Solving Nurse Rostering Optimization Problem using Reinforcement Learning - Simulated Annealing with Reheating Hyper-heuristics Algorithm

Nurse rostering is the process of scheduling hospital nurses for a certain period of time. This is a complex and interesting task that involves multiple perspectives, including those of the hospital, nurses, and patients. In addition, nurse scheduling must consider existing constraints to obtain a schedule that is close to optimal. Common constraints include the number of available nurses, their abilities, labor regulations, and hospital policies. The numerous constraints in nurse scheduling lead to a problem known as the Nurse Rostering Problem (NRP), which is difficult to solve conventionally and is considered an NP-hard problem. Many researchers have attempted to solve NRP computationally, as this approach reduces the time required to create nurse schedules. The less time needed to create nurse schedules, the more time can be allocated to improving medical services, which ultimately impacts the quality of healthcare in the hospital's surrounding area. Several approaches have been used to solve NRP, including hyper-heuristic methods, which are search methods used to solve optimization problems. The aim of this study is to employ a hyper-heuristic approach using Reinforcement Learning - Simulated Annealing with Reheating algorithm to solve the nurse scheduling problem in hospitals in Norway. The results show that the Reinforcement Learning - Simulated Annealing with Reheating algorithm can produce schedules with an 82% improvement in solutions. This algorithm also outperforms the comparison algorithms, with a 3% to 47% improvement over the Simple Random - Hill Climbing algorithm, a 2% to 12% improvement over the Reinforcement Learning - Hill Climbing algorithm, and a 0.7% to 7% improvement over the Reinforcement Learning - Simulated Annealing algorithm.