Rerostering of nurses with intelligent agents and iterated local search

Abstract

ABSTRACTThe nurse rerostering problem is a special case rostering problem. Rerostering occurs when a disruption to a current nurse roster requires its reconstruction. This article presents a modified agent-based nurse rostering system that solves both the nurse rostering and rerostering problem. Similar to existing nurse rerostering methods, this agent system minimizes the differences between the initial roster and the reconstructed roster through the use of negotiations and iterated local search. This system differs from existing solutions because it seeks to reconstruct the roster so that it also minimizes the negative impact on nurse preferences. The agent-based system, called Competitive Nurse Rostering and Rerostering (CNRR), was tested on three sets of 30 random experiments which included over 200 schedule disruptions. CNRR found solutions for over 90% of all disruptions and over 98% of fixable disruptions. The generated solutions prevented any reductions to nurse preference satisfaction in over half our experimental runs. In 28% of our experiments, there was a significant nurse preference utility loss of five points or more. Five points of utility loss equates to an average of an 11% reduction in a nurse's preference satisfaction.