Local search for shift design

Designing shifts is one of the important stages in the general workforce scheduling process. In this paper we consider solving the shift design problem by using local search methods. First we propose a set of move types that give rise to a composite neighbourhood relation. In the move selection process, we make use of the basic prohibition mechanisms of tabu search. In addition, in order to avoid having to explore the whole neighbourhood which could be prohibitively large, we evaluate the moves in decreasing order of their promise to yield some improvement. Furthermore, we propose an algorithm for generating a good initial solution, which also exploits knowledge about requirements and shift structure. Experimental results on both real-life and randomly-generated instances show the advantages of these ingredients. The solver is part of a commercial product and has shown to work well in practical cases.