A self-organising hyper-heuristic framework

Hyper-heuristics can be thought of as heuristic management methodologies which are motivated by the goal of building a high level of generality in the scope of computational search methodologies [2] [7]. Most meta-heuristics (and other search methods) perform a search over the solution space directly, whereas, hyper-heuristics as high level strategies search a heuristic space. One of the hyper-heuristic frameworks makes use of multiple perturbative (improvement) heuristics in a single-point iterative search setting. At each step, one of the low level heuristics is selected and applied to the candidate solution at hand. The resulting move is either accepted or rejected based on some problem independent criteria. This study focuses on such perturbative hyper-heuristics. In [5], two types of low level heuristics are identified: mutational heuristics and hill climbers. A mutational heuristic perturbs a given candidate solution in a systematic way using a stochastic component without any expectation regarding the change in the quality of the solution. On the other hand, a hill climber returns a solution that has either better or equal quality with the input solution. [5] investigates four different types of perturbative hyper-heuristic frameworks that aim to make better use of multiple mutational and hill climbing heuristics as illustrated in Figure 1: FA, FB , FC and FD. In the type A, all low level heuristics are utilised simultaneously. The type B framework functions similarly. The only difference is that if the heuristic selection