| Popis: |
The article describes the methods of increasing the search capabilities of the genetic algorithm. The purpose of this work is to find ways to accelerate the work of genetic algorithms. The relevance of the work lies in the fact that today the increase in the search capabilities of genetic algorithms is the main problem when using such algorithms. Often, with manipulations with an algorithm, the probability of getting into the local extremum of the function under investigation increases. The method described in the paper is to change the point of view of the natural selection process. The classical genetic algorithm implements the action of natural selection (EO) at the level of individuals. However, in microbiology, natural selection is presented as a selection of genes; this viewpoint is not widely used in the theory of genetic algorithms. This paper presents an algorithm that implements natural selection at the level of genes. The measure of the fitness of a gene in its work is taken as its stability in the process of changing generations, which can be traced based on the Shewhart maps. The algorithm uses a set of fuzzy rules, with the help of which the dynamically changing parameters of the algorithm are controlled, in particular, the probability of hitting the next generation. Based on the conclusion that genes are statistically manageable, a prediction block has been implemented in the algorithm. To increase the speed of the algorithm, you can enter the internal prediction of the genome. The prediction decision is made on the basis of a fuzzy rule: if the time series of the i-th representative of the population is controlled according to Schuhart over the last L generations, then add an individual to the population whose genome consists of the predicted values of the genes for K generations ahead. Thus, the algorithm manages dynamically changing parameters (mutation, population size), and also predicts the most adapted genes based on third-party GA. The results are confirmed by an experiment conducted on test functions for optimization algorithms. On the basis of the conducted experiments, it can be concluded that the algorithm is practically applicable in search and optimization problems. |
