relaxation of the mixed-integer programs. The impact of these cuts on computational times depends much on the problem.