Solution of the Fracture Detection Problem by Machine Learning Methods

Abstract

Inverse problems of fracture exploration seismology are solved using machine learning methods. A single fracture of fixed size and subvertical orientation is considered in the two-dimensional case. The spatial position and the inclination angle of the fracture are determined using a neural network. The training set consists of solutions of direct problems produced by the grid-characteristic method on regular rectangular meshes in the form of synthetic seismograms obtained by measuring the vertical velocity on the surface of the medium.