Performance analysis of hyperheuristics and their use with hill-climbers

Hyperheuristics are iterative approaches that are proposed as a higher level abstractionas compared to the metaheuristics. Hyperheuristic methods manage a set of heuristics forsolving a problem. A typical iteration in a hyperheuristic framework consists of two phases:heuristic selection and move acceptance. After the selection and application of an appropriateheuristic to a single candidate solution at hand, a decision is made whether to keep the newcandidate solution, or not. This decision is based on only nonproblem-specific data, such as,fitness change or heuristic execution time. In this thesis, the traditional framework is extendedand three new frameworks are proposed in order to make better use of hill-climbers. Theseframeworks and several heuristic selection method and acceptance criterion combinations areanalyzed in depth. Their performances are evaluated on well-known benchmark functions.The performance variances of the hyperheuristics are further investigated on the examinationtimetabling benchmark problem instances.