Imaging of steel casing's conductivity using surface electrical data and a deep learning approach

Li，Yinchu; Yang，Dikun

10.1190/segam2020-3426575.1

2020

1052-3812

1949-4645

SEG Technical Program Expanded Abstracts

2020-October

636-640

It is critical to ensure wellbore integrity in oil and gas fields for the purpose of reducing the risks associated with fluids leakage in resource extraction, environmental damage, and waste disposal. By energizing the wellhead and using steelcased wells as long electrodes (LE), electrical methods show the obvious ability to enhance the sensitivity of surface data to the deep anomaly. The conductivity value of steel casing is very high, which is usually reasonably assumed, but in practice may not be correct. Moreover, the damaged part of the casing can be equivalent to locally reduced conductivity in the conductivity distribution of the entire casing. Therefore, in this work, we introduce deep learning (DL) technique in the form of a fully convolutional network (FCN) to reconstruct 1D integrated conductivity distribution along a well casing. A number of synthetic data sets are generated by a fast 3D dc forward modeling algorithm using an equivalent resistor network (RESnet) where the casing is treated as a line-like object with a variable cross-sectional area integrated conductivity. We collect surface timedifferential electric field data on a survey line across the wellhead as the input for our neural network. After the training phase, the neural network can transform new sets of surface data to 1D casing conductivity models, which reflects the damage or corrosion along the well. Our synthetic results show the promising potential of combining surfaced LE electrical methods and DL as a non-invasive technology in the on-site evaluation and real-time monitoring of casing integrity conditions. This data-driven inversion can also be used as a tool to recover the casing conductivity, which has not been studied as a geophysical inversion target. Presentation Date: Wednesday, October 14, 2020 Session Start Time: 9:20 AM Presentation Time: 11:25 AM Location: Poster Station 7 Presentation Type: Poster.

Electrical/resistivity Inversion Machine learning Time-lapse Wells

第一 ; 通讯

英语

EI

20214611177969

Boreholes ; Deep learning ; Gas industry ; Oil wells ; Steel corrosion ; Three dimensional computer graphics ; Wellheads

Ergonomics and Human Factors Engineering:461.4 ; Oil Fields:512.1.1 ; Gas Fuels:522 ; Metals Corrosion:539.1 ; Steel:545.3 ; Data Processing and Image Processing:723.2 ; Computer Applications:723.5

2-s2.0-85119081624

Scopus

Department of Earth and Space Sciences,Southern University of Science and Technology,China

Li，Yinchu,Yang，Dikun. Imaging of steel casing's conductivity using surface electrical data and a deep learning approach[C],2020:636-640.