| Author's Name: AN Hongyi1, WEN Xin1, LI Juzheng1, ZHANG Jingzhe1, ZHANG Linzhi1,
FANG Pingchao1, DU Tianwei2, ZHANG Kui3, WANG Qunwu3 |
| Institution: 1. Exploration Department, PetroChina Southwest Oil & Gasfield Company；
2. Kunlun Digital Technology Co., Ltd；3. Beijing Precise Energy Technology Co., Ltd |
| Abstract: |
| Fault interpretation is one of the core tasks in oil and gas exploration and development. However, with the
increase of exploration scale, traditional artificial fault interpretation and conventional fault detection methods are unable
to meet practical needs. Deep learning methods provide an important approach for intelligent seismic fault recognition,
among which deep network models represented by Unet have achieved many successful cases in this type of task.
However, due to the particularity of convolution operations, this method loses some information in the feature extraction
process, resulting in the need for further improvement in the accuracy and robustness of fault recognition. In this paper, we
design a CNN-Transformer hybrid module and embed it into the Unet network framework, proposing a hybrid network
model based on U-CNNformer. The hybrid network model improves the mining ability of both global features and local
details in the sample set, overcomes the limitations of the conventional Unet network in weak information correlation in
fault recognition, and improves the robustness of the model while ensuring the accuracy of fault recognition. Testing on
the publicly available North Sea F3 data and applying with actual data in a certain area of Sichuan Basin in China
demonstrate that the proposed hybrid network model not only accurately detects fault features but also provides a more
detailed characterization of fault distribution, achieving high-precision intelligent fault recognition with excellent
application effectiveness. |
| Keywords: fault recognition; deep learning; Unet; CNN; Transformer; model training; data test |
|
投稿时间：
2025-07-01 |
|
