摘要
致密砂岩储层具有低孔低渗特点,由孔隙度变化引起的弹性参数和地震响应特征变化较弱,为解决常规的基于岩石物理关系线性映射孔隙度预测或者基于多属性融合的概率映射孔隙度预测方法难以表征孔隙度与地震属性间复杂非线性关系的问题,提出了基于深度前馈神经网络的孔隙度预测方法。该方法首先以测井计算的有效孔隙度曲线作为训练目标,以井旁的地震数据属性和反演弹性属性作为训练特征构成训练样本;其次通过优选评价确定复杂结构深度前馈神经网络模型参数,建立井旁地震数据与孔隙度之间的非线性映射关系;最后将训练优良的深度网络模型应用到整个数据体,得到有效孔隙度预测成果,进而实现致密砂岩优质储层定量表征。松辽盆地北部三角洲前缘沉积的致密砂岩应用实例表明,基于深度学习的孔隙度预测结果与井资料吻合较好,相对误差为8.1%,较常规基于岩石物理关系的线性映射孔隙度预测方法误差减小8.2%;证明了该方法对致密砂岩储层孔隙度预测的有效性。研究成果可为井位部署及方案优化设计提供理论指导与技术参考。
Tight sandstone reservoirs have low porosity and low permeability,with less variation of elastic parame-ters and seismic response characteristics caused by porosity variation.Conventional porosity prediction methods of linear mapping based on rock physical relationship or probabilistic mapping based on co-simulation have difficulty to determine complex nonlinear relationship between porosity and seismic attributes.Therefore,a porosity predic-tion method based on deep feedforward neural network is proposed.Firstly,taking effective porosity curve calculat-ed by logging as training target,near-well seismic data attribute and inversion elastic attribute are used as training characteristics to form training samples in this method.Secondly,the parameters of deep feedforward neural net-work model of complex structure are determined through optimization evaluation,and nonlinear mapping relation-ship between near-well seismic data and porosity is established.Finally,the well-trained deep network model is ap-plied to the whole data volume to obtain effective porosity prediction results so as to achieve quantitative character-ization of high-quality tight sandstone reservoirs.Case of tight sandstone deposited in delta front of northern Songliao Basin shows good correspondence between deep learning-based porosity prediction and well data,with rel-ative error of 8.1%,decreasing by 8.2% of conventional probability mapping porosity prediction based on co-simula-tion,and proving effectiveness of this method for porosity prediction of tight sandstone reservoir.The research pro-vides theoretical guidance and technical reference for optimization design of well deployment.
作者
李奎周
王团
赵海波
唐晓花
田得光
郑绪瑭
高天宇
LI Kuizhou;WANG Tuan;ZHAO Haibo;TANG Xiaohua;TIAN Deguang;ZHENG Xutang;GAO Tianyu(Exploration and Development Research Institute of PetroChina Daqing Oilfield Co Ltd,Daqing 163712,China;School of Earth Sciences,Northeast Petroleum University,Daqing 163318,China;Heilongjiang Provincial Key Laboratory of Continental Shale Oil,Daqing 163712,China)
出处
《大庆石油地质与开发》
CAS
北大核心
2023年第5期140-146,共7页
Petroleum Geology & Oilfield Development in Daqing
基金
中央引导地方科技发展专项“黑龙江省陆相页岩油重点实验室建设项目”(ZY20B13)。
关键词
致密砂岩储层
孔隙度预测
深度前馈神经网络
非线性映射
tight sandstone reservoir
porosity prediction
deep feedforward neural network
nonlinear mapping
