摘要
碳酸盐岩地层因其复杂的沉积成岩演化过程,岩石类型较为多样,同时受到孔隙及流体性质的影响,岩性测井响应特征变化较大,给岩性的识别带来了困难。针对传统方法对测井参数垂向结构特征以及岩性特征多样性表达能力不足的问题,引入胶囊网络以提升复杂碳酸盐岩岩性识别的效果。胶囊网络通过卷积结构提取测井参数的时序特征,并用胶囊向量表达特征的不变性和共变性,能够有效挖掘测井参数和岩性在空间序列结构特征之间的深度内在关系,从而构建高精度的岩性识别模型。以鄂尔多斯盆地苏里格气田东区W区块碳酸盐岩储层为例,开展胶囊网络在岩性识别中的应用研究。首先,基于对岩性信息敏感的6种测井参数划分训练集和测试集;其次,构建基于多尺度卷积和跳跃连接结构的胶囊网络模型。与传统机器学习方法和常规深度学习方法相比(K近邻、朴素贝叶斯、支持向量机、BP神经网络和卷积神经网络等),基于胶囊网络的岩性识别模型正确率达到96.65%,识别精度提高1.59%~32.06%。实验结果表明,胶囊网络能够有效地提取测井数据的时序特征和垂向结构特征,为复杂碳酸盐岩岩性识别提供一种新的思路。
The sedimentary and diagenetic evolution process of carbonate rock formations is complicated,and the rock types are relatively diverse. The logging response characteristics of rocks vary greatly,which brings difficulties to lithology identification. Traditional methods have insufficient ability to express the vertical structure characteristics of logging parameters and the diversity of lithological characteristics. We propose a deep learning model based on capsule network to improve the recognition effect of complex carbonate lithology. The capsule network extracts the time series characteristics of logging parameters through the convolution structure,and expresses the invariance and covariation of the features with the capsule vector. The model we proposed can effectively dig out the internal relationship between logging parameters and lithology in the spatial sequence structure characteristics,thereby construct a high-precision lithology recognition model. Taking the carbonate reservoir in block W in the eastern area of Sulige Gas Field as an example,the application research of capsule network in lithology identification is carried out. First,the training set and the test set are divided based on six logging parameters that are sensitive to lithology information. Secondly,build the capsule network model based on multi-scale convolution and jump connection structure. Compared with traditional machine learning methods and conventional deep learning methods(K Nearest Neighbors,Naive Bayes,Support Vector Machines,BP Neural Networks and Convolutional Neural Networks,etc.),the recognition accuracy of the lithology recognition model based on the capsule network reaches 96.65%. The recognition accuracy is improved by 1.59%-32.06%. The experimental results show that the capsule network can effectively extract the time series characteristics and vertical structure characteristics of the logging data,and provide a new idea for the identification of complex carbonate lithology.
作者
周恒
张春雷
张欣
武中原
马乔雨
ZHOU Heng;ZHANG Chun-lei;ZHANG Xin;WU Zhong-yuan;MA Qiao-yu(School of Science,China University of Geosciences(Beijing),Beijing 100083,China;Beijing Zhongdi Runde Petroleum Technology Co.,Ltd.,Beijing 100083,China;School of Statistics,Beijing Normal University,Beijing 100875,China)
出处
《天然气地球科学》
EI
CAS
CSCD
北大核心
2021年第5期685-694,共10页
Natural Gas Geoscience
基金
国家科技重大专项“鄂尔多斯盆地大型岩性地层油气藏勘探开发示范工程”(编号:2016ZX05050)资助.
关键词
岩性识别
胶囊网络
碳酸盐岩储层
测井参数
深度学习
Lithology identification
Capsule network
Carbonate reservoir
Logging parameters
Deep learning
