基于LS-SVR算法的多源监测数据高铁隧道围岩参数反分析方法  

Back-analysis method of rock mass properties in tunnel engineering using multiple monitoring data based on LS-SVR algorithm

作　　者：李照众 畅翔宇 王浩[1] 茅建校[1] Li Zhaozhong;Chang Xiangyu;Wang Hao;Mao Jianxiao(Key Laboratory of Concrete and Prestressed Concrete Structures of Ministry of Education,Southeast University,Nanjing 211189,China;China Railway 24th Bureau Group Co.,Ltd.,Shanghai 200071,China)

机构地区：[1]东南大学混凝土及预应力混凝土结构教育部重点实验室,南京210096 [2]中铁二十四局集团有限公司,上海200071

出　　处：《Journal of Southeast University(English Edition)》2023年第1期1-7,共7页东南大学学报（英文版）

基　　金：The National Natural Science Foundation of China(No.51978155);the Science and Technology Program of Ministry of Housing and Urban-Rural Development(No.2020-K-125).

摘　　要：为了准确估计岩体性质,依托阳山高速铁路隧道,提出了一种基于最小二乘支持向量回归(LS-SVR)的多源监测数据高铁隧道围岩参数反分析方法.以均方根误差(RMSE)和绝对百分比误差(MAPE)为评价指标,将参数反分析结果与BP神经网络和高斯过程回归模型结果进行比较.结果表明,对于单一类型的监测数据,考虑拱顶沉降的LS-SVR模型的RMSE和MAPE值最低.随着监测数据类型的增加,LS-SVR反分析模型的RMSE值逐渐减小,且采用拱顶沉降、收敛和仰拱隆起3种监测数据的反分析模型的RMSE值最小.相比于BP神经网络和高斯过程回归模型,LS-SVR模型具有较低的RMSE和MAPE值.相较于现有围岩力学参数反分析方法,考虑多源监测数据的LS-SVR模型具有更高的参数反分析精度.To accurately estimate the rock mass properties of a high-speed railway tunnel,a back-analysis method using multiple monitoring data based on the least-squares support vector regression(LS-SVR)algorithm is presented.The root mean square error(RMSE)and mean absolute percentage error(MAPE)are used as evaluation indices.The results of the parameter estimation are compared with those of the back propagation neural network(BPNN)and Gaussian process regression(GPR).The results show that for the single type of monitoring data,the LS-SVR model with vault settlement has the lowest RMSE and MAPE values.Moreover,as the data type increases,the RMSE value of the LS-SVR decreases,especially for the model with the mixed data of vault settlement,convergence,and floor heave.The comparison results show that the presented model has lower RMSE and MAPE values than BPNN and GPR.The LS-SVR model using multiple monitoring data shows better performance than existing back-analysis methods,improving the accuracy of the estimation of rock mass properties.

关 键 词：隧道工程 参数反分析方法 围岩力学参数 最小二乘支持向量回归算法 

分 类 号：U456[建筑科学—桥梁与隧道工程]