Replication Data for: Hybrid and Automated Machine Learning Approachesfor Oil Fields Development

Since seismic data interpretation is complex and requires expert knowledge, it is very relevant to conduct direct searches for hydrocarbon deposits using a seismic data-driven method. This issue has been widely discussed in the field of seismic exploration for many years. Direct detection of hydrocarbon deposits by registered amplitude anomalies is possible only when the recording equipment and systems obtain records in actual ("raw'') amplitudes. оne of the most important stages of interpretation of seismic survey data is the search for the contact boundary between two layers of different geological elements of a rock mass that have different physical properties. This boundary is usually called the horizon. In order to find such horizons, it is usually used to analyze seismic sections (in image format) of a seismic cube, with the help of an expert geologist. The purpose of this type of analysis is to search for patterns in the images of the seismic sections to highlight various geological elements of a rock mass. The problem of finding the contact boundary between 2 geological elements can be reduced to the problem of binary semantic segmentation of two-dimensional representations of seismic time slices. We have images of the seismic time slice and corresponding masks obtained with the help of experts. Masks contain information about the location of the contact boundary on a given seismic time slice.