摘要
目前低渗透率储层是我国油气开发领域的重点,其流体性质的识别对油田勘探开发具有重要指导意义。低渗透率储层岩石物理特征复杂、测井响应特征表现不明显,导致流体性质识别困难。集成学习因其强大的非线性能力和高效性成为储层智能评价的有力工具,但最终评价效果受限于样本质量。针对低渗透率储层的标签数据分布不均匀和稀缺的问题,提出了一种基于SMOTE(Synthetic Minority Oversampling Technique)采样和集成学习的低渗透率储层流体性质识别方法。利用SMOTE采样合理增加岩心标签数据,以符合集成学习模型的训练需求,进而优选集成学习模型,实现对低渗透率储层流体性质的准确识别。基于SMOTE采样和集成学习的流体识别方法在东营凹陷Y9XX井组的应用结果表明,该方法能有效识别低渗透率储层的流体性质,其准确率达87.44%。在此基础上,对东营凹陷的Y94X井进行盲井测试,最终的分类结果满足实际测井解释对精度的需求。SMOTE采样结合集成学习的流体识别模式为后续机器学习在储层评价的广泛应用提供了依据。
At present,low-permeability reservoirs are the focus of oil and gas development in China.The identification of their fluid properties is of great guiding significance for oilfield exploration and development.The petrophysical characteristics of low-permeability reservoirs are complex,and the logging response characteristics are not obvious,resulting in difficulties in fluid property identification.Integrated learning,with its powerful non-linear ability and high efficiency,has become a powerful tool for intelligent reservoir evaluation.However,its final effect is limited by the quality of samples.Aiming at the problems of uneven distribution and scarcity of labeled data in low-permeability reservoirs,a method for identifying fluid properties in low-permeability reservoirs based on SMOTE sampling and integrated learning technology is proposed.The SMOTE sampling is used to reasonably expand the core labeled data to meet the training requirements of the integrated learning model.Then,the integrated learning model is optimized to achieve the identification of fluid properties in low-permeability reservoirs.The application results of the method for identifying fluid properties in low-permeability reservoirs based on SMOTE sampling and integrated learning in the Y9XX well group of the Dongying depression show that,this method can effectively identify the fluid properties of low-permeability reservoirs,with an accuracy rate of 87.44%.On this basis,a blind well test is carried out on Y94X well in the Dongying depression,and the final classification results meet the accuracy requirements of actual logging interpretation.The fluid identification model combining SMOTE sampling and integrated learning provides a basis for the wide application of subsequent machine learning in reservoir evaluation.
作者
杨文凯
孙建孟
杜钦波
张宇昆
罗歆
YANG Wenkai;SUN Jianmeng;DU Qinbo;ZHANG Yukun;LUO Xin(School of Earth Sciences and Technology,China University of Petroleum(East China),Qingdao,Shandong 266580,China;Logging Technology Research Institute,China National Logging Corporation,Xi'an,Shaanxi 710077,China;Science and Technology Management Department,China National Logging Corporation,Xi'an,Shaanxi 710077,China;Well Logging Technology Pilot Test Center,China National Petroleum Corporation,Xi'an,Shaanxi 710077,China)
出处
《测井技术》
2025年第1期1-9,共9页
Well Logging Technology
基金
国家自然科学基金项目“深部低阻砂岩气藏渗流与导电机理模拟分析研究”(42174143)
中国石油集团测井有限公司开放基金课题“基于电成像数据的地层产状计算算法研究”(CNLC2022-9C06)。
关键词
流体性质识别
集成学习
SMOTE采样
样本不均匀
东营凹陷
fluid property identification
integrated learning
SMOTE sampling
sample unevenness
Dongying depression
