A study of evoluntionary perturbative hyper-heuristics for the nurse rostering problem.

Abstract

Hyper-heuristics are an emerging field of study for combinatorial optimization. The aim of a hyper-heuristic is to produce good results across a set of problems rather than producing the best results. There has been little investigation of hyper-heuristics for the nurse rostering problem. The majority of hyper-heuristics for the nurse rostering problem fit into a single type of hyper-heuristic, the selection perturbative hyper-heuristic. There is no work in using evolutionary algorithms employed as selection perturbative hyper-heuristics for the nurse rostering problem. There is also no work in using the generative perturbative type of hyper-heuristic for the nurse rostering problem. The first objective of this dissertation is to investigate the selection perturbative hyper-heuristic for the nurse rostering problem and the effectiveness of employing an evolutionary algorithm (SPHH). The second objective is to investigate a generative perturbative hyper-heuristic to evolve perturbation heuristics for the nurse rostering problem using genetic programming (GPHH). The third objective is to compare the performance of SPHH and GPHH. SPHH and GPHH were evaluated using the INRC2010 benchmark data set and the results obtained were compared to available results from literature. The INRC2010 benchmark set is comprised of sprint, medium and long instance types. SPHH and GPHH produced good results for the INRC2010 benchmark data set. GPHH and SPHH were found to have different strengths and weaknesses. SPHH found better results than GPHH for the medium instances. GPHH found better results than SPHH for the long instances. SPHH produced better average results. GPHH produced results that were closer to the best known results. These results suggest future research should investigate combining SPHH and GPHH to benefit from the strengths of both perturbative hyper-heuristics.