期刊
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
卷 170, 期 -, 页码 315-330出版社
ELSEVIER
DOI: 10.1016/j.petrol.2018.06.038
关键词
Borehole; Image; Logs; Deep; Network; Stacked; Autoencoder; Permeability; Porosity; Estimation
资金
- CENPES/PETROBRAS [5850.0106789.18.9]
- CBPF [5850.0106789.18.9]
- GPU card
Rock permeability and porosity are some of the most important features to be determined during the exploitation of a certain hydrocarbon reservoir, as these physical properties help petrophysicists understand the most likely distribution and/or presence of oil/gas reservoirs, as well as decide whether that certain field might be exploitable or not. Despite their importance, in order to obtain permeability and porosity values it is usually required to run tests on borehole core samples that are either destructives or that need to be done under very controlled conditions, which makes them very expensive and time consuming. In the last decades, some authors have been able to estimate both these properties measurements from well-logging reservoir curves - such as Gamma-Ray response (GR), Bulk Density (RHOB) or Neutron Porosity (NPHI)-. In this study, we present a novel method that can be used to estimate formation permeability and porosity from borehole image logs - instead of simple logs-using deep autoencoders. The data used in this work belongs was extracted from a Brazilian pre-salt carbonate well. First, we use a 3-level stacked autoencoder to extract internal features from both ultrasonic and microresistivity images, thus characterizing the features contained inside these images. Afterwards, we use a support vector machine regressor (SVM) to link the encoded features extracted from the autoencoders for each type of data (ultrasonic and microresistivity borehole images) to the petrophysical measurements logs. Using this method we were able to achieve a correlation coefficient between original and estimated validation samples of R-2 = 96.30% for effective porosity logs and R-2 = 96.06% for permeability logs, and a normalized squared mean error of NMRSE = 7.30% and NMRSE = 5.51%, respectively, all results in the blind test sample.
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