基于机器学习的复杂储层微小断裂系统识别方法研究与应用  

Research and application of micro-fault system detection based on machine learning

作　　者：杜炳毅 张广智[2] 王磊[1] 雍学善[1] 王腾飞 DU Bingyi;ZHANG Guangzhi;WANG Lei;YONG Xueshan;WANG Tengfei(Northwest Branch,Research Institute of Petroleum Exploration & Development,Petrochina,Lanzhou 730020,China;School of Geosciences,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]中国石油天然气股份有限公司勘探开发研究院西北分院,甘肃兰州730020 [2]中国石油大学(华东)地球科学与技术学院,山东青岛266580

出　　处：《石油物探》2021年第4期621-631,共11页Geophysical Prospecting For Petroleum

基　　金：国家油气重大专项(2016ZX05027004-001);国家自然科学基金项目(41674130)共同资助。

摘　　要：针对复杂断裂储层分布规律复杂、横向特征变化剧烈、微小断裂识别难的问题,结合各向异性岩石物理模型的微观特征和断裂敏感属性的宏观特征,开展了基于机器学习的复杂储层微小断裂系统特征预测方法研究。根据断裂系统的发育规律建立复杂储层各向异性岩石物理模型,预测井位处的各向异性参数,计算各向异性梯度并将其作为微小断裂系统指示因子;从叠后地震数据中提取构造类地震属性并进行优化处理,运用相关聚类算法优选微小断裂系统属性集;选取已知井点处的微小断裂系统属性集和微小断裂系统指示因子作为训练数据,利用机器学习(支持向量机)算法建立敏感属性集与微小断裂系统指示因子的非线性映射关系,实现复杂储层微小断裂系统特征的准确刻画。四川盆地某工区碳酸盐岩复杂断裂储层预测结果表明,预测的微小断裂系统发育规律符合研究工区的地质认识,并且与测井资料解释结果吻合程度较高,为复杂储层的微小断裂系统识别提供了有效的技术手段。The development of micro-fault systems is one of the main factors determining the existence of complex reservoirs.These types of reservoirs are characterized by a complex spatial distribution and pronounced heterogeneity.Moreover,in these reservoirs,the identification of micro-fault systems is difficult.Consequently,detection and comprehensive prediction are the main research directions in the exploration and development of complex reservoirs.In this study,a novel detection method for micro-fault systems based on machine learning was proposed.The method accounts for rock-physics anisotropy and the selection of sensitive attributes.First,an anisotropic rock physics model was built based on the complex reservoir properties.The curves of elasticity and anisotropy can be estimated from well data.The gradient of anisotropy is a parameter that is sensitive to micro-fault development.Subsequently,the seismic structural attributes were extracted from post-stack seismic data,and seismic optimization was carried out using a correlation clustering algorithm,which was used to select attributes sensitive to faults and fractures.Finally,the attribute set and micro-fault system index factor as the input data were applied to establish the non-linear mapping using a support vector machine,which is a type of machine learning algorithm.The micro-fault system index factor,which was used to describe properties of the micro-fault system,was estimated by nonlinear mapping.A test on actual data from a complex,fault-fractured carbonate reservoir in the Sichuan Basin,southwestern China,yielded predicted micro-fault system characteristics that were consistent with available geological information.In addition,the result agreed well with a log interpretation,confirming that the proposed method can provide support for reliable predictions micro-fault systems on complex reservoirs.

关 键 词：复杂断裂储层 岩石物理 地震属性 机器学习 支持向量机 断裂识别 储层预测 

分 类 号：P631[天文地球—地质矿产勘探]