摘要
准噶尔盆地玛湖凹陷风城组岩性复杂,为准确预测其岩石力学参数,提出了一种自适应权重组合预测方法。首先分析、对比传统方法和不同机器学习算法(BP神经网络、XGBoost、支持向量机(SVM)、随机森林(RF)、卷积神经网络(CNN)、决策树(CART)、长短时记忆神经(LSTM)网络等)的预测效果,传统方法难以准确预测岩石力学参数,而不同机器学习算法的预测效果不同,其中抗压强度、抗张强度和脆性指数预测的最优机器学习算法模型为SVM,弹性模量为BP,泊松比为RF,内聚力为XGBoost,内摩擦角和断裂韧性为LSTM网络;单一机器学习算法难以实现对多个岩石力学参数的同步准确预测。在此基础上,通过对不同岩石力学参数选取不同预测基模型,再根据基模型预测效果赋予权重并进行组合,以开展自适应权重组合预测。结果表明,该方法能够有效提升机器学习算法的预测精度和泛化性能,可实现复杂岩性地层多个岩石力学参数的同步准确预测。
The lithology of the Fengcheng Formation in Mahu Sag,Junggar Basin is complex.To accurately predict its rock mechanical parameters,this paper proposes an adaptive weight combination forecast method.Firstly,the paper analyzes and compares the predictive performance of traditional methods and different ma⁃chine learning algorithms(BP neural network,XGBoost,support vector machine(SVM),random fores(t RF),convolutional neural network(CNN),Classifation and regression tree(CART),long⁃short term memory neural(LSTM)network,etc.).Traditional methods are difficult to achieve accurate forecasts of rock mechanical pa⁃rameters,while different machine learning algorithms have different predictive effects.The optimal machine learning algorithm model for predicting compressive strength,tensile strength,and brittleness index is SVM.The optimal models for predicting elastic modulus,Poisson’s ratio,and cohesion are BP,RF,and XGBoost,respectively.The optimal model for predicting internal friction angle and fracture toughness is LSTM network.A single machine learning algorithm is difficult to achieve synchronous and accurate forecasts of multiple rock mechanical parameters.On this basis,adaptive weight combination forecast is carried out by selecting different forecast base models for different rock mechanical parameters,assigning weights based on the forecast effect of the base models,and combining them.The results show that this method can effectively improve the forecast accuracy and generalization performance of machine learning algorithms and can achieve synchronous and ac⁃curate forecasts of multiple rock mechanical parameters in complex lithological formations.
作者
唐俊方
熊健
刘向君
甘仁忠
罗德江
梁利喜
TANG Junfang;XIONG Jian;LIU Xiangjun;GAN Renzhong;LUO Dejiang;LIANG Lixi(National Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Southwest Petroleum University,Chengdu,Sichuan 610500,China;National Center for International Research on Deep Earth Drilling and Resource Development,Ministry of Science and Technology China University of Geosciences(Wuhan),Wuhan,Hubei 430074,China;PetroChina Xinjiang Oilfield Company,Karamay,Xinjiang 834000,China;Geomathematics Key Laboratory of Sichuan Province,Chengdu University of Technology,Chengdu,Sichuan 610059,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2024年第1期1-11,共11页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“陆相页岩油储层热-化-力耦合作用诱导致裂与增渗机理研究”(42272190)
中国石油科技创新基金项目“基于‘数据+知识’协同驱动的深层陆相页岩油储层工程甜点预测方法研究”(2023DQ02⁃0101)
数学地质四川省重点实验室2022年开放基金课题“页岩储层可压裂性定量表征与测井预测”(scsxdz2022⁃05)
科技部地球深部钻探与深地资源开发国际联合研究中心开放课题“基于地质力学特性的陆相页岩油储层可压裂性评价方法研究”(DE DRD⁃2023⁃03)联合资助。
关键词
岩石力学参数
复杂岩性地层
机器学习
自适应组合预测
rock mechanical parameters
complex lithologic formations
machine learning
adaptive combination forecast
