极限学习机驱动的地震多属性融合识别曲流带单一点坝

识别高弯度曲流河内部单一点坝对于研究曲流河演化特征、指导油气藏开发等具有重要意义。为此，以孤东油田馆陶组曲流河为例，综合地震属性聚类分析与优选、极限学习机算法属性融合和地震正演模拟等，构建一种利用极限学习机算法驱动的地震多属性融合技术，识别复杂曲流带内部单一点坝。钻井和开发动态检验表明：①该方法在实例区的砂岩厚度预测单井吻合率达到93.3%，优于支持向量机和BP神经网络方法； ②研究区曲流带内部发育三种点坝叠加模式，即相向迁移点坝叠置、同向迁移废弃河道残留和同向迁移废弃河道无残留，三者地震相在反射连续性和振幅强度方面存在差异； ③研究区曲流带内部发育有5个单一点坝，点坝间残留的废弃河道形成渗流屏障影响剩余油开发。该方法研究成果可为油藏开发提供地质依据。

Single point bar interpretation in meandering belt with extreme learning machine driven multiple seismic attributes fusion

The identification of single point bars in high-curvature meandering rivers is of great significance for understanding the evolution and characteristics of meandering rivers and guiding the oilfield deve-lopment. With the meandering river of Guantao Formation in Gudong oilfield as an example, a fusion technology of multiple seismic attributes with the extreme learning machine algorithm is constructed to recognize single point bars in complex meandering belts. The technology integrates the clustering analysis of seismic attributes, seismic attributes fusion by the extreme learning machine algorithm, and seismic forward modeling. Through the drilling and dynamic verification, the following results can be obtained. 1) The method can improve the prediction accuracy of sandstone thickness with the coincidence rate of a single well reaching 93.3% which is higher than that of SVM and BP neural network. 2) Three combination models for the point bar in the meandering belt are studied, including the migration pattern of point bars in the opposite direction, the migration patterns with and without abandoned channels in the same direction. The three differ in reflection continuity and amplitude. 3) There are five single point bars in the study area and the abandoned channels between adjacent point bars form seepage barriers influencing the remaining oil development. The methods and results provide direct geological support for oil development.