基于门控循环单元的全断面掘进机稳定段掘进性能预测  

Predicting Tunneling Performance of TBM Stable Stage Based on Gated Recurrent Unit

作　　者：张弛[1] 李艳[1] 王鹏[1] 刘沛 梁科森 ZHANG Chi;LI Yan;WANG Peng;LIU Pei;LIANG Ke-sen(Mechanical and Electrical Engineering College,Central South University,Changsha 410083,China)

机构地区：[1]中南大学机电工程学院,长沙410083

出　　处：《科学技术与工程》2022年第32期14443-14450,共8页Science Technology and Engineering

基　　金：国家重点研发计划(2018YFB1702500)。

摘　　要：全断面隧道掘进机(tunnel boring machine, TBM)一个正常掘进循环分为空推段、上升段和稳定段3个阶段,其中稳定掘进段为主要施工阶段,稳定段掘进性能的好坏是TBM掘进的关键。为实现TBM安全高效掘进,建立一种基于门控循环单元(gated recurrent unit, GRU)神经网络的预测模型,预测TBM稳定段掘进性能。模型以新疆某供水工程Ⅱ标段TBM施工数据为依托,5种掘进循环上升段主要参数的时间序列数据作为主要输入,围岩等级作为辅助输入来考虑岩体对掘进性能的影响,输出为稳定段的总推进力和刀盘扭矩,为稳定段TBM性能预判提供参考。为显示预测效果,对比传统循环神经网络(recurrent neural network, RNN)预测模型,并分析不同长度时间序列输入对模型预测精度的影响。结果表明:GRU模型预测拟合优度均在0.9以上,平均绝对百分比误差均小于12.25%,同时能够适用不同长度时序输入。由此可见,所建模型具有较高预测精度,泛化能力较好,能够辅助预判掘进机稳定段性能。A normal tunneling cycle of a tunnel boring machine(TBM)is divided into three stages:empty pushing stage,ascending stage and stable stage.The stable stage is the main construction stage,and the performance of the stable stage is the key to TBM tunneling.In order to achieve safe and efficient tunneling of TBM,a prediction model based on gated recurrent unit(GRU)neural network was established to predict the tunneling performance of the stable section.The model was based on the TBM construction data of section II of a water supply project in Xinjiang,time series data of five main parameters of the ascending stage were taken as the main input and the surrounding rock grade was taken as auxiliary input to consider the influence of the rock mass on the tunneling performance.The output of the model is the predicted values of thrust and torque,which provide a reference for the prediction of TBM performance in the stable stage.In order to show the prediction effect,the traditional recurrent neural network(RNN)prediction model was compared,and the influence of time series input with different lengths on the prediction accuracy of the model was analyzed.The results show that the goodness of fit of GRU model predictions is above 0.9,and the mean absolute percentage error is less than 12.25%.At the same time,it can be applied to time series inputs of different lengths.It can be concluded that the model has high prediction accuracy and good generalization ability,and can assist in predicting the performance of the stable stage of the TBM.

关 键 词：地下工程 全断面掘进机(TBM) 门控循环单元(GRU)神经网络 掘进性能预测 围岩等级 

分 类 号：U455[建筑科学—桥梁与隧道工程]