Inverse Problems of Heterogeneous Geological Layers Exploration Seismology Solution by Methods of Machine Learning

This article is devoted to solving the inverse problems of exploration seismology of fractures and their uniformly oriented systems using convolutional neural networks. The use of convolutional neural networks is optimal due to the multidimensionality of the studied data object. A training sample was formed using mathematical modeling. In the numerical solution of direct problems, a grid-characteristic method with interpolation on unstructured triangular meshes was used to form a training sample. The grid-characteristic method most accurately describes the dynamic processes in exploration seismology problems, since it takes into consideration the nature of wave phenomena. The approach used makes it possible to construct correct computational algorithms at the boundaries and contact boundaries of the integrational domain. Fractures were set discretely in the integration domain in the form of boundaries and contact boundaries. The article presents the results of solving inverse problems for single fracture length, placement and orientation detection and for system of fractures with variations in the angle of inclination of fractures, height of fractures, density of fractures in the system.