Train dispatchers are responsible for guaranteeing safety and increasing efficiency in railway operations. In such a dynamic and stressful working environment, adopting proper workload policies and accurately implementing them is vital. An important aspect to address to improve the dispatcher’s working condition is the quality of their working shifts. 

Given the high safety concerns of their profession, dispatchers’ working shifts should fulfil several legal and operational constraints, such as bounds on the length of shifts and on the resting time periods between shifts. Constructing shift schedules for dispatchers is a complex and time-consuming process that is currently done manually. Research-wise, many researchers suggested optimization models to automatize shift scheduling for several professions. 

We formulated an Integer Programming (IP) problem to optimize shift schedules for Malmö dispatching centre using artificial data of a real-world size. The objective function was to minimize the number of used dispatchers to cover a set of geographical areas during a single working day. We develop a stronger formulation that reduces the runtime, then we add several constraints to increases the attractiveness of the schedule. The improvements in attractiveness consist in avoiding too short shifts, eliminating starting times at night (between 00 and 6 am), and minimizing the number of unnecessary handovers, where these occur when a dispatcher changes the assigned controlled geographical area during the working shift. 

We proposed four approaches for tracking the handovers and we ran several experiments starting with a small instance. We realized that two of the approaches do not solve the problem within a reasonable time, then we decided to go further with more experiments using the two remaining approaches. In one group of experiments, we fixed the number of areas and increased the time resolution within a single day, while in a second group we did the opposite. Although all the approaches resulted in improved schedules with the fewest number of handovers, the runtime is still an important factor for determining the most efficient approach. Moreover, we used the approach with the lowest runtime to solve an instance with the real-world size of 15 areas and a time resolution of 12 periods a day. Finally, we ran the same instance using our old formulation to compare the two models.