| Abstract: |  Hyper-heuristics are an emerging that has received increasing attention in the last years. As they are black box optimization techniques that work on higher level of abstraction, they have many real world application. This work aims to explore the possibilities of application of evolutionary algorithms and related methods in the field of hyper-heuristics. Their properties make them a particularly promising candidates. They can explore large solution spaces and at the same time are very straightforward, which gives the user greater control and easier extensibility. An approach based on exploration of neighborhoods in solution spaces using evolution of low level heuristics. Five version in total were proposed: a baseline version and four more complex extensions. This aims to pin-point the most contributing extensions. All of these algorithm versions were evaluated using a benchmark framework HyFlex supplied by the organizers of CHeSC 2011 challenge, which aimed at finding an algorithm capable of solving the widest spectrum of optimization problems. The benchmark consisted of solving problems from 6 different domains, each containing 5 problem instances. All of the 4 extended version performed very well, placing between 6th and 4th place out of 21 competing algorithms. This confirmed that the proposed method is capable of finding results of quality comparable to current state-of-the-art methods. Even though no implementation capable of outperforming all other was found, it was proven that each extension contributed on different types of problems.
|
|---|