摘要
随着复杂储层地震资料特征筛选的机器学习技术的进步,如何有效地对参与地震属性优选和储层反演的地震样本进行采集和分析,成为目前智能地震预测领域的一个研究热点。目前的方法多着重于模型分类算法的改进,在标签的制作和采集方面不仅耗费大量时间进行人工标注,还存在标签不平衡情况下类内可靠性、类间平衡性不强等问题。为此,提出基于稀疏强特征提取的三维地震数据完备方法。首先,基于多数决原则的样本分割(Sample Segmentation Based on Majority Rule,SSMR)寻迹多尺度、多标签三维地震样本,进行采集、自动标注;然后,改进标签洗牌平衡方法(Improved Label Shuffling Balance Method,ILSB),通过“2+1”的样本增广平衡策略进行数据完备处理,改善样本采样不平衡性导致的模型训练偏向性;最后,利用基于最小L_(1)范数稀疏表示对奇异值分解结果进行强特征提取(Minimum L_(1)-norm Based Sparse Representation for Feature Extraction,L_(1)-SRFE)和可视化表示。实际资料应用表明,实钻井与验证井预测结果吻合度高,该方法具有较高的标签分类准确率。
As machine learning technology for screening seismic data features of complex reservoirs develops,how to effectively collect and analyze seismic samples involved in seismic attribute optimization and reservoir inversion has currently become a hot research topic in the field of intelligent prediction based on seismic data.Existing methods mostly focus on improving model classification algorithms,which not only consume a lot of time for manual labeling in the production and collection of labels but also suffer from poor intra-class reliability and inter-class balance in the case of label imbalance.Therefore,a 3D seismic data completion method based on sparse strong feature extraction is proposed.First,sample segmentation based on majority rule(SSMR)is used to trace multi-scale and multi-label 3D seismic samples for collection and automatic labeling.Then,the improved label shuffling balance(ILSB)method is used to complete the data by a"2+1"sample broadening and balancing strategy,so as to improve the model training bias caused by unbalanced sample sampling.Finally,minimum L_(1)-norm based sparse representation for feature extraction(L_(1)-SRFE)and visual representation of the singular value decomposition results are performed.Application of the actual data shows that the predicted results of the actually drilled wells and the validation wells are in good agreement,and the method has a high accuracy of label classification.
作者
崔雪鹏
黄捍东
罗亚能
成锁
郝亚炬
崔刚
CUI Xuepeng;HUANG Handong;LUO Yaneng;CHENG Suo;HAO Yaju;CUI Gang(State Key Laboratory of Petroleum Resources and Prospecting,China University of Petroleum(Beijing),Beijing 102249,China;College of Geophysics,China University of Petroleum(Beijing),Beijing 102249,China;Geophysical Research&Development Center,BGP Inc.,CNPC,Zhuozhou,Hebei 072751,China;Exploration and Production Research Institute,Tarim Oilfield Company,PetroChina,Korla,Xinjiang 841000,China;School of Geophysics and Measurement-Control Technology,East China University of Technology,Nanchang,Jiangxi 330013,China;The First Production Plant,Huabei Oilfield Company,PetroChina,Renqiu,Hebei 062552,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2023年第2期263-276,共14页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“裂缝性储层地震定量预测及流体识别方法研究”(41974124)
“组稀疏结构和等效衰减模型双重约束下的叠前Q值反演方法研究”(42004114)
江西省自然科学基金项目“基于压缩感知的地震数据自适应压缩及反射系数快速反演”(20202BAB-211010)联合资助。
