Using hyper-heuristics for evaluation of operators in evolutionary algorithms

Razvoj na področju evolucijski algoritmov je še vedno v porastu in ni zaznati, da bi se kmalu umiril. Področje evolucijskih algoritmov se krepi z vedno novimi in boljšimi algoritmi iz dneva v dan. Praktična uporaba le teh se seli v težja in zahtevnejša okolja. Pričakovanja splošne in strokovne javnosti so vse večja. Zadnja družina algoritmov je znana pod imenom hiper-hevristika. Za to skupino algoritmov je značilna sočasna uporaba večjega števila algoritmov pri reševanju enega problema. Cilj je združiti znanja večih algoritmov v eno povezano celoto. Predlagana in razvita je nova vrsta orodja. Razvito orodje omogoča razvoj algoritmov po novih smernicah. Algoritmi razviti po predlaganih usmeritvah so preglednejši, kompaktnejši, poenoteni, prenosljivi, razširljivi. Te odlike lahko pričakujemo od novo nastalih algoritmov v bližnji prihodnosti. Glavni doprinos orodja je neodvisno zaganjanje posameznih delov evolucijskega al- goritma. Obstoječi evolucijski algoritmi so preoblikovani v posamezne enote. Orodje skrbi za vrstni red in trajanje zagona za vsako enoto posebej. S povezovanjem različnih delov, dobljenih iz različnih algoritmov, pridobivamo nove algoritme. Glede na uspešnost algoritma, lahko sklepamo o uspešnosti njegovih enot. Development in the field of evolutionary algorithms is still on the rise and it does not seems to calm down soon. Field of evolutionary algorithms is growing with new and better algorithms on daily basis. The practical application of new algorithms are used in the difficult and demanding environments. Expectations of general and professional public are high. Latest family of algorithms is known as hyper-heuristics. Characteristics for this group of algorithms is the simultaneous use of a large number of algorithms for solving single problem. The aim is to combine knowledge of multiple algorithms into one cohesive environment. Proposed and developed is new type of tool. The developed tool enables the de- velopment of algorithms with new guidelines. Algorithms developed by the proposed guidelines are transparent, compact, standardized, portable, extensible. These qualities can be expected from newly created algorithms in a near future. The main contribution of the tool is independent execution of the individual parts of the evolutionary algorithm. Existing evolutionary algorithms are transformed into individual units. The tool takes care of the order and runtime for each unit. By linking different parts of different algorithms, new algorithms are created. Depending on the performance of the algorithm, we can assume the performance of its units.