AdaBoost算法在致密砂岩水淹层识别中的应用  

Application of AdaBoost algorithm in recognition of water flooded tight sandstone layer

作　　者：杨明任 申辉林[1] 曲萨 孙启鹏 章利民 肖淑明[2] YANG Mingren;SHEN Huilin;QU Sa;SUN Qipeng;ZHANG Limin;XIAO Shuming(School of Geoscience and Technology,China University of Petroleum,Qingdao,Shandong266580,China;Xianhe Oil Production Plant,Shengli Oilfield Company,Sinopec,Dongying,Shandong257068,China)

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]中石化胜利油田分公司现河采油厂,山东东营257068

出　　处：《中国海上油气》2021年第4期62-69,共8页China Offshore Oil and Gas

摘　　要：目前各类水淹层的测井响应特征不明显,特别是致密砂岩水淹层测井曲线主要反映储层岩性特征,而反映储层流体性质的信息十分有限,导致水淹层解释多解性极强,传统的交会图技术、测井曲线对比分析等方法识别水淹级别精度低,严重影响油田开发生产。为此,本文提出了基于AdaBoost算法的预测模型,该算法分类速度快,不需要繁杂的调参过程,不会出现过拟合情况,应用多个弱分类器的线性组合,综合判断分类结果,能够有效地提高致密砂岩水淹级别的识别精度。首先,结合长庆油田长6段致密砂岩现场资料把水淹层细划分为未水淹、低水淹、中水淹、高水淹等4个水淹级别,并将这4个水淹类型拆解为二分类问题,通过不断的迭代得到样本分布,然后将弱分类器经过线性加权平均得到强分类器。该方法应用到长庆致密砂岩水淹层识别中,水淹层预测精度达88.7%,在研究常规测井资料水淹层识别中具有重要意义。The logging response characteristics of all kinds of water flooded layer are not obvious,especially the logging curves of water flooded tight sandstone layer mainly reflect the reservoir lithologic characteristics,but the information reflecting the reservoir fluid properties is very limited,which leads to multiple solutions of water flooded layer interpretation.The traditional crossplot technique,logging curve comparative analysis and other methods all have low levels of precision in recognizing water flooded layer,which seriously affects oilfield development and production.Therefore,aprediction model based on the Adaboost algorithm is proposed in this paper.The algorithm has a fast classification speed,does not require complex parameter adjustment process,and does not happen over-fitting circumstances.The linear combination of several weak classifiers is applied to comprehensively judge the classification results,which can effectively improve the recognition accuracy of the water flooded level of tight sandstone.First,combined with the field data of tight sandstone in Chang 6 Member of Changqing oilfield,the water flooded layers are divided into four levels:unflooded,low flooded,medium flooded and high flooded,and further the four levels of water flooded type are decomposed into binary classifications.The sample distribution is obtained by repeated iteration,then the strong classifier is obtained by linear weighted average of the weak classifiers.This method is applied to the recognition of water flooded tight sandstone layers in Changqing oilfield,and the prediction accuracy of water flooded layers reaches 88.7%,which is of great significance in the study of recognition of water flooded layers with conventional logging data.

关 键 词：ADABOOST算法 分类器 致密砂岩 水淹层识别 线性组合 

分 类 号：TE132.1[石油与天然气工程—油气勘探]