Self-configuring genetic programming algorithm for solving symbolic regression problems

Abstract

In most cases, the only way to study and solve practical problems is to investigate them with the help of models. However, this method is also to be a complicated problem where a significant part of the effort is aimed at finding the functional dependencies between input and output variables. But a great number of methods of numerical identification solve problems by developing a model that, in fact, is a black box. The reduction different types of problems to the problem of symbolic regression allow us to overcome the lack of these methods. The algorithm of genetic programming is applied for solving the problems of symbolic regression. The given paper considers an algorithmic complex that includes a genetic algorithm and an algorithm for genetic programming for solving symbolic regression problems. The uniform crossover operator is applied for these methods; it ensures the flexibility of the algorithm due to the greater diversity of structures resulting from crossover. The opportunity for selection two or more parents for recombination is realized. To automate the selection of the algorithms parameters a self-configuring procedure at the population level is realized and the efficiency of its application is proved for test problems of symbolic regression. The practical implementation of algorithms for solving classification problems and differential equations is carried out.