Personnel scheduling problem with respect to the workforce demand and rostering has been a subject of research for a long time. Due to the high complexity of the problem, the optimal solution cannot be found in a reasonable amount of time. That is why new methods or their combinations are suggested for automatic scheduling with respect to the known set of constraints. Obviously, employers want employees to perform as many tasks as possible during their working time. On the other hand, they pay attention to the employee’s requests by suggesting flexible working hours, desirable sequence pattern of working days and work time. Optimization with a large set of constraints is not practical if a combination of evolutionary algorithms with the greedy approach is applied. It is planned to design a machine learning based algorithm and implement its prototype during this project. This algorithm would be based on extracting the favourable patterns in the schedule (roster combinations) and using them in the later calculations.
KTU Science and Innovation Fund
Reinforcement-based algorithm to solve personnel rostering problem was created in this project. Numerical experiments were carried out for the benchmark dataset instances published in schedulingbenchmarks.org/, which describe roster requirements (shift demand and constraints) and are popular in scientific research. Although the best-known solutions have not been obtained using this algorithm, the idea to transfer the information from previously generated schedules with different number of employees and problem horizon was confirmed.
The novelty of the project is that the created algorithm enables to generate a schedule under consideration of the experience accumulated while generating other schedules under the analogous set of constraints. The description of the state by the set of categorical variables enables to apply the reward estimations to employees with various parameter values which define number of maximum consecutive shifts, minimum number of consecutive days off, etc.
Period of project implementation: 2020-04-14 - 2020-12-31