Characterizing Subsurface Damage Zones From 3D Seismic Data Using Artificial Neural Network Approach

Abstract

Summary To further improve the quality and efficiency of subsurface fault zone image and study its geometry. Herein we adopted post-stack seismic data conditioning and a combination of seismic multi-attribute for producing a new hybrid attribute through a supervised multilayer perceptron (MLP) neural network in the Jurassic formation of Cai36 3D prospect located in the eastern part of the Junggar Basin. We first conditioned original seismic data by using the dip-steering cube extracted from the original seismic data. Secondly, we extracted conventional seismic attributes from the conditioned data sensitive to fault zone signatures. Thirdly, we selected a set of “picks” at a time slice representing the presence or absence of fault zones. Then we adopted the supervised MLP neural network to train over the selected seismic attributes extracted at the fault zone and non-fault zone positions. We obtained a new fault probability cube as new attributes. Finally, we analyzed a typical strike-slip fault zone using the new attributes. This study provides an effective way of fault zone imaging from seismic data and adds new insights into its geometry. Therefore, the workflows used here could be widely applied to other 3D surveys.