期刊
PETROLEUM SCIENCE AND TECHNOLOGY
卷 41, 期 24, 页码 2299-2327出版社
TAYLOR & FRANCIS INC
DOI: 10.1080/10916466.2022.2117380
关键词
artificial neural network (ANN); coherency; curvature; dip-steering; energy; multilayer perceptron (MLP); seismic attributes; similarity; supervised and unsupervised learning; unsupervised vector quantizer (UVQ)
This study used a combination of a multilayer perceptron and an unsupervised vector quantizer to analyze 3D seismic volume and identify faults and fractures. By integrating the most relevant attributes, more precise and trustworthy results were obtained.
Faults and fractures play a significant role in oilfield drilling operations, hydrocarbon trapping, and reservoir development. Exploring faults quickly and accurately can help reach the target more efficiently. In this study, applicable seismic attributes were combined using a multilayer perceptron and an unsupervised vector quantizer and applied to a 3D seismic cube to identify discontinuities. First, high-probability faulted areas were picked manually on a seismic section as an input pattern for the MLP. Then, particular seismic attributes (dip-steering, similarity, coherency, curvature, instantaneous) were applied to the data. Consequently, the MLP and UVQ were used to determine the most contributed attributes. Using the MLP and UVQ, the most relevant attributes were integrated to find faults and fractures in the 3D seismic volume. In contrast to some fault-identifying methods and prior studies, this study used not just steered attributes but also compared supervised and unsupervised neural networks. Eventually, comparing the outputs of the MLP, faulted and non-faulted cubes, with the initial seismic section and the UVQ's output revealed discrepancies. For a specific set of attributes, the MLP was obviously superior to the UVQ in terms of creating detailed outputs, analyzing time, and rendering more precise and trustworthy results.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据