基于迁移学习的非均质储层参数预测方法  

A Prediction Method for Heterogeneous Reservoir Parameters Based on Transfer Learning

作　　者：高国海 王欣 蒋薇 王永生[1] 张恩莉[2] 周燕[2] 李亮 Gao Guohai;Wang Xin;Jiang Wei;Wang Yongsheng;Zhang Enli;Zhou Yan;Li Liang(Southwest Petroleum University,Chengdu,Sichuan 610500,China;PetroChina Southwest Oil&Gasfield Company,Chengdu,Sichuan 610041,China;CNPC Chuanqing Drilling Engineering Company Limited,Chengdu,Sichuan 610051,China)

机构地区：[1]西南石油大学,四川成都610500 [2]中国石油西南油气田分公司,四川成都610041 [3]中国石油川庆钻探工程有限公司,四川成都610051

出　　处：《特种油气藏》2024年第5期59-66,共8页Special Oil & Gas Reservoirs

基　　金：国家自然科学基金优秀青年基金“页岩气多尺度非线性渗流力学”(52222402);西南石油大学油气藏地质及开发工程全国重点实验室开放性研究课题“面向油气产量预测的非平稳时间序列模型构建及应用”(PLN 2022-33)。

摘　　要：针对传统储层参数预测方法常忽略渗流机理和参数相关性的问题,提出了一种融合渗流理论与迁移学习的储层参数预测模型。通过SMOTE过采样算法,有效处理了样本不均衡问题;利用随机森林算法建立了岩性及渗流能力判别模型,为储层参数预测提供渗流机理信息;结合参数相关性,运用迁移学习技术搭建储层参数预测模型。结果表明,通过引入岩性和渗流能力判别技术,能够进行储层参数之间的相关性分析,可以有效提升储层参数的预测精度。该模型的储层孔隙度和渗透率的误差为3.51%和15.17%,预测准确率较传统的储层参数预测方法有显著提升。该方法可有效解决非均质储层参数预测难题,为人工智能技术与物理模型结合研究提供参考。To address the issue of traditional methods neglecting flow mechanisms and parameter correlations,a prediction model of reservoir parametert hat integrates seepage theory with transfer learning is proposed.By using the oversampling algorithm of SMOTE,the issue of imbalanced samples is effectively addressed.The discriminative model of lithology andseepage capacity is established by using random forest,which provides the information of seepage mechanism for reservoir parameter predicting.Combined with parameter correlation,the technology of transfer learningis used to build a reservoir parameter prediction model.The results show that the correlation analysis between reservoir parameters can be conducted by introducing lithology and seepage capacity discrimination technology,which can effectively improve the prediction accuracy of reservoir parameters.The prediction error of the model in porosity and permeability parameters is 3.51%and 15.17%,respectively,and the prediction accuracy is significantly improved.This method can effectively address the issues of parameters prediction in heterogeneous reservoir,and provide reference for the research combining artificial intelligence technology with physical models.

关 键 词：储层参数预测 迁移学习 SMOTE 随机森林 神经网络 

分 类 号：TE311[石油与天然气工程—油气田开发工程]