基于DBN模型的火山岩岩性识别方法——以车排子地区为例  

Lithology identification method of volcanic rocks based on DBN model:an case in Chepaizi area

作　　者：赵军[1] 王伟明 杨志冬 张梦露 ZHAO Jun;WANG WeiMing;YANG ZhiDong;ZHANG MengLu(School of Geoscience and Technology,Southwest Petroleum University,Chengdu 610500,China;The First Oil Production Plant of Xinjiang Oilfield Company CNPC,Karamay 834000,China)

机构地区：[1]西南石油大学地球科学与技术学院,成都610500 [2]中国石油新疆油田分公司采油一厂,克拉玛依834000

出　　处：《地球物理学进展》2022年第1期328-337,共10页Progress in Geophysics

基　　金：中国海油湛江分公司科技项目“文昌9、10区低孔低渗储层测井精细评价及潜力分析”(CCL2019ZJFN0823)资助。

摘　　要：准噶尔盆地西北缘红车断裂带石炭系火山岩储层历经火山多期喷发形成,岩石矿物组成与岩石结构复杂、岩石类型繁多,导致常规方法识别岩性存在困难.为了能提高火山岩岩性识别精度,提出一种基于深度置信网络(DBN)识别岩性的方法.本文通过对薄片鉴定资料、岩心观察资料及测井数据统计分析,根据火山岩类型总结的测井响应特征绘制交会图,发现其识别火山岩岩性效果较差.在此基础上,优选反映岩性变化的5种常规测井敏感曲线(GR、CNL、DEN、AC、RT)作为特征向量,建立岩性识别总样本数据.其中选取80%的数据样本点作为构建DBN模型的训练样本,剩下的20%数据样本点作为测试样本验证该模型的预测精度.实验结果表明,利用深度置信网络(DBN)识别火山岩岩性的方法预测正确率可达到85.9%.与常规交会图、Fisher判别模型及BP神经网络模型相比,基于DBN模型的岩性识别方法效果更好,能为研究区火山岩岩性识别提供一定借鉴.The Carboniferous volcanic rock reservoirs in the Hongche fault zone on the northwestern margin of the Junggar Basin was formed by volcanic eruptions in multiple periods. The rock mineral composition and rock structure are complex, and the rock types are diverse, making it difficult to identify lithology by conventional methods. In order to improve the accuracy of volcanic rock lithology identification, a method based on Deep Belief Network(DBN) to identify lithology is proposed. In this article, through statistical analysis of thin section identification data, according to the logging response characteristics summarized by the volcanic rock types drawing the cross plot, and it is found that the volcanic rock lithology identification effect is poor.On this basis, five conventional logging sensitivity curves(GR, CNL, DEN, AC, RT) reflecting lithological changes are selected as feature vectors, and to established the total sample data of lithology recognition. 80% of the data samples are selected as training samples to build the model, and the remaining 20% data samples are used as test samples to verify the prediction accuracy of the model. The experimental results show that the prediction accuracy rate of the method using the Deep Belief Network(DBN) to identify volcanic rock lithology can reach 85.9%. Compared with the conventional cross-plot, Fisher discriminant model and BP neural network model, the lithology identification method based on DBN model has a better effect, and can provide certain reference for volcanic rock lithology identification in the study area.

关 键 词：火山岩储层 岩性识别 常规测井资料 交会图 深度置信网络(DBN) 

分 类 号：P631[天文地球—地质矿产勘探]