A system for mapping natural filtration channels based on seismic data and machine learning methods

Abstract

In this paper the system for solving the problem of interpretation of fracture zones was proposed, using the analysis of seismic sections on two-dimensional images based on of deep machine learning techniques. The proposed system is based on the filtering channels mapping on seismic data, the algorithm for selecting features using an auto-encoder and the algorithm for synthesizing new images using on generative adversarial networks. Auto-encoders provide high information content with low intra-class and high inter-class variability. To train the neural network, we used a data sample based on seismic database with the location of previously mapped fracture zones in the volume of 70 sections of the wave field for oil and gas fields in the Yamalo-Nenets and Khanty-Mansi Autonomous Okrug. Computer simulation for the problem of identification of geological layers on two-dimensional sections of a seismocube is carried out, the obtained simulation results were presented and discussed.