Using control add-ons based on an artificial neural network and the random forest algorithm to improve the efficiency of the genetic algorithm
Abstract
Using control add-ons based on an artificial neural network and the random forest algorithm to improve the efficiency of the genetic algorithm
Incoming article date: 07.10.2024The article provides a rationale for the hypothesis about the possibility of influencing changes in the destructive ability of genetic algorithm (GA) operators on the trajectory of population movement in the solution space directly during the operation of the evolutionary procedure for labor-intensive tasks. To solve this problem, it is proposed to use a control superstructure from an artificial neural network (ANN) or the "random forest" algorithm. The hypothesis is confirmed based on the results of computational experiments. This study presents the results obtained with calculations on CPU and CPU + GPGPU in a resource-intensive task of synthesizing dynamic simulation models of business processes using the mathematical apparatus of the Petri net theory (PN), and a comparison with the operation of GA without a control superstructure, GA and a control superstructure based on ANN of the RNN class, GA and the "random forest" algorithm. To simulate the operation of GA, ANN, the "random forest" algorithm, business process models, it is proposed to use a graph representation using various extensions of PN, examples of modeling the selected methods using the proposed mathematical apparatus are given. For the operation of the ANN and the random forest algorithm for recognizing the state of the GA population, a number of rules are proposed that allow the management of the solution synthesis process. Based on the computational experiments and their analysis, the strengths and weaknesses of using the proposed machine learning algorithms as a control superstructure are shown. The proposed hypothesis was confirmed based on the results of computational experiments.
Keywords: "Petri net, decision tree, random forest, machine learning, Petri net theory, bipartite directed graph, intelligent systems, evolutionary algorithms, decision support systems, mathematical modeling, graph theory, simulation modeling