利用随机森林回归算法预测总有机碳含量  

Predicting Total Organic Carbon Content by Random Forest Regression Algorithm

作　　者：冯明刚[1] 严伟[1] 葛新民[2] 朱林奇 FENG Ming-gang;YAN Wei;GE Xin-min;ZHU Lin-qi(Explorationand Research Institute of Sinopec,Chengdu 610041,China;School of Geoscience,China University of Petroleum,Shandong Qingdao 266580,China;Key Laboratory of Exploration Technologies for Oil and Gas Resources,Ministry of Education,Hubei Cooperative Innovation Center of Unconventional Oil and Gas,Yangtze University,Wuhan 430100,China)

机构地区：[1]中国石化勘探分公司勘探研究院,成都610041 [2]中国石油大学(华东)地球科学与技术学院,山东青岛266580 [3]长江大学油气资源与勘探技术教育部重点实验室非常规油气湖北省协同创新中心,武汉430100

出　　处：《矿物岩石地球化学通报》2018年第3期475-481,共7页Bulletin of Mineralogy, Petrology and Geochemistry

基　　金：国家科技重大专项(2017ZX05036005);中石化科技部项目(P16110)

摘　　要：针对现有页岩气储集层总有机碳含量预测模型存在的模型泛化能力弱、稳定性差的问题,提出了一种利用随机森林回归算法预测储集层总有机碳含量的方法。该方法使用地球物理测井提供的密度、铀含量、钍含量、自然伽马及光电吸收截面吸收指数等测井响应值作为输入,岩芯实验总有机碳含量作为输出,通过学习输入曲线与总有机碳含量的函数关系,动态预测整口井的总有机碳含量曲线。通过对焦石坝地区两口页岩气探井建模及预测可知,当随机森林中树的数量达到500时,建立的模型即可对训练样本中输入与输出的函数关系进行完全学习。通过训练结果及预测结果可知,随机森林回归方法不易发生过拟合现象,泛化能力极强,同时预测得到的曲线更为平滑,预测总有机碳含量较其他方法更为准确,有效地提高测井信息预测总有机碳含量模型的精度,对页岩气储集层评价提供帮助。The existing TOC prediction model of shale gas reservoir has weak generalization ability and is not stable,thus,a method for predicting reservoir TOC using random forest regression algorithm is proposed. The method takes logging responses of density,uranium and thorium contents,gamma ray and photoelectric absorption cross section as input and TOC content as output. By studying the function relation between input and TOC,the variation of TOC of the whole well is predicted dynamically. The modeling and prediction results for two wells in the Jiaoshiba area have shown that the established model can completely study the input and output functions when the number of trees in random forest reaches 500. Through the comparison between training and prediction results,random forest regression method is difficult to be overfitting and generalization ability is very strong. At the same time,TOC forecast can generate more accurate results and smoother predicted curve than other methods. The above results show that the random forest regression algorithm is powerful to improve the accuracy of logging prediction TOC model,and can be helpful for the exploration and development of shale gas.

关 键 词：页岩气 总有机碳含量 随机森林回归 机器学习 

分 类 号：P631[天文地球—地质矿产勘探]