Stability Study of a Neuro-Fuzzy Output System Based on Ratio Area Method

Abstract

Purpose of research is to study the hypothesis about the possibility of changing the type of transition process during training in a neuro-fuzzy inference system based on area ratio method, and to study the properties of weight coefficient influence on its stability.Methods. An apparatus of fuzzy logic is used for the development of a neuro-fuzzy output system. At the same time, input and output variables are described by triangular membership functions. Mamdani implication model was used in the compositional rule. A linear model of area ratio was used in defasification. The reverse error propagation method was used during training.Results. In experimental studies, it was found that the proposed neuro-fuzzy model based on area ratio method allows to change the type of transition process, namely, to transform oscillatory process into an aperiodic (monotonic) process. In experimental studies, it was also found that the stability of neuro-fuzzy output system is more influenced by the weight coefficient determined in calculating the total area of membership output functions. Thus, the obtained results prove: first, that the proposed neuro-odd output system ensures the transformation of transfer characteristics, and second, ensures its stability in a given range of weight coefficient characteristics.Conclusion: The architecture of an adaptive neuro-fuzzy output system based on a linear method of area ratio is described. A distinctive feature of the proposed architecture is the use of an odd system of triangular accessory functions at inputs and outputs. Analysis of the simulation process of its training showed that it s important to ensure stability during training. It is also necessary to establish permissible values of the weight coefficient, numerical values of which in its turn affect the transformation of transfer characteristics of a neuro-fuzzy output system.