基于神经网络的双模盾构复合地层掘进参数预测模型  

Tunneling Parameter Prediction Model of Dual-mode Shield in Composite Strata Based on Neural Network

作　　者：孙峻枫 陈凡[1] 宋天田 毋晨 方勇[1] 王士民[1] SUN Junfeng;CHEN Fan;SONG Tiantian;WU Chen;FANG Yong;WANG Shimin(Key Laboratory of Transportation Tunnel Engineering of Ministry of Education,Southwest Jiaotong University,Chengdu 610031,China;Shenzhen Metro Group Co.,Ltd.,Shenzhen 518026,China;CCCC First Highway Xiamen Engineering Co.,Ltd.,Xiamen 361021,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,成都610031 [2]深圳市地铁集团有限公司,深圳518026 [3]中交一公局厦门工程有限公司,厦门361021

出　　处：《铁道标准设计》2023年第5期100-108,共9页Railway Standard Design

基　　金：国家重点研发计划项目(2016YFC0802201);国家自然科学基金面上项目(51578461)。

摘　　要：依托深圳市轨道交通12号线怀德站-福永站区间隧道工程,基于EPB/TBM双模盾构穿越地质参数和现场掘进监测数据,采用BP神经网络方法建立双模式盾构掘进参数预测模型,分别对地层参数及掘进模式进行量化,将刀盘扭矩、刀盘转速、螺旋机转速、总推进力、隧道埋深、围岩等级、岩石单轴饱和抗压强度及不同掘进模式作为输入参数,预测出在不同掘进模式及不同地层条件下的设备掘进速率,针对3类典型地层的预测结果进行可视化分析验证,并对预测模型精度进行改进分析。结果表明:神经网络预测模型在TBM模式下的微风化段平均相对误差为8.6%,EPB模式下的强风化段平均相对误差为10.6%,EPB模式下的中风化段平均相对误差26.2%;该模型对强风化段及微风化段等地层强度变化较为稳定的地层预测精度较高,同时,该预测模式适用于22个隐层神经元并对掘进速率采用直接放缩的方法。Relying on the tunnel project between Huaide Station and Fuyong Station of Shenzhen Rail Transit Line 12,the BP neural network method is used to establish the dual-mode shield tunneling parameter prediction model based on the EPB/TBM dual-mode shield tunneling geological parameters and on-site driving monitoring data.The model quantifies the formation parameters and the driving mode,and takes the cutter head torque,cutter head rotation speed,screw machine rotation speed,total propulsion force,tunnel burial depth,surrounding rock grade,rock uniaxial saturated compressive strength and different driving modes as input parameters to predict the equipment excavation rate under different excavation modes and different strata conditions.The visual analysis and verification for the prediction results of three types of typical strata are conducted,and then the accuracy of the prediction model is improved.The results show that the average prediction error of the model under the TBM model is 8.6%for the lightly weathered segment,the average prediction error for the strongly weathered segment under the EPB model is 10.6%,and the average relative error for the moderately weathered segment under the EPB model is 26.2%.The model is more accurate in predicting the stratum of strong weathered segment and light weathered segments,where the stratum intensity changes are relatively stable.At the same time,the prediction model is more suitable for 22 hidden layer neurons and adopts the direct scaling method of the driving rate.

关 键 词：盾构隧道 双模盾构 掘进参数 BP神经网络 复合地层 

分 类 号：TU478[建筑科学—结构工程] U455.43[建筑科学—土工工程]