基于Stacking集成学习的声波时差测井曲线复原研究  

Research on Acoustic Moveout Logging Curves Restoration Based on Stacking Ensemble Learning

作　　者：曹志民[1,2,3] 丁璐 韩建 CAO Zhi-min;DING Lu;HAN Jian(SANYAOfshore Oil&Gas Research Institute,Northeast Petroleum University;School of Physics and Electronic Engineering,Northeast Petroleum University;Research Center for Oil&Gas Testing and Measurement Technology and Instrumentation of Heilongjiang Province)

机构地区：[1]东北石油大学三亚海洋油气研究院 [2]东北石油大学物理与电子工程学院 [3]黑龙江省油气测试计量技术及仪器仪表工程技术研究中心

出　　处：《化工自动化及仪表》2024年第3期470-476,共7页Control and Instruments in Chemical Industry

基　　金：海南省科技专项(批准号:ZDYF2022GXJS220,ZDYF2022GXJS222)资助的课题。

摘　　要：声波时差测井曲线在石油勘探中发挥着不可或缺的作用,但是受地质或仪器的影响,经常会出现部分甚至完整的声波测井曲线缺失的情况。针对这一问题,提出了一种基于Stacking集成学习的声波时差测井曲线复原方法,该模型使用随机森林(RF)、梯度提升决策树(GBDT)、轻量梯度提升机(LightGBM)和极限梯度提升(XGBoost)作为基学习器,支持向量回归(SVR)作为元学习器,同时采用5折交叉验证的方法。实验选取了大庆油田某区块的实际测井数据,分别进行了同井和异井间的缺失声波时差测井曲线复原实验,结果表明,所提方法比单一模型预测更加准确,验证了此方法的可行性。Acoustic moveout logging curves play an indispensable role in petroleum exploration,but the influence from geology and instruments results in the loss of partial or even complete acoustic logging curves.In this paper,a method of acoustic moveout logging curve restoration based on Stacking ensemble learning was proposed.The model employs random forest(RF),gradient lifting decision tree(GBDT),lightweight gradient lifting machine(LightGBM)and extreme gradient lifting(XGBoost)as the base learners,takes support vector regression(SVR)as the meta-learner and adopts a five-fold cross validation method.In the experiment,the a section's actual logging data in Daqing Oilfield was selected to respectively implement the restoration experiments of the same well and different wells'missing acoustic transit time logging curves.The experimental results show that,the method proposed outperforms the single model in the prediction and it verifies the feasibility of this method.

关 键 词：声波时差测井曲线 Stacking集成学习 测井曲线复原 5折交叉验证 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]