摘要
针对地质雷达图像解译依赖于专家经验、费时费力且易受主观因素影响精度的问题,提出基于卷积神经网络理论算法针对超前预报中富水破碎带自动化定位预测方法:设计了残差网络结构、损失函数与训练策略;通过对云南文马高速公路中3条重点隧道的地质雷达图像进行收集筛选,构建了富水破碎带深度特征提取的超前预报网络模型;在Pytorch深度学习框架下,采用训练预热策略及自适应矩估计(Adam)优化器实现了模型参数稳定收敛的高效训练.通过测试验证及对比实验,表明这种方法对隧道地质雷达图像中的富水破碎带不良地质体特征具有较好的检测精度.文马高速望城坡隧道实践证明,该方法可辅助判定隧道施工过程中富水破碎带,识别定位不良地质区域及概率置信度,为实际工程提供决策依据.
Aiming at the accuracy problem of geological radar image interpretation that relies on expert experience,is time-consuming and labor-intensive,and is easily affected by subjective factors,a method based on convolutional neural network theory is proposed to automatically locate and predict the water-rich fracture zone in advanced prediction:a residual network structure is designed,loss function and training strategy;through the collection and screening of geological radar images of three key tunnels in the Wenma Expressway,an advanced prediction network model for the extraction of deep features of the water-rich fracture zone is constructed;under the Pytorch deep learning framework,training is adopted Warm-up strategy and adaptive moment estimation(Adam)optimizer realize efficient training of stable convergence of model parameters.Test verification and comparative experiments show that this method has good detection accuracy for the poor geological features of the water-rich fracture zone in the tunnel geological radar image.The practice of the Wangchengpo Tunnel of Wenma Expressway has proved that this method can assist in the determination of the water-rich fracture zone during the tunnel construction process,identify poorly located geological areas and probability confidence,and provide a basis for decision-making in actual projects.
作者
陈培帅
袁青
张子平
杨林
陈再励
吴立
CHEN Peishuai;YUAN Qing;ZHANG Ziping;YANG Lin;CHEN Zaili;WU Li(CCCC Second Harbor Engineering Co.,Ltd.,Wuhan 430040,China;Faculty of Engineering,China University of Geosciences,Wuhan 430074,China;Research and Development Center of Transport Industry of Intelligent Manufacturing Technologies of Transport Infrastructure,Wuhan 430040,China)
出处
《应用基础与工程科学学报》
EI
CSCD
北大核心
2022年第1期196-207,共12页
Journal of Basic Science and Engineering
基金
国家自然科学基金项目(41672260)。
关键词
隧道工程
卷积神经网络
富水破碎带
地质雷达
超前地质预报
不良地质体
图像解译
tunnel engineering
convolutional neural network
water-rich fracture zone
ground penetrating radar
geological advance prediction
unfavorable geologic bodies
image interpretation