Projects & Grants
| Study of influence of the evolutionary algorithms parameters change in dependence on individuals distribution in the search space | |
|---|---|
| Project Id | SGS06/ÚVAFM/2018 |
| Main solver | doc. RNDr. Petr Bujok, Ph.D. |
| Period | 1/2018 - 12/2018 |
| Provider | Specifický VŠ výzkum |
| State | finished |
| Anotation | In many fields of science, research and commercial sphere arises tasks need to be optimized. In the case of hard problems, as has been demonstrated, the techniques based on nature systems - evolutionary algorithms (EA) have been successfully applied. Although these stochastic heuristics are robust tools they are often unable to find the sufficiently precise solution of tasks described by a hard objective function or defined in a high dimension space. The efficiency of EA depends on the setting of their control parameters that need to be set depending on the task being solved. Appropriate adaptation of parameters can improve EA efficiency.In order to adaptation of the parameters to be appropriate, it is necessary to react to certain changes in the population of individuals. One of the fundamental features of the population is the size of its convex hull which can be understood as a degree of individuals? distribution. By studying this parameter we can identify the certain phase of the population and adapt the parameters of the algorithm to it.In terms of the number of extreme occurrences, objective functions can be divided into unimodal and multimodal. In the case of a search for minimal of the multimodal functions algorithms often get stuck at local extrema. Consecutive identification of local extrema areas and their separation from further search is promising to achieve better results.Evolutionary algorithms developed in recent years achieved good results even in the case of solution the artificially-developed tasks. The main goal of the evolutionary algorithms? development should be their application to real problems arising from the human world. |
