APPLICATION OF NEURO-FUZZY NETWORKS FOR DETERMINATION OF RATIONAL PARAMETERS OF NUMERICAL SOLUTION OF SYSTEMS OF DIFFERENTIAL EQUATIONS

Abstract

Purpose. The purpose of the work is to improve the computational methods of calculating the systems of differential equations, which describe the accumulation of geometrical defects of structures, which function in an aggressive environment. Obtaining a numerical result with predetermined flexibility requires numerical integration parameters that would ensure the required accuracy. Methodology. The calculation costs of solving the problem of predicting the lon-gevity of corrosive structures are related to this system of differential equations. In cases where the problem of optimal design is solved, then the selection of optimal parameters of numerical procedures with control over the accuracy be-comes essential. To improve the efficiency of computational methods for this class of differential equation systems, the work is suggested to use elements of the computational intellect, which are neural networks. It is suggested to use an adaptive neural network for obtaining parameters of numerical integration on the basis of fuzzy logical inference, which allows formalizing the multidimensional data used for setting up a fuzzy system. Results. This approach allows using a neuro-fuzzy network for forecasting the durability timeframe taking into account the parameters that influence it. Orig-inality. The results of numerical experiments show that the adaptive neuro-fuzzy system after the training is able to summarize the input data and propose the parameters of numerical procedures, which ensure the required accuracy of the obtained result. Numerical experiments based on the comparison of reserved data and results of network operation prove that the adaptive system can be used to improve the efficiency of calculating methods when carrying out this type of task. References 21, tables 2, figures 2.