基于BP神经网络的黏土矿物预测模型  

BP Neural Network⁃Based Models to Predict Clay Minerals

作　　者：李鑫羽 欧阳传湘[1] 杨博文 赵鸿楠 聂彬[1] LI Xinyu;OUYANG Chuanxiang;YANG Bowen;ZHAO Hongnan;NIE Bin(School of Petroleum Engineering,Yangtze University,Wuhan,Hubei 430100,China;No.8 Oil Production Plant,Changqing Oilfield Company,PetroChina,Xi’an,Shaanxi 710000,China)

机构地区：[1]长江大学石油工程学院,武汉430100 [2]中国石油长庆油田分公司第八采油厂,西安710000

出　　处：《新疆石油地质》2021年第5期624-629,共6页Xinjiang Petroleum Geology

基　　金：国家自然科学基金(51804044)。

摘　　要：黏土矿物的准确预测是深层钻探和油气层保护的关键,为明确塔里木盆地库车凹陷北部构造带侏罗系阿合组黏土矿物分布规律,利用自然伽马能谱测井参数、阳离子交换能力、含氢指数和光电吸收截面指数,构建了基于BP神经网络的测井模型和组合模型,其平均绝对误差分别为5.34%和2.38%。将构建的模型应用于依南5井,结合X射线衍射资料,2个模型的平均绝对误差分别为4.64%和3.45%。模型预测的依南5井黏土矿物含量从高到低依次为伊利石、绿泥石、伊蒙混层和高岭石,在油藏开发中应预防储集层速敏和酸敏伤害。Accurate prediction of clay minerals is the key to deep drilling operation and pay zone protection.In order to determine the dis⁃tribution law of clay minerals in the Jurassic Ahe formation in the northern tectonic zone of the Kuqa depression,Tarim basin,a well log⁃ging model and a combined model based on BP neural network were constructed by using GR logging parameters,cation exchange capacity,hydrogen index and photoelectric absorption cross⁃section index.The average absolute errors of the two models are 5.34%and 2.38%,re⁃spectively.Applied to Well Yinan 5,the average absolute errors of the models are 4.64%and 3.45%,respectively,by considering X⁃ray dif⁃fraction data.The prediction result shows that the clay mineral contents from high to low are illite,chlorite,illite/smectite mixed layer and kaolinite in Well Yinan 5.Damages from velocity sensitivity and acid sensitivity should be prevented in reservoir development.

关 键 词：黏土矿物 自然伽马能谱测井 BP神经网络 阳离子交换能力 含氢指数 光电吸收截面指数 

分 类 号：TE112[石油与天然气工程—油气勘探]