基于XGBoost的测井曲线重构方法

Reconstruction of well logs based on XGBoost

测井数据在地层评价中具有十分重要的作用,但是实际应用中由于地质、工程等因素的影响,经常出现部分测井数据缺失甚至漏测的情况。基于传统的经验模型和多元回归分析的测井曲线重构方法精度不够,因此提出利用机器学习方法进行测井曲线的重构;考虑到神经网络的局限性,基于XGBoost构建了测井曲线重构模型。以渤海湾盆地定向井为例验证重构模型的效果:首先进行了测井曲线补全和生成实验,并通过K折交叉验证将XGBoost模型性能与梯度提升决策树(GBDT)、随机森林(RF)和全连接神经网络(FNN)三种方法进行对比,然后结合地质背景分析预测效果。验证结果表明,基于XGBoost的测井曲线重构方法在准确性和稳定性方面都取得了更好的效果,并且表现出较强的泛化能力。

Well logs play an important role in the formation evaluation.However,well log data might be missing or incomplete due to operational and geological issues in the logging process.As the reconstruction of well logs based on traditional empirical models and multiple regression method is less accurate,machine learning is proposed.Considering the limitation of the traditional neural network,XGBoost is utilized to build the reconstruction model for well logs.The directional wells in Bohai Bay Basin are exemplified to verify the model with the experiments of well log imputation and generation.The proposed model is compared with traditional machine learning models such as gradient lifting decision tree(GBDT),random forest(RF),and fully connected neural network(FNN)through K-fold cross-validation.The prediction effect is analyzed combined with the geological background.Results show that reconstruction of well logs based on XGBoost achieves high prediction accuracy,stability,and strong generalization ability.