基于深度双向循环神经网络的储层孔隙度预测  

Reservoir porosity prediction based on deep bidirectional recurrent neural network

作　　者：王俊[1,2] 曹俊兴[1,2] 周欣[1,2] WANG Jun;CAO JunXing;ZHOU Xin(State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation of Chengdu University of Technology,Chengdu 610059,China;College of Geophysics,Chengdu University of Technologyy Chengdu 610059,China)

机构地区：[1]油气藏地质及开发工程国家重点实验室(成都理工大学),成都610059 [2]成都理工大学地球物理学院,成都610059

出　　处：《地球物理学进展》2022年第1期267-274,共8页Progress in Geophysics

基　　金：国家自然科学基金重点项目(42030812,41430323)和国家自然科学基金面上项目(41974160)联合资助。

摘　　要：储层孔隙度是描述储层特征的重要参数之一,根据测井资料进行准确的孔隙度预测对于储层精细描述至关重要.为此,发展一种基于深度双向循环神经网络的储层孔隙度预测方法,并利用实际井数据验证其有效性和准确性.将测井数据看成纵向上具有联系的时序数据,利用双向循环网络建立测井数据与储层孔隙度之间的非线性映射关系,同时引入"丢弃"和"早停止"策略防止过拟合.研究结果表明,相较于多元线性回归方法和全连接深度神经网络,该方法不仅有效解决了孔隙度预测中的空间尺度问题,而且弥补了传统深度网络无法提供上下文信息的缺陷,提高了孔隙度预测的准确性和稳定性.Reservoir porosity is one of the critical parameters to describe reservoir characteristics. High-efficiency and low-cost porosity prediction based on logging data is very important for fine reservoir description. Therefore, a deep learning method based on the Deep Bidirectional Recurrent Neural Network(DBRNN) is proposed to perform the porosity prediction task, and actual data verify its validity and accuracy. From the perspective of sedimentary reservoir continuity, well log data are considered as vertical time sequences. DBRNN establishes the nonlinear mapping relationship between logging data and reservoir porosity. At the same time, Dropout and Early stopping strategies are introduced to avoid overfitting, and the Nadam optimization algorithm is used to optimize the network to obtain the optimal DBRNN model. The results show that, compared with the Multiple Linear Regression(MLR) method and fully connected Deep Neural Network(DNN), this method not only effectively solves the problem of spatial scale of porosity prediction but also makes up for the defect that the traditional deep network can not provide context information, and improves the accuracy and stability of porosity prediction.

关 键 词：人工智能 深度学习 双向循环神经网络 孔隙度预测 

分 类 号：P631[天文地球—地质矿产勘探]