Neuro-fuzzy identification of nonlinear dependencies

Abstract

The paper proposes a fuzzy multilayer perceptron (MLP) and a modified algorithm for its training for solving problems of identification of nonlinear dependencies. The obtained results show a sharp reduction in the search for the optimal parameters of the neuro-fuzzy model compared to classical MLP and increase its accuracy. In the work, questions of optimization of the rule base of the neuro-fuzzy model are also investigated and the temporal and spatial complexity of the proposed algorithm is analyzed. The results of computational experiments show that the number of training epochs has sharply decreased, and productivity has increased compared to the well-known MLP models.