AN APPROACH TO SELECT AN ARTIFICIAL NEURAL NETWORK ARCHITECTURE TO APPROXIMATE THE DEPENDENCE OF SURFACE WAVE PHASE VELOCITY ON AN EARTH ELASTIC MODEL

The paper is dedicated to the development of an approach for tuning the architecture and hyperparameters of a fully connected multilayer artificial neural network used to solve the problem of restoring one-dimensional velocity models of the shear wave of the upper part of the geological section by inverting the dispersion curves of the frequency dependence of the phase velocities of the surface wave. The approach is developed using a statistical analysis of errors distributions in determining the parameters of one-dimensional velocity models of the earth, constructed from the results of numerical studies. The developed approach is universal and can be used to solve similar issues.