强容噪性随机森林算法在地震储层预测中的应用  

Strong tolerance random forest algorithm in seismic reservoir prediction

作　　者：宋建国[1,2] 杨璐 高强山 刘炯 

机构地区：[1]中国石油大学(华东)地球科学与技术学院,山东青岛266580 [2]海洋矿产资源评价与探测技术功能实验室,山东青岛266071 [3]中国科学院地球化学研究所月球与行星科学研究中心,贵州贵阳550081 [4]中国石化石油勘探开发研究院,北京100083

出　　处：《石油地球物理勘探》2018年第5期954-960,共7页Oil Geophysical Prospecting

基　　金：国家"十三.五"科技重大专项(2017ZX05049002);国家自然科学基金项目(41674125);国家自然科学基金委员会中国石油化工股份有限公司石油化工联合基金资助项目(U1663207);国家重点基础研究发展计划项目("973"计划项目)(2014CB239104)联合资助

摘　　要：地震数据和测井数据中的噪声与有效信号难以有效界定,决定了地震储层预测需采用强容噪性算法。通过训练样本中加入随机噪声证实随机森林算法具有较好容噪性,但不能据此推知它在地震储层预测中仍有很强容噪性。基于F3工区实际数据,从噪声较强的原始地震数据中提取含噪样本,由经过倾角中值滤波处理的地震数据提取去噪样本,建立多种地震属性与孔隙度参数之间的随机森林回归模型;由构建的含噪模型和去噪模型分别与原始地震数据去噪前后两个数据体进行运算,得到4种不同情况下的孔隙度数据体。结果表明:由含噪模型得到的两个预测结果受噪声干扰较大;去噪模型的两个预测结果受噪声影响较小,能有效刻画储层特征,表现出强容噪性。随机森林模型对异于样本数据的异常值具有强的容忍度。可知随机森林算法应用于地震储层预测的关键是样本数据不含噪声,而估算过程中地震数据体是否做了去噪处理对预测结果影响较小。Since it is difficult to effectively define noise and signal on seismic and logging data,seismic reservoir prediction needs good noise tolerance algorithms.The random forest(RF)algorithm with strong noise tolerance is proved by adding noise to training samples.However this does not mean RF has good noise tolerance in seismic reservoir prediction as well.First we extract noise samples from original seismic data with strong noise in the Survey F3,and extract denoised samples from the data processed by the dip-steered median filter.Then we establish random forest regression models between seismic attributes and the porosity parameter.After processing the original seismic data and the filtered seismic data with the noise sample model and denosied sample model,we estimate four different porosity parameter cubic data.The results reveal that the two data sets obtained with the noise model are more disturbed by noise,and the other two data sets obtained with the denoised model are much less affected by noise.On these data sets,reservoir geological characteristics can be effectively characterized which proves the random forest model has strong robustness and perfect tolerance to abnormal data differing from the sample data.The key issue in the application of the random forest algorithm to seismic reservoir prediction is that the training data does not contain noise.In other words,the input variable of sample data being denoised is much more significant,whereas whether seismic data were denoised or not has less effects on the prediction result.

关 键 词：地震属性 随机森林 容噪性 储层预测 噪声 

分 类 号：P631[天文地球—地质矿产勘探]