摘要
地震属性蕴含大量储层信息,融合多种地震属性可提高储层预测精度。由于地下地质结构复杂、非均质性强,依据单一的地震属性融合方法难以精细刻画储层特征。为此,提出了一种基于Stacking集成学习的分频地震属性融合储层预测方法。该方法主要包括3个部分:①根据不同厚度储层的振幅与频率关系,利用多个频率的地震信息,降低地震属性的多解性;②联合相关性分析和无监督聚类技术优选地震属性,剔除冗余属性特征;③利用能够综合多个差异化模型优势的Stacking集成学习模型,融合不同频段的地震属性,提高地震属性的解释精度。将该方法用于渤海湾盆地埕岛油田,并使用线性公式定量分析法进一步评估Stacking模型的泛化效果。结果显示:与单类预测模型相比,Stacking模型的综合预测性能和可靠性均有显著提升;对应的地震属性融合结果高值区形态更加清晰,融合属性与砂体厚度的相关系数可达到0.92,这表明该方法具有良好的应用前景。
Seismic attributes contain a large amount of reservoir information,and the fusion of multiple seismic attributes can improve the precision of reservoir prediction.However,due to the complex underground geologi⁃cal structure and strong heterogeneity,a single fusion method of seismic attribute is difficult to describe the res⁃ervoir characteristics.Therefore,a reservoir prediction method of fusing frequency-decomposed seismic attri⁃butes using Stacking ensemble learning is proposed.The method consists of three main parts:①Based on the amplitude and frequency relationship of the reservoir with different sand thicknesses,seismic data with varying frequencies and bandwidths are considered to reduce the uncertainties of seismic attribute interpretation.②Jointly preferring seismic attributes based on correlation analysis and unsupervised clustering is performed to eliminate redundant information of seismic attributes.③A Stacking ensemble learning model,which can com⁃bine the advantages of various models,is designed to fuse seismic attributes with different frequencies and band⁃widths for improving seismic attribute interpretation resolution.The proposed method is applied by taking the Chengdao Oilfield in the Bohai Bay Basin as an example.The quantitative linear formula analysis method is pro⁃posed to further evaluate the generalization effect of the Stacking model.The results prove that comprehensive prediction accuracy and the reliability of the Stacking model are significantly improved compared with those of the single-class models.The corresponding high-value areas of the fusion attribute are more evident,and the correlation coefficient between the fusion attribute and sand thickness can reach 0.92,indicating that the method has great potential in reservoir prediction.
作者
刘磊
李伟
杜玉山
岳大力
张雪婷
侯加根
LIU Lei;LI Wei;DU Yushan;YUE Dali;ZHANG Xueting;HOU Jiagen(College of Artificial Intelligence,China University of Petroleum(Beijing),Beijing 102249,China;National Key Labora-tory of Petroleum Resources and Engineering,China University of Petroleum(Beijing),Beijing 102249,China;College of Geosciences,China University of Petroleum(Beijing),Beijing 102249,China;Shengli Oil Field Company,SINOPEC,Dongying,Shandong 257015,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2024年第1期12-22,共11页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“坳陷湖盆洪水型湖相重力流沉积演化机理及差异构型模式”(42272186)
“坡度与水深主控的河流辫—曲转换机理及其沉积响应”(42202109)
中国石油战略合作专项“鄂尔多斯盆地致密油—页岩油储层非均质成因机理与表征技术”(ZLZX2020-02)
北京市科协青年人才托举工程(BYESS2023460)
博士后创新人才支持计划项目“少井条件下沉积过程与地震正演驱动的地震智能反演方法”(BX20220351)联合资助。
关键词
地震属性
储层预测
STACKING
集成学习
分频
智能融合
seismic attribute
reservoir prediction
Stacking ensemble learning
frequency-decomposed
intelli⁃gent fusion