Machine seismic: an automatic approach for the identification of subsurface structural models

Abstract

Subsurface structure identification and modeling in three-dimensional dataset is a very complex problem for the geoscientist. Manual dealing of seismic data involves different complexities and requires more time and effort in order to address the errors and mistakes associated with the quality of seismic data. It is sometimes not straightforward to classify the features of interest especially when the data are of poor quality or associated with complex micro-tectonic regime. This brings the need of automatic working engine which can handle such issues and can find the target-oriented features with the more accuracy. Therefore, advance machine learning approach has been considered as a powerful tool which extensively used in different fields of data science to enhance the quality of data and fast processing. In this study, a new technology has been introduced in the seismic processing and development to make processes in automated routine. This method is based on deep neural network (DNN) that directly translates the raw seismic data into final fault models. A test example of deep semi-supervised learning method was performed for seismic faults and reflector identification without prior information and without going into complete dataset for labeling. This example explains the challenges to the issue and gives the solution to the problem related to the structure and tectonic modeling to explain the performance gain. In this approach of semi-supervised learning, only one of any reflectors and one or two of the faults has been introduced in the learning phase to identify such features in the complete 3D cube. This is an example of deep learning neural network for machine seismic in real case study and allows exploration geoscientists to easier mark the hot spots of their interest before mining or drilling.