摘要
由于孔渗统计回归和测井解释方法在致密砂岩储层参数预测中表现不佳,人工智能方法被广泛应用于致密砂岩储层参数预测中。然而,可用的岩心数据很难满足人工智能大量学习样本的要求。因此,提出了基于高斯混合模型的虚拟样本生成方法,以解决缺乏训练样本的问题。该算法的通过拟合原始样本的分布来生成虚拟样本,填充了小样本数据之间的信息缺失。通过标准函数测试,该方法能有效生成训练数据,实际工区孔隙度和渗透率预测对比试验表明,经过虚拟样本扩充数据集后,模型的预测准确率分别提高了9.7%和18.6%,表明所提出的方法可以有效地提高小样本条件下的模型预测精度。
Due to the poor performance of pore permeability statistical regression and logging interpretation method in predicting parameters of tight sandstone reservoirs,artificial intelligence methods are widely used in predicting parameters of tight sandstone reservoirs.However,the available core data is difficult to meet the requirements of artificial intelligence for learning a large number of samples.Therefore,a virtual sample generation method based on Gaussian mixture model is proposed to solve the problem of lacking training samples.This algorithm generates virtual samples by fitting the distribution of the original samples,filling in the information gaps between small sample data.Through standard function testing,this method can effectively generate training data.Comparative experiments on predicting porosity and permeability in actual work areas show that after expanding the dataset with virtual samples,the prediction accuracy of the model has increased by 9.7%and 18.6%,respectively.This indicates that the proposed method can effectively improve the prediction accuracy of the model under small sample condition.
作者
韩旭东
张广智
刘飞
郭彦民
刘太伟
杜磊
朱孔斌
徐帅
HAN Xudong;ZHANG Guangzhi;LIU Fei;GUO Yanmin;LIU Taiwei;DU Lei;ZHU Kongbin;XU Shuai(Seventh Oil Production Plant of Changqing Oilfield Branch Company of China National Petroleum Corporation,Xi'an 710200,Shaanxi,China;School of Geosciences,China University of Petroleum School,Qingdao 266580,Shandong,China;Exploration and Development Research Institute of Liaohe Oilfield Branch Company of China National Petroleum Corporation,Panjin 124010,Liaoning,China)
出处
《矿产与地质》
2024年第1期195-204,共10页
Mineral Resources and Geology
基金
国家自然科学基金项目(编号:42074136,U23B6010)
国家科技重大专项(编号:2016ZX05002-005)
中国石油大学(华东)研究生创新基金(编号:YCX2020014)等共同资助。
关键词
储层参数预测
高斯混合模型
虚拟样本生成
深度学习
reservoir parameter prediction
Gaussian mixture model
virtual sample generation
deep learning
