Mapping System to Interprete a Fracture Zones in Seismic Data Based on an Autoencoder and Generative Adversarial Networks

Our research was aimed at developing methods and algorithms for identifying and mapping fracture zones of the Jurassic and pre-Jurassic complexes based on seismic data and methods of machine learning and artificial intelligence. In this paper it were developed and tested: the feature selection algorithm based on an autoencoder for interpreting data on seismic slices with low intraclass and high interclass variability, as well as the algorithm for synthesizing new images based on generative adversarial networks and a seismic database. The experiments carried out have shown that the algorithms have good accuracy and speed of operation, which makes them potentially suitable for processing large volumes of seismic information. The proposed autoencoder for candidate images with a high recall value both for the search variant with the micro-averaging metric and for the search with the macroaveraging metric. The automatic interpretation of seismic data on seismic slices of the Salymskaya well was carried out. The system for processing, analyzing and classifying 3D seismic information was developed based on the proposed methods and algorithms. The optimal predictive model was tuned based on machine learning and artificial intelligence methods.