基于动态线性分段表示的钻进参数围岩分级特征提取方法  

Classification Feature Extraction Method of Surrounding Rock of Drilling Parameters Based on Dynamic Linear Piecewise Representation

作　　者：何永义 王明年[1,2] 凌学鹏 易文豪 夏覃永 李泽星 童建军[1,2] 赵思光 HE Yongyi;WANG Mingnian;LING Xuepeng;YI Wenhao;XIA Qinyong;LI Zexing;TONG Jianjun;ZHAO Siguang(Key Laboratory of Transportation Tunnel Engineering of Ministry of Education,Southwest Jiaotong University,Chengdu Sichuan 610031,China;School of Civil Engineering,Southwest Jiaotong University,Chengdu Sichuan 610031,China;China Railway No.3 Engineering Group Co.,Ltd.,China Railway Group Limited,Taiyuan Shanxi 030000,China;Railway Engineering Technical Standards Institute,China Railway Economic and Planning Research Institute,Beijing 100038,China)

机构地区：[1]西南交通大学交通隧道工程教育部重点实验室,四川成都610031 [2]西南交通大学土木工程学院,四川成都610031 [3]中国中铁股份有限公司中铁三局集团有限公司,山西太原030000 [4]中国铁路经济规划研究院有限公司铁路工程技术标准所,北京100038

出　　处：《中国铁道科学》2025年第1期96-106,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课题(K2020G035,K2021G024)。

摘　　要：依托来自宜昌—郑万高铁联络线隧道工程的1 765份钻进参数样本,在分析钻进参数时序曲线特征的基础上,结合贝叶斯置信区间检验方法、动态线性分段表示方法、卡尔曼滤波方法和线性分段均值处理方法,提出一种基于动态线性分段表示的钻进参数围岩分级特征提取方法;对比该方法处理前后掌子面钻进参数样本的离散性和差异性,以及采用该方法处理前后6种不同机器学习算法下的围岩分级模型准确性,验证该方法的应用效果。结果表明:钻进参数时序体现出明显的纵向分段、区间波动和随机离散特征;采用该方法处理后,相同围岩级别下的样本数据标准差平均降低28.72%~82.68%,不同围岩级别下的样本类间距离均值提升66.79%~77.37%,6种机器学习算法下围岩分级模型得到的分级准确率由85.3%~88.8%提高到88.1%~89.9%。作为一种基础数据处理方法,该方法能够避免各种非地质因素对围岩分级精度的影响,较好地体现了钻进参数和围岩质量间的良好响应关系,并提升了具体实践中的围岩质量评价准确性。Relying on 1765 drilling parameter samples from the Yichang-Zhengwan High-Speed Railway tie-line tunnel project,based on the analysis of the time series curve characteristics of drilling parameters,a classification feature extraction method of surrounding rock of drilling parameters based on dynamic linear piecewise representation is proposed by combining Bayesian confidence interval test,dynamic linear piecewise representation,Kalman filtering and linear piecewise mean processing methods.The application effect of this method is verified through comparing discreteness and difference of drilling parameter samples of tunnel face before and after processing,and the accuracy of surrounding rock classification model under 6 different kinds of machine learning algorithms before and after processing.The results show that the time series of drilling parameters has obvious characteristics of longitudinal piecewise,interval fluctuation and random dispersion.After processing with this method,the average standard deviation of sample data under the same surrounding rock level is reduced by 28.72%-82.68%,and the average distance between sample classes under different surrounding rock level is increased by 66.79%-77.37%.The classification accuracy of surrounding rock classification model under 6 machine learning algorithms is improved from 85.3%-88.8%to 88.1%-89.9%.As a basic data processing method,it can avoid the influence of various non-geological factors on the classification accuracy of surrounding rock,better reflect the good response relationship between drilling parameters and surrounding rock quality,and improve the accuracy of surrounding rock quality evaluation in concrete practice.

关 键 词：隧道 围岩分级 钻进参数 时间序列 动态线性分段表示 贝叶斯置信区间检验 卡尔曼滤波 

分 类 号：U451.2[建筑科学—桥梁与隧道工程] U452.27[交通运输工程—道路与铁道工程]