水平定向钻载荷下天然气管道失效智能预测研究  

Intelligent prediction of natural gas pipeline failure under horizontal directional drilling loads

作　　者：何婷 杨松 张开 陈利琼 黄坤 陈星宇 HE Ting;YANG Song;ZHANG Kai;CHEN Liqiong;HUANG Kun;CHEN Xingyu(School of Petroleum and Natural Gas Engineering,Southwest Petroleum University,Chengdu 610500,China;China National Petroleum Transport Corporation Sichuan Branch,Chengdu 610000,China)

机构地区：[1]西南石油大学石油与天然气工程学院,成都610500 [2]中国石油运输有限公司四川分公司,成都610000

出　　处：《安全与环境学报》2025年第4期1370-1379,共10页Journal of Safety and Environment

摘　　要：为研究水平定向钻载荷下埋地天然气管道的失效条件,结合水平定向钻施工特点和天然气管道实际情况,建立了钻头-土体-管道有限元模型,得到包括影响水平定向钻载荷作用的六个关键特征变量的失效数据集。基于该数据集,建立智能失效预测模型,通过树结构Parzen估计器(Tree-structured Parzen Estimator,TPE)优化极端梯度提升(Extreme Gradient Boosting,XGBoost)模型性能。结果表明,与支持向量积、随机森林、贝叶斯回归等算法相比,XGBoost-TPE算法性能最好,平均绝对误差(Mean Absolute Error,MAE)为7.8127 MPa,均方根误差(Root Mean Square Error,RMSE)为11.3256 MPa,决定系数(R2)为0.9891。研究可为天然气管道交叉工程及周边工程水平定向钻施工风险定量评价及安全管理提供理论支撑。Third-party directional drilling is a significant contributor to failures in buried natural gas pipelines,making accurate failure prediction essential for ensuring pipeline safety.Current research primarily relies on finite element simulations,which,although complex and computationally intensive,are not widely implemented in pipeline safety management due to their limited practicality.This paper integrates finite element analysis with machine learning algorithms to create a dataset of factors influencing pipeline failure derived from finite element simulations.This dataset is then utilized to train a model for accurate prediction and analysis of pipeline failures.First,a finite element model of the drill bit,soil,and pipeline is developed based on the characteristics of horizontal directional drilling and the actual conditions of natural gas pipelines,applying the fourth-strength theory as the failure criterion for analysis.The fourth-strength theory was chosen as the failure criterion for this analysis.The control variable method was employed to examine the effects of six key factors on Mises stress.By extracting the maximum Mises stress under various feature vector conditions,a failure dataset was compiled,consisting of 428 feature vectors.This dataset will aid in the development of the prediction model.Extreme Gradient Boosting(XGBoost)was used as the predictive model,with the Tree-structured Parzen Estimator(TPE)employed to optimize hyperparameters,leading to the creation of the XGBoost TPE failure prediction model.The model was compared against three other machine learning models—Random Forest(RF),Support Vector Regression(SVR),and Bayesian Regression(BR)—to validate its performance.Among the eight algorithms tested,XGBoost TPE delivered the best results.Specifically,it achieved a Mean Absolute Error(MAE)of 7.8127 MPa,a Root Mean Square Error(RMSE)of 11.3256 MPa,and a coefficient of determination(R 2)of 0.9891.These results underscore the exceptional predictive accuracy and robustness of the XGBoost TPE

关 键 词：安全工程 天然气管道 水平定向钻载荷 有限元模拟 极端梯度提升 树结构Parzen估计器优化 

分 类 号：X937[环境科学与工程—安全科学]