Improving the shift-scheduling problem using non-stationary queueing models with local heuristic and genetic algorithm

Abstract

We improve the shift-scheduling process by using nonstationary queueing models to evaluate schedules and two heuristics to generate schedules. Firstly, we improved the fitness function and the initial population generation method for a benchmark genetic algorithm in the literature. We also proposed a simple local search heuristic. The improved genetic algorithm found solutions that obey the delay probability constraint more often. The proposed local search heuristic also finds feasible solutions with a much lower computational expense, especially under low arrival rates. Differently from a genetic algorithm, the local search heuristic does not rely on random choices. Furthermore, it finds one final solution from one initial solution, rather than from a population of solutions. The developed local search heuristic works with only one well-defined goal, making it simple and straightforward to implement. Nevertheless, the code for the heuristic is simple enough to accept changes and cope with multiple objectives.