基于VNet深度学习架构的低序级断层智能识别方法

低序级断层识别是油气勘探开发的重要环节，传统的相干体、谱分解、曲率、蚂蚁体、边缘检测等方法虽然能够提高断层识别的效果和精度，但是对断距较小的低序级断层识别效果不佳。基于人工智能技术的全卷积神经网络(FCN)深度学习方法，为识别低序级断层提供了新的途径。在UNet基础上提出的VNet模型深度学习架构，可以在上、下采样过程中增加信号的感受野，尽可能地在提取大尺度断层信息的同时保留和提取小尺度断层信息。选用正演模拟数据和实际地震数据分别对UNet模型、VNet模型进行测试，通过选择合适的损失函数、迭代次数，优选合适模型权重参数对两者进行模型训练和断层识别效果对比，结果表明，基于VNet模型方法提取的信息更丰富，在识别低序级断层方面更有效。

Intelligent recognition method of low-grade faults based on VNet deep learning architecture

The recognition of low-grade faults is an important link in oil and gas exploration and development. Coherent volume, spectral decomposition, curvature, aunt body, edge detection, and other traditional methods have greatly improved the effect and accuracy of fault recognition, but they cannot effectively recognize low-grade faults with small fault distances. However, as an artificial intelligence technology, the deep learning method based on a full convolution neural network provides a new way for low-grade fault recognition. Based on UNet, the proposed VNet deep learning architecture can increase the receptive field of signals during the up and down sampling, extract large-scale fault information as much as possible, yet retain and extract small-scale fault information at the same time. Furthermore, this paper uses forward modeling data and actual seismic data to test UNet and VNet models, selects appropriate loss function, iteration times, and model weight parameters to compare the effects of model training and fault recognition. The results show that the VNet-based method can extract rich information and is more effective in low-grade fault recognition.