Search Algorithms for Nurse Scheduling with Genetic Algorithms

Abstract

This paper investigates the use of a genetic algorithms (GA) approach to a multi-objective optimization problem, in particular the nurse scheduling problem (NSP). Because GAs are computationally intensive algorithms, there is a strong need to make it effective. We define effective as producing good results in a short time. Our efforts so far revealed that for solving of the NSPs using a GA approach, unwanted premature convergence occurs in the early stages of the search. This is due to the presence of strong constraints and limiting of the search to feasible regions only. Therefore, diversification of the solution space plays a very important role for exploration of the potentially unexplored regions of the solution space. A mutation operator, or some kind of niching method is supposed to overcome this problem. The authors’ task is to enhance optimization of the search performance of GA for NSPs. First a simplified and later a more useful version of the problem are examined. The existence of the global optimum for the simplified version of the problem is known. However, the existence of the global optimum for the later version of the problem is unknown. Therefore, for the later version of the problem the objective is to acquire solutions as close to the Pareto efficient front as possible. So far the best results have been acquired by applying a simple and cost effective block-wise mutation operator that we call an escape operator. Throughout computer simulations, the aforementioned difficulties are analyzed and the efficiency of the so called escape operator is confirmed.