基于SA-VNet卷积神经网络的低序级断层识别方法
Low-order fault identification method based on SA-VNet convolutional neural network
张陈强 1贺锡雷 1谌洪平 1秦思萍 1张祖豪 1周云秋1
作者信息
- 1. 成都理工大学地球勘探与信息技术教育部重点实验室,成都 610059
- 折叠
摘要
在地震构造解释中,断层识别起着十分重要的作用,是寻找油气有利区域的基础和关键.在受走滑断层影响的区域,低序级断层发育,导致区域构造复杂.常规的断层识别方法和传统的卷积神经网络对低序级断层的识别效果较差,这会对复杂地区的油气藏勘探开发造成严重的影响.为了克服这一问题,并且提高训练效率,在V-Net网络的基础上加入了空间注意力机制(Spatial Attention)和动态调整学习率,提出了SA-VNet网络的断层识别方法.该方法是采用图像处理领域的语义分割技术,通过判断输入数据体的每个数据点为断层的概率,得到断层概率体.使用动态调整学习率能实现用更少的训练轮次得到更好的训练效果.通过理论模型验证了提出方法的可行性.随后,将SA-VNet 网络应用于实际工区,结果表明该方法对低序级断层的识别能力更强,准确率更高.
Abstract
Fault identification is a crucial step in seismic structure interpretation,serving as the basis for locating promising areas for oil and gas exploration.In strike-slip fault regions,low-order faults can develop,which are difficult to identify using conventional fault identification methods and traditional convolutional neural networks,posing significant challenges for oil and gas exploration and development in complex areas.In order to overcome this issue and improve training efficiency,we propose a novel fault identification method,SA-VNet,which incorporates the Spatial Attention mechanism into the V-Net network.The proposed method utilizes semantic segmentation technology to classify each data point of the input data body as either fault or non-fault,generating a probability body of the fault distribution.We introduce dynamic learning rate adjustment to improve the training efficiency and achieve superior training outcomes with fewer rounds.The feasibility of the proposed method is verified through theoretical modeling,and we further demonstrate its effectiveness in identifying low order faults in actual working areas.
关键词
断层识别/低序级断层/卷积神经网络/SA-VNetKey words
Fault identification/Low order fault/Convolutional neural network/SA-VNet引用本文复制引用
出版年
2024
NETL
NSTL
万方数据