基于EnKF与ABAQUS数据同化方法的岩土参数反演分析  

Geotechnical Parameter Inversion Analysis Based on EnKF and ABAQUS Data Assimilation Method

作　　者：王昌 曹忠辉 王定 宋辉 WANG Chang;CAO Zhonghui;WANG Ding;SONG Hui(Department of Engineering Mechanics,Hohai University,Nanjing 210098,China;Hunan Provincial Water Transportation Construction&Investment Group Co.Ltd.,Changsha 410118,China)

机构地区：[1]河海大学力学与工程科学学院,南京210098 [2]湖南省水运建设投资集团有限公司,长沙410118

出　　处：《河南科学》2024年第8期1179-1191,共13页Henan Science

摘　　要：岩土参数呈现空间变异性,反演分析能够获取关键地质参数的准确空间分布,对结构安全评估至关重要.提出了一种基于集合卡尔曼滤波(Ensemble Kalman Filter,EnKF)与数值模拟软件ABAQUS的数据同化方法,使用参数随机场作为初始输入,通过集成ABAQUS以及其用户子程序,实现不同工况下的参数反演分析,并且在逐步计算过程中,算法可以利用新获取的观测值加速参数场收敛到真实场.通过与现有Fortran有限元代码框架下的EnKF算法对比,基于EnKF与ABAQUS数据同化方法的更新场更接近于参考场,并且RMSE值约为Fortran有限元代码框架下EnKF算法的55%,证明了算法的有效性.此外,通过参数场分布云图及RMSE值对比,对两个关键参数集合数量和随机场水平与垂直方向相关长度的值,进行了敏感性分析.最后,将该方法应用于双隧洞分步开挖中,其参数场分布云图更接近于参考场,同时RMSE值随更新次数不断减小,更新两次后的RMSE值仅有初始场的29%,证明了其在该类型问题中反演分析的有效性.Geotechnical parameters exhibit spatial variability,and inverse analysis is crucial for obtaining accurate spatial distributions of key geological parameters,which is essential for structural safety assessment.This study proposes a data assimilation method based on the ensemble Kalman filter(EnKF)and the numerical simulation software ABAQUS.Using a random parameter field as the initial input,the method integrates ABAQUS and its user subroutines to perform inverse analysis under different working conditions.During the step-by-step calculation process,the algorithm can utilize newly acquired observations to accelerate the convergence of the parameter field towards the true field.Compared with the EnKF algorithm under the Fortran finite element code framework in existing studies,the updated field based on the EnKF and ABAQUS data assimilation method is closer to the reference field,and the RMSE value is approximately 55% of that under the Fortran finite element code framework,demonstrating the effectiveness of the algorithm.Furthermore,this study conducts a sensitivity analysis of two key parameters:the ensemble size and the ratio of the correlation lengths in the horizontal and vertical directions of the random field,using the distribution cloud maps of the parameter field and RMSE values for comparison.Finally,the method is applied to the stepwise excavation of a twin tunnel,where the parameter field distribution cloud map is closer to the reference field,and the RMSE value continuously decreases with the number of updates,with the RMSE after two updates being only 29% of that of the initial field,proving the effectiveness of the inverse analysis for this type of problem.

关 键 词：空间变异性 地质参数 数据同化 集合卡尔曼滤波 

分 类 号：TB12[理学—工程力学]