基于支持向量机的大断面岩质隧道掌子面围岩非均一性判识方法  

Inhomogeneity Identification Method for Surrounding Rock of Large-Section Rock Tunnel Face Based on Support Vector Machine

作　　者：易文豪 王明年[1,2] 童建军[1,2] 赵思光 李佳旺 桂登斌 张霄 YI Wenhao;WANG Mingnian;TONG Jianjun;ZHAO Siguang;LI Jiawang;GUI Dengbin;ZHANG Xiao(School of Civil Engineering,Southwest Jiaotong University,Chengdu Sichuan 610031,China;Key Laboratory of Transportation Tunnel Engineering,Ministry of Education,Southwest Jiaotong University,Chengdu Sichuan 610031,China)

机构地区：[1]西南交通大学土木工程学院,四川成都610031 [2]西南交通大学交通隧道工程教育部重点试验室,四川成都610031

出　　处：《中国铁道科学》2021年第5期112-122,共11页China Railway Science

基　　金：中国国家铁路集团有限公司科技研究开发计划重大课题(K2018G014,K2020G035);国家自然科学基金资助项目(51878567,51878568)。

摘　　要：为快速判识高速铁路大断面岩质隧道施工过程中常表现出的掌子面围岩显著非均一性特征,提出1种大断面岩质隧道掌子面围岩非均一性判识方法。依托郑万高铁5条试验隧道,采集得到掌子面围岩炮眼钻孔过程样本299份,相应形成大断面岩质隧道掌子面围岩分级样本库;分别构建基于支持向量机(SVM)和2种神经网络的大断面岩质隧道掌子面围岩分级模型,采用同样的数据对比3种模型性能;依托SVM分级模型,利用炮孔名义分级和名义分区分级名词描述掌子面围岩非均一性判识方法,并针对5条试验隧道进行判识计算,提出大断面岩质隧道支护结构的局部优化建议。结果表明:相比于2种神经网络分级模型,SVM分级模型在样本数量较少条件下的分级准确度更高,平均准确度为87.9%;更换样本后,SVM分级模型准确度依然如此,且泛化性更强;本文方法判识得到2次爆破过程中的围岩显著非均一性特征,判识结果与爆破终点掌子面实际情况基本一致,可为支护参数局部优化提供依据。In order to quickly identify the significant non-homogeneous characteristics of the tunnel face surrounding rock often revealed during the construction of large-section rock tunnel in high-speed railway,a method was proposed to identify the inhomogeneity for the surrounding rock of large-section rock tunnel face.299 samples of blasthole drilling process in the working face of Zhengzhou-Wanzhou high-speed railway tunnel were collected to form a classification sample library for the surrounding rock of large-section rock tunnel accordingly.The classification models for the surrounding rock of large-section rock tunnel face were constructed respectively by using support vector machine(SVM)and 2 neural networks,and the performance of the 3 models were compared with the same data.Based on the SVM classification model,the nominal classification and nominal zoning classification terms of blasthole were used to describe the inhomogeneity identification method for the surrounding rock of tunnel face,The identification calculation was carried out for 5 test tunnels of Zhengzhou-Wanzhou high-speed railway to propose the local optimization suggestions for the support structure of large-section rock tunnel.The results show that compared with the 2 neural network classification models,the SVM classification model has higher accuracy under the condition of fewer samples,with an average accuracy of 87.9%.After replacing the samples,the SVM classification model still has higher accuracy and stronger generalization.The method used in this paper identifies the significant inhomogeneity characteristics of the surrounding rock during the two blasting processes,and the results are basically consistent with the actual tunnel face situation at the end of blasting,which can provide a basis for the local optimization of support parameters.

关 键 词：岩质隧道 围岩 非均一性 支持向量机 郑万高铁 

分 类 号：U458.3[建筑科学—桥梁与隧道工程]