Choosing the Fittest Subset of Low Level Heuristics in a Hyperheuristic Framework

A hyperheuristic is a high level procedure which searches over a space of low level heuristics rather than directly over the space of problem solutions. The sequence of low level heuristics, applied in an order which is intelligently determined by the hyperheuristic, form a solution method for the problem. In this paper, we consider a hyperheuristic-based methodology where a large set of low level heuristics is constructed by combining simple selection rules. Given sufficient time, this approach is able to achieve high quality results for a real-world personnel scheduling problem. However, some low level heuristics in the set do not make valuable contributions to the search and only slow down the solution process. We introduce learning strategies into hyperheuristics in order to select a fit subset of low level heuristics tailored to a particular problem instance. We compare a range of selection approaches applied to a varied collection of real-world personnel scheduling problem instances.