基于BP神经网络的隧道下穿既有运营地铁线注浆施工参数预测——以广州地铁18号线隧道注浆工程为例  

Prediction of Grouting Construction Parameters of Tunnel Undercrossing Existing Subway Lines Based on BP Neural Network

作　　者：梁粤华 翟利华 刘旭 张聪 王树英[3] LIANG Yue-hua;ZHAI Li-hua;LIU Xu;ZHANG Cong;WANG Shu-ying(Guangzhou Metro Design&Research Institute Co.,Ltd.,Guangzhou 510010,China;School of Civil Engineering,Central South University of Forestry and Technology,Changsha 410004,China;School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]广州地铁设计研究院股份有限公司,广州510010 [2]中南林业科技大学土木工程学院,长沙410004 [3]中南大学土木工程学院,长沙410075

出　　处：《科学技术与工程》2023年第7期3045-3052,共8页Science Technology and Engineering

基　　金：广东省城市轨道交通工程建造新技术企业重点实验室资助项目(2017B030302009)。

摘　　要：新建隧道下穿既有运营地铁线施工过程中极易对既有运营地铁线产生不利影响,而广泛采用的超前预注浆尚处于以经验性选取注浆施工参数的阶段,导致工程事故频发。为此,首先以开挖段地层物性参数、地层位移变化值作为输入层,注浆施工参数为输出层,构建了基于BP(back propagation)神经网络的注浆施工参数预测模型;其次,以MAPE(mean absolute percentage error)作为预测精度评价指标,采取试算法对BP神经模型参数(隐含层节点数目、学习率)进行了探讨;最后,将提出的BP神经网络用于指导工程实践。研究结果表明:当BP神经网络预测模型隐含层节点数为9、学习率为0.01、训练次数为20 000以及精度目标值为1×10^(-4)时,模型适用性评价显示预测值与监测值之间最大相对误差为19,平均相对误差均低于13,说明提出的BP神经网络预测模型可行;进一步的工程应用结果表明:采用预测的注浆施工参数进行注浆后掌子面稳定、开挖过程中未发生隧道塌方等事故,满足相关规范要求。研究成果也可在隧道下穿其他结构或建筑物灾害防控注浆工程中得到推广应用。The construction processes of new tunnel crossing the existing operation subway line is easy about adversely affecting the existing operation subway line,and the widely used advanced pre-grouting is still in the stage of empirical selection of grouting construction parameters,resulting in frequent engineering accidents.Therefore,the physical parameters of the excavation section and the change value of the stratum displacement were used as the input layer,and the grouting construction parameters were used as the output layer.A method of determining the grouting construction parameters based on BP neural network was constructed.Secondly,MAPE(mean absolute percentage error)was taken as the prediction accuracy evaluation index,the BP neural model parameters(the number of hidden layer nodes,learning rate,training times,accuracy target value,etc.)were discussed by a trial algorithm.Finally,the proposed BP neural network was used to guide engineering practice.The results show that when the number of hidden layer nodes of BP neural network prediction model are 9,the learning rate is 0.01,the training number is 20000 and the accuracy target value is 1×10^(-4),the applicability evaluation of the model shows that the maximum relative error between the predicted value and the monitoring value is 19,and the average relative error is less than 13,which indicated that the BP neural network prediction model is feasible.The further engineering application results showed that the predicted grouting construction parameters were used to stabilize the tunnel face after grouting,and there were no accidents such as tunnel collapse during excavations.The displacement change values of the existing operating subway line was between-0.5 mm and 1.0 mm,which met the requirements of relevant specifications.The research result can also be popularized and applied to the disaster prevention and control grouting engineering of tunnel under crossing other structures or buildings.

关 键 词：隧道 下穿 既有运营地铁线 BP神经网络 注浆施工参数 

分 类 号：U455[建筑科学—桥梁与隧道工程]