基于树形回归法预测页岩油藏压裂水平井产能研究  

Research on predicting the productivity of fractured horizontal wellsin shale reservoirs based on the tree regression method

作　　者：李菊花 陈镜有[1,2] 秦顺利 陈雨 梁成钢 陈依伟 LI Juhua;CHEN Jingyou;QIN Shunli;CHEN Yu;LIANG Chenggang;CHEN Yiwei(Hubei Key Laboratory of Production Engineering for Oil and Gas Drilling(Yangtze University),Wuhan 430100,Hubei;School of Petroleum Engineering,National Engineering Research Center for Oil&Gas Drilling and Completion Technology,Yangtze University,Wuhan 430100,Hubei;Jiqing Oilfield Operation Area,Xinjiang Oilfield Branch,CNPC,Jimsar 831700,Xinjiang)

机构地区：[1]油气钻采工程湖北省重点实验室(长江大学),湖北武汉430100 [2]长江大学石油工程学院油气钻完井技术国家工程研究中心,湖北武汉430100 [3]中国石油新疆油田分公司吉庆油田作业区,新疆吉木萨尔831700

出　　处：《长江大学学报（自然科学版）》2024年第3期47-54,共8页Journal of Yangtze University(Natural Science Edition)

基　　金：中国石油天然气股份有限公司重大科技专项“吉木萨尔凹陷页岩油勘探开发示范工程”(2019E-2609)。

摘　　要：页岩油井水平段体积压裂段数多且产能差异大,常规产能预测和压裂效果评价难度大,借助机器学习建立一种稳定、高效的智能产能预测方法是提升页岩油藏开发的有效途径。使用吉木萨尔页岩油藏的91口生产井的地质参数、工程参数和生产数据库,基于热力学图版和特征参数相关性分析对参数进行验证,从14个特征参数中确定6个涵盖地质因素与施工因素的最佳主控因素。采用树形回归法的决策树(DT)、随机森林(RF)和梯度提升决策树(GBDT)三种机器学习方法进行产量预测建模,利用均方根误差对模型性能进行评估。研究结果表明,含水率、含油饱和度、加砂量、压裂液用量、压裂段簇数和压裂级数是影响压裂水平井产能的主控因素;随机森林模型的预测效果最好,预测准确度达到94%,测试集均方根误差为0.934;三种树形模型中的随机森林模型优于决策树模型和梯度提升决策树模型,解决了其他树形模型的过拟合问题。The number of horizontal section volume fracturing sections in shale oil wells is large and the production capacity varies greatly,so conventional production capacity prediction and fracturing effect evaluation are difficult,and establishing a stable and efficient intelligent production capacity prediction method with the help of machine learning is an effective way to enhance the development of shale reservoirs.In this study,we used the geological parameters,engineering parameters,and production databases of 91 production wells in the Jimsar shale reservoirs,the data were validated based on thermodynamic plat and characteristic parameter correlation analysis,and determined the 6 best main control factors covering geological factors and construction factors from the 14 feature parameters.Three machine learning methods,decision tree(DT)with tree regression method,random forest(RF),and gradient boosting decision tree(GBDT)were used for yield prediction modeling,and the model performance was evaluated using root mean square error.The results showed that water content,oil saturation,sand addition,fracturing fluid dosage,the number of fracturing section clusters,and the number of fracturing stages were the main controlling factors affecting the production capacity of fractured horizontal wells.The random forest tree regression method had the best prediction effect,with 94%prediction accuracy and 0.934 root mean square error for the test set.The random forest model of the self-sampling method among the three tree regression methods outperformed the decision tree model and the gradient boosting decision tree model,which solved the problem of overfitting problem of other tree models.

关 键 词：页岩油藏 产能预测 随机森林 决策树 压裂水平井 

分 类 号：TE34[石油与天然气工程—油气田开发工程]