基于人工智能算法的地质分层技术及应用效果  

Geological Stratification Technology Based on Artificial Intelligence Algorithms and Its Application Effects

作　　者：高源[1] 姚卫华[1] 蔡少锋 李良[1] 薛媛[1] 赵佩佩 王肖洋 魏炜 GAO Yuan;YAO Weihua;CAI Shaofeng;LI Liang;XUE Yuan;ZHAO Peipei;WANG Xiaoyang;WEI Wei(Exploration and Development Research Institute,PetroChina Changqing Oilfield Company,Xi’an,Shaanxi 710016,China;Xi’an Shiwen Software CO.LTD.,Xi’an,Shaanxi 710075,China)

机构地区：[1]中国石油长庆油田公司勘探开发研究院,陕西西安710016 [2]西安石文软件有限公司,陕西西安710075

出　　处：《测井技术》2024年第2期204-214,共11页Well Logging Technology

基　　金：中国石油天然气集团有限公司攻关性应用性科技专项“陆相页岩油规模增储上产与勘探开发技术研究”(2023ZZ15)。

摘　　要：地质分层是地质研究的基础,实际工作中地质人员进行大量井的地层划分,工作量大,且不同地质研究人员分层差异较大,分层结果不稳定。根据地质分层任务的特征,对比卷积神经网络、关注分层边界的卷积神经网络、掩码自编码器这3种大数据分析算法,优选掩码自编码器算法进行地质建模。根据地质分层任务的特殊性,结合地质构造,加入条件随机场进行层位顺序约束,对掩码自编码器算法进一步优化。以XX油田A、B区块作为研究示范区,建立地层划分样本,构建地质分层模型,实现二级地层全井段划分以及三级小层精细划分,均取得较好的预测效果,并实现了利用探评井预测开发井。该技术能有效解决分层结果稳定性的问题,并且可以实现批量井快速预测。Geological stratification is the foundation of geological research.In the actual work,geological personnel carry out a large number of well stratigraphic divisions,which requires a large amount of work.Moreover,there are significant differences in stratification among different geological researchers,and the stratification results are unstable.Based on the characteristics of geological stratification tasks,three big data analysis algorithms,convolutional neural network,convolutional neural network focusing on hierarchical boundaries and mask autoencoder,are compared,and the mask autoencoder algorithm is selected for geological modeling.Based on the particularity of geological stratification tasks and geological structures,conditional random fields are added to constrain the layer order,and the mask autoencoder algorithm is further optimized.Taking the block A and B in XX oilfield as research demonstration areas,formation division samples are established and geological stratification models are established to realize the division of the entire well section of the second level strata and fine division of the third level small layers.Good prediction results are achieved,and exploration and evaluation wells are used to predict development wells.This technology can effectively solve the problem of stability in stratification results,and can realize rapid prediction of batch wells.

关 键 词：人工智能 机器学习 掩码自编码器 地质分层 智能分层技术 

分 类 号：P631.84[天文地球—地质矿产勘探]