Application of genetic algorithm in problems of approximation of complex multidimensional dependencies and identification of parameters of theoretical models

Abstract

The article proposes a method for constructing an analytical approximation of n-dimensional data, based on the use of a genetic algorithm. A feature of the method is that the encoding of the search space is performed in the form of a parsing tree for an algebraic expression by the parser of the context-free grammar of the class LR (1). In addition, during the evolutionary process, in addition to the use of structure mutations (subject to their positive influence), the stage of mutation of the coefficients is performed, which allows avoiding the target function falling into local extremum. And also at each step of the evolutionary process, there is a stage for searching for an extremum in the space of coefficients and a stage for simplifying the analytical model.