An adaptive method with local search for nurse rostering problem

The nurse rostering problem (NRP) is a well known NP-hard problem. This paper presents an adaptive method with local search to generate schedules for the NRP. Nurses are scheduled one by one according to the order of them. During the construction of schedules for each nurse, the planning period is divided into several weeks. Valid weekly shift patterns for each week are generated in advance and then combined together to form one´s complete schedule. In order to find the best combination of these weekly shift patterns quickly and thus construct a schedule with low penalty value, the local search is developed. After all schedules for each nurse are constructed by the local search, they combine together to form a roster. A coverage repairing procedure is used to make the roster feasible and a greedy search is used to improve its quality while ensuring its feasibility. Then nurses are ordered adaptively based on the roster built and the new order is fed into the next iteration. The resulting rosters show that our proposed method can obtain good quality solutions in reasonable time.