An advanced model and novel meta-heuristic solution methods to personnel scheduling in healthcare

Constructing timetables of work for personnel in healthcare institutions is known to be a highly constrained and difficult problem to solve. In this thesis, we introduce a model for the practical rostering problem in Belgian hospitals. It is general enough to cope with the large set of constraints and to meet varying objectives encountered in practice. We set up a solution framework that consists of a modifiable and explanatory evaluation function, many options for handling initialisation parameters and for formulating various objectives, and meta-heuristics to search solutions. A set of neighbourhoods was designed for organising an effective exploration of the search space. We combined different local search heuristics with these neighbourhoods and managed to find scenarios that produce algorithms for widely varying problem settings. The hybrid tabu search approach deserves special attention because it is applied in practice, as part of a software package based on the model proposed in this thesis. A range of new memetic approaches for rostering is introduced. They use local search improvement heuristics within a genetic framework. We identify the best evolutionary operators of a memetic algorithm for the rostering problem, particularly the nature of effective recombination, and show that these memetic approaches can handle initialisation parameters and a range of instances more effectively than usual tabu search algorithms, at the expense of longer computation times. Having presented cost function based search heuristics, we finally introduce a new multi criteria approach which overcomes some practical difficulties for automated nurse rostering. The developed multi criteria approach, incorporated in a meta-heuristic, takes into account the fact that some constraints are easier to satisfy than others while allowing schedulers to control compensation of constraints. By automating the nurse rostering problem, the scheduling effort and time are reduced considerably in comparison with the manual approach that was previously used. The software based on the model developed in this thesis provides an unbiased way of generating the schedules for all the personnel members. It enables simple verification of the constraints, helps redefining unrealistic hard constraints, and thus leads to an overall higher satisfaction of the personnel.