基于全卷积神经网络的走滑断裂识别技术——以托甫台地区为例

The technology for identifying strike-slip faults based on fully convolutional neural network:Taking Tuoputai Area as an example

对断裂系统进行准确识别是地震资料解释中的重要内容,解释效率及好坏直接影响着油气勘探开发工作的进展。针对传统相干体、曲率等属性在复杂的走滑断裂发育区域分辨率低、识别能力差等缺点,提出了一种基于全卷积神经网络的走滑断裂识别方法:首先通过调整参数生成大量不同类型的三维合成断层样本用于训练全卷积神经网络,并对比U-Net与SegNet这2种全卷积神经网络的断裂识别效果,优选出训练网络SegNet;然后利用构造导向滤波方法对实际地震数据进行断裂增强处理,目的是进一步清晰地刻画走滑断裂,同时提高地震资料的信噪比,使网络能更好地学习到实际断裂的特征。在塔河油田托甫台地区的走滑断裂识别实践表明,无论从分辨率、抗噪性还是连续性,该方法都优于传统的单一属性的断裂识别方法,并且可以实现快速精准的断裂识别,为实际生产提供有效指导。

Accurate identification of fault systems is an important part in seismic data interpretation,and the efficiency and quality of the interpretation directly affect the progress of oil and gas exploration and development.To overcome the shortcomings of low resolution and poor recognition ability of traditional coherence and curvature attributes in the complex strike-slip fault development area,a method for identifying strike-slip faults based on fully convolutional neural network was proposed.Firstly,a large number of different types of 3 D synthetic fault samples were generated by adjusting parameters for training the fully convolutional neural network,and the training network SegNet was preferred by comparing the fault recognition effects of U-Net and SegNet.Then the structure-oriented filtering method was used to enhance the actual seismic data,aiming to further characterize the strike-slip faults and improve the signal-to-noise ratio of the seismic data,so that the network can better learn the characteristics of the actual faults.The practice of strike-slip fault identification in the Tuoputai Area of Tahe Oilfield shows that the proposed method is better than the traditional single-attribute fault identification method in terms of resolution,noise immunity and continuity,and the technique can achieve fast and accurate fault identification and provide effective guidance for practical production.