Hyperheuristics for managing a large collection of low level heuristics to schedule personnel

We investigate the performance of several hyperheuristics applied to a real-world personnel-scheduling problem. A hyperheuristic is a high-level search method which manages the choice of low level heuristics, making it a robust and easy to implement approach for complex real-world problems. We need only to develop new low level heuristics and objective functions to apply a hyperheuristic to an entirely new problem. Although hyperheuristic methods require limited problem-specific information, their performance for a particular problem is determined to a great extent by the quality of low level heuristics used. We address the question of designing the set of low level heuristics for the problem under consideration. We construct a large set of low level heuristics by using a technique which allows us to "multiply" partial low level heuristics. We apply hyperheuristic methods to a trainer scheduling problem using commercial data from a large financial institution. The results of the experiments show that simple hyperheuristic approaches can successfully tackle a complex real-world problem provided that low level heuristics are carefully selected to treat various constraints. We also examine how the choice of different sets of low level heuristics affects solution quality.