基于随机森林的浊积岩储层预测方法  

The Method of Turbidite Reservoir Prediction Based on Random Forest

作　　者：任雄风 刘杨[1] 张军华[1] 谭明友[2] 张云银[2] 于正军[2] REN Xion-feng;LIU Yang;TAN Ming-you;ZHANG Yun-yin;YU Zhengjun(School of Geosciences,China University of Petroleum(East China),Qingdao,266580,China;Geophysical Research Institute of Shengli Oilfield Company,Sinopec,Dongying,257015,China)

机构地区：[1]中国石油大学(华东)地学院,青岛266580 [2]胜利油田物探研究院,东营257015

出　　处：《科学技术与工程》2019年第25期68-74,共7页Science Technology and Engineering

基　　金：国家科技重大专项(2017ZX05009-001、2017ZX05072-001)资助

摘　　要：由于用单一地震属性描述浊积岩储层厚度有很大不确定性,基于多种地震属性,将随机森林算法引入对浊积岩储层厚度的预测中。通过试验,优选出弧长、能量半时、均方根振幅、最大振幅、平均能量和道积分等六种地震属性,构建井旁道地震属性与浊积岩厚度之间的关系模型,对浊积岩储层厚度进行预测。研究结果表明,随机森林方法对异常值和噪声具有很好的容忍度,训练速度快,泛化误差小,不易出现过拟合现象,预测精度高于神经网络方法,有较好的推广价值。Due to the great uncertainty of describing turbidite reservoir thickness with a single seismic attribute,the random forest algorithm was introduced into the prediction of turbidite reservoir thickness based on a variety of seismic attributes. Through experiments,six kinds of seismic attributes such as arc length,energy half time,rootmean-square amplitude,maximum amplitude,average energy and trace integral are selected,and the relationship model between well side channel seismic attributes and turbidite thickness was built to predict the thickness of turbidite reservoir. The results show that the random forest method has good tolerance to outliers and noises,high training speed,small generalization error,and is less prone to overfitting. The prediction accuracy is higher than the neural network method,so it has good promotion value.

关 键 词：浊积岩 随机森林 神经网络 地震属性 厚度预测 

分 类 号：P631[天文地球—地质矿产勘探]