摘要
为解决隧道超前钻探地质预报在解译过程中存在的主观判断性强、定量数据利用率低、评判标准不统一等问题,通过引入机器学习中的极限梯度提升集成算法模型(extreme gradient boosting,XGBoost),结合钻探数据开展隧道围岩完整程度与围岩级别的双层质量评价研究。一方面,采取数据降噪、等距分割、二级指标计算等数据预处理手段对11 233条原始钻探采样数据进行规律发掘和质量提升;另一方面,结合遗传算法(genetic algorithm,GA)与分类器链(classifier chains,CC)构建GA-CC-XGBoost模型,实现复杂机器学习模型的超参数组合自动寻优以及多标签分类的内在相关性考虑。最终所构建训练集的完整程度与围岩级别2项标签的分类准确率分别为95.91%、97.95%,综合分类准确率为93.88%。经过实际隧道工程应用表明,该模型预测结果满足现场超前钻探地质预报的解译需求。
The interpretation of advance drilling forecasts for tunnels involves many problems, such as highly subjective judgment, low quantitative data utilization, and inconsistent evaluation standards. Therefore, the extreme gradient boosting ensemble model( XGBoost) in machine learning is introduced, and the double-layer( integrity and classification) quality evaluation of the surrounding rock of the tunnel is conducted based on the drilling data. First, the 11 233 original drilling sampling data are preprocessed to improve data quality using data noise reduction, equidistant segmentation, and secondary index calculation methods. Second, a genetic algorithm(GA)-classifier chain(CC)-XGBoost model is constructed. This model realizes the automatic optimization of the hyperparameter combination of the complex machine-learning model and considers the inherent relevance of multilabel classification. The classification accuracy of the GA-CC-XGBoost model in the double-layer label is 95.91% and 97.95%, and the comprehensive classification accuracy is 93.88%. The actual tunnel engineering application indicates that the prediction results of the model meet the interpretation requirements of geological forecasts for on-site advance drilling.
作者
梁铭
彭浩
解威威
宋冠先
朱孟龙
张亚飞
LIANG Ming;PENG Hao;XIE Weiwei;SONG Guanxian;ZHU Menglong;ZHANG Yafei(Guangxi Road and Bridge Engineering Group Co.,Ltd.,Nanning 530000,Guangxi,China)
出处
《隧道建设（中英文）》
CSCD
北大核心
2022年第8期1443-1452,共10页
Tunnel Construction
关键词
隧道工程
超前地质预报
水平钻探
模型优化
多标签分类
tunnel engineering
advance geological forecast
horizontal drilling
model optimization
multilabel classification
作者简介
第一作者:梁铭(1992-),男,广西河池人,2018年毕业于中南大学,地质工程专业,硕士,工程师,现从事隧道超前地质预报的定量解译技术工作。E-mail:csuliangming@csu.edu.cn;通信作者:解威威,E-mail:214395494@qq.com。
