DEVELOPMENT OF A DEVICE FOR ARC WELDING WITH CONTROLLED MECHANICAL TRANSFER USING A METAL-CERAMIC STRIP ELECTRODE

networks are: ANN are well suited to solving complex multifactorial problems that may not have a clear explanation within physical models; An increasing number of interpolation problems does not have a direct algorithmic solution, but a set of points for which the measured values are known; ANN can be used to solve similar problems and help to identify relationships that are geologically relevant. The use of ANN for the approximation of nonlinear dependencies is particularly attractive in the context of geophysical data analysis. Forecasting of petrophysical quantities and restoration of geological and geophysical regularities is carried out in the interwell space. It is best to combine several types of neural networks for this. For example, generalized regression neural network (GRNN) and multilayered neural network (MNN), which perform a complementary approximation (since the MNN allows us to construct a global approximation of an unknown function with some extrapolation, and the GRNN performs a local approximation with some boundaries). ANN are an alternative to linear models based on the multiple linear regression method to solve petrophysical prediction problems based on seismic data and geophysical well exploration. Conclusions. In the process of the second stage of research, the efficiency of applying the algorithm of estimation of significance with the help of a neural network is shown to solve the problem of choosing informative seismic attributes. It has been proven that this method can be effectively applied to reduce the number of seismic attributes without losing useful information.