Seismic technologies

[SE-02] Shape Carving Methods of Geologic Body Interpretation from Seismic Data Based on Deep Learning

*Sergei Petrov¹, Tapan Mukerji¹, Xin Zhang², Xinfei Yan² (1. Stanford University (United States of America), 2. Research Institute of Petroleum Exploration & Development, PetroChina (China))

The conventional approach to building a discrete facies reservoir model with seismic data involves interpreting seismic data, building a 3D cell model of a reservoir, distributing properties in it with geostatistical algorithms, and eventually using these properties to obtain the prediction of reservoir performance. A problem with this approach is that assumptions and approximations are introduced at each stage. Another issue is a tradeoff between reproducing statistical information from existing hard data and obtaining geologically sound results when applying geostatistical algorithms. Overall, obtaining a meaningful and realistic result in the form of a 3D reservoir model using the conventional workflow is a challenging and time-consuming task. Machine learning tools can help to get around the seismic interpretation stage and to go directly from raw seismic data to reservoir characteristics of interest. Recently techniques involving convolutional neural networks have started to gain momentum. Convolutional neural networks are particularly efficient at pattern recognitions within images, and this is why they are suitable for the seismic facies classification task. As there are currently a variety of architectures available, we performed a comparative study by experimenting with three different architectures based on convolutional layers using different synthetic and field datasets. The comparison was made in terms of the quality of the seismic interpretation results and computational efficiency. The three architectures tested were a) 2D convolutional network with dilated convolution b) 3D convolutional network and c) a U-net architecture.