Machine learning derived AVO analysis on marine 3D seismic data over gas reservoirs near South Korea

Abstract

Amplitude variation with offset (AVO) analysis has become a crucial step in determining gas well locations. However, large amounts of time and effort are required to confirm AVO anomalies, and interpretations can be inconsistent. To avoid the need for the long processes involved in conventional manual analysis, we developed an automatic AVO analysis method for common midpoint (CMP) gathers through using machine learning (ML) with a convolutional neural network (CNN). To deal with complicated seismic data, the network was constructed based on VGG16 network architecture, which includes 16 layers. The resulting CNN-based algorithm was applied to two sets of three-dimensional (3D) seismic data acquired off the east coast of South Korea. One dataset, which was obtained over gas reservoirs and confirmed to show multiple AVO class III anomalies by comparison with geophysical well logging data, was used for training and evaluation of the proposed CNN-based algorithm. The resulting trained model was tested using the second dataset, which was obtained over an area near the gas reservoirs with a different depositional environment. To demonstrate the applicability of the model to raw and final migrated CMP gathers, AVO class III anomalies predicted by the ML-derived analysis were confirmed by manual AVO analysis of the test data.