基于地震属性集成学习的自然伽马测井数据预测方法  

Natural Gamma Well Logging Prediction Based on Seismic Attribute Ensemble Learning

作　　者：单立群[1] 祁妍嫣 姜淑贤 刘彦昌[1] 刘修远 SHAN Liqun;QI Yanyan;JIANG Shuxian;LIU Yanchang;LIU Xiuyuan(Northeast Petroleum University Qinhuangdao,Qinhuangdao,Hebei 066004,China;Petroleum Engineering School,Southwest Petroleum University,Chengdu,Sichuan 610500,China;Integrated Management Department,Tongxin Group,Dagang Oilfield Company,Tianjin 300280,China)

机构地区：[1]东北石油大学秦皇岛校区,河北秦皇岛066004 [2]西南石油大学石油与天然气工程学院,四川成都610500 [3]大港油田公司同欣集团综合管理部,天津300280

出　　处：《测井技术》2021年第4期394-398,共5页Well Logging Technology

基　　金：东北石油大学“国家基金”培育基金“基于大数据的碳酸岩流动单元及储层特征研究”(2018GP2D-04)。

摘　　要：地球物理测井中的地震反演是油气藏定性识别的一个重要步骤,非均质储层测井资料的识别是地质工作者和工程技术人员面临的重大挑战。提出了一种基于地震属性集成学习的测井数据预测方法,对大港油田X区块A区明化镇组高产砂岩储层进行应用分析。对研究区125口代表性井的地震体和测井数据进行了岩石物理性质分析,将地震数据与测井数据在同一尺度进行了集成。利用XGBoost算法对地震属性进行重要性分析,选择对测井数据预测影响最大的地震属性作为模型输入。分别使用XGBoost、AdaBoost和随机森林这3种集成学习算法建立预测模型。研究结果表明预测结果与真实值接近,与传统的方法相比具有更好的预测性能,基于地震属性的集成学习算法可以明显提高自然伽马测井数据的预测精度。Seismic inversion in geophysics is an important part in the process of characterization of oil and gas reservoirs.Identification of well logs in heterogeneous reservoirs presents a great challenge to geologists and engineers.A new method for predicting well logs by seismic attributes is proposed in this study,based on ensemble learning algorithms within the most productive sandstone reservoir in the Minghuazhen formation in region A,block X,in Dagang oilfield,China.Petrophysical properties derived from seismic volume and logs from 125 representative wells are analyzed.The seismic and logging data are integrated in the same domain.Importance analysis of seismic attributes to select the largest impact is performed with XGBoost algorithm to obtain trained data sets.Predictive models are then established by training XGBoost,AdaBoost,and Random Forest model.The result from this study is consistent with real logging data,and is more accurate than that from previous investigations.It is concluded that using the ensemble learning method can improve the accuracy of prediction of GR well logs.

关 键 词：测井解释 地震属性 测井曲线 集成学习 随机森林 XGBoost算法 ADABOOST算法 

分 类 号：P631.84[天文地球—地质矿产勘探]