基于贝叶斯优化XGBoost模型陡坡桥梁桩基变形可靠性分析方法  

Reliability Analysis of Piles Foundation of Steep Slope Bridge Based on Deformation Surrogate Model of Bayesian Optimization XGBoost Algorithm

作　　者：张晓忠 刘文杰 卢骜 唐仁华 陈昌富[1] 钟专 ZHANG Xiaozhong;LIU Wenjie;LU Ao;TANG Renhua;CHEN Changfu;ZHONG Zhuan(School of Civil Engineering,Hunan University,Changsha 410082,China;China Construction Ffth Engineering Division Co.,Ltd.,Changsha 410000,China;School of Civil Engineering,Changsha University,Changsha 410022,China)

机构地区：[1]湖南大学土木工程学院,湖南长沙410082 [2]中国建筑第五工程局有限公司,湖南长沙410000 [3]长沙学院土木工程学院,湖南长沙410022

出　　处：《湖南科技大学学报(自然科学版)》2025年第1期45-54,共10页Journal of Hunan University of Science And Technology:Natural Science Edition

基　　金：湖南省教育厅优秀青年科学研究项目资助(24B0790);国家自然科学基金面上项目资助(52278349);中国建筑第五工程局有限公司技术开发项目资助(中建050120220539020)。

摘　　要：陡坡桥梁桩基的变形(沉降和水平位移)是影响桥梁运营安全的关键因素,为解决按变形控制的陡坡桥梁桩基可靠度计算问题,较全面地考虑地层、基桩和荷载各方面的不确定性因素,引入XGBoost模型构建基桩变形代理模型,提出了一种按变形控制的陡坡桥梁基桩可靠性分析方法.首先,采用均匀试验设计法设计陡坡桥梁基桩数值模拟计算方案;然后,按计算方案建立相应的三维有限元数值模型,并由计算结果得到桩顶沉降和水平位移样本数据集;其次,基于样本数据集,采用贝叶斯优化的XGBoost算法构建基桩的沉降和水平位移代理模型;最后,基于变形代理模型,采用蒙特卡洛直接抽样法对陡坡桥梁基桩进行可靠性分析计算.结果表明:基于贝叶斯优化的XGBoost构建的基桩竖向和水平变形代理模型的拟合和测试验证效果良好,在训练集和测试集上的可决系数R^(2)均达到了0.95以上;提出的方法有效地提高了陡坡桥梁基桩沉降和水平位移可靠度计算效率.研究成果可为陡坡桥梁基桩的设计和计算提供更加科学的理论支撑.The settlement and horizontal displacement of bridge pile foundations on steep slopes are critical factors that impact the safety of bridge operations.To address the reliability calculation issue associated with deformation-controlled bridge pile foundations on steep slopes,this study comprehensively considers uncertainties related to strata,pile foundations,and loads.It introduces an XGBoost model to develop a proxy model for foundation pile deformation.Additionally,it presents a reliability analysis method for bridge foundation piles on steep slopes controlled by deformation.This method utilizes the uniform test design approach to establish a numerical simulation calculation scheme and corresponding three-dimensional finite element numerical model for bridge foundation piles on steep slopes.Subsequently,sample data sets of pile top settlement and horizontal displacement are obtained from the calculation results.Based on these data sets,a proxy model for setlement and horizontal displacement of foundation piles is constructed using the Bayesian optimization XGBoost algorithm.Finally,the Monte Carlo direct sampling method is employed to analyze and calculate the reliability of bridge foundation piles on steep slopes based on the deformation proxy model.The results demonstrate that vertical and horizontal deformation proxy models based on Bayesian optimization XGBoost exhibit excellent fitting and prediction effects,with coefficient of determination R^(2) values above 0.95 for both training set and test set scenarios.This proposed method effectively enhances the efficiency of reliability calculations for pile settlement and horizontal displacement in bridge foundation construction on steep slopes.The research findings can offer enhanced scientific theoretical backing for the design and computation of bridge foundation piles situated on steep slopes.

关 键 词：桩 沉降 水平位移 代理模型 均匀设计 机器学习 

分 类 号：TU473[建筑科学—结构工程]