To effectively solve challenging staff rerostering problems, we propose to enhance a large neighborhood search (LNS) with a machine learning guided destroy operator. This operator uses a conditional generative model to identify variables that are promising to select and combines this with the use of a special sampling strategy to make the actual selection. Our model is based on a graph neural network (GNN) and takes a problem-specific graph representation as input. Imitation learning is applied to mimic a time-expensive approach that solves a mixed-integer program (MIP) for finding an optimal destroy set in each iteration. An additional GNN is employed to predict a suitable temperature for the destroy set sampling process. The repair operator is realized by solving a MIP. Our learning LNS outperforms directly solving a MIP with Gurobi and yields improvements compared to a well-performing LNS with a manually designed destroy operator, also when generalizing to schedules with various numbers of employees.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.