摘要
盾构施工参数与隧道掘进过程中的稳定关联密切,为避免掘进参数完全依靠经验,并为参数动态调整提供参考及依据。依托南通地铁1号线针对刀盘推力以及刀盘扭矩,将刀盘转速等7个关键参数的历史数据作为训练集,基于循环神经网络(RNN)以及长短期记忆网络(LSTM)两种算法分别建立关于刀盘推力和刀盘扭矩的预测模型。两种模型所用数据集均基于滑动窗口插补进行预处理,RNN模型采用均方误差(MSELoss)作为损失函数,LSTM模型则采用PyTorch库中的均方误差自定义损失函数。结果表明,两种模型对于盾构总推进力的学习预测能力均优于对刀盘扭矩的预测能力,其中RNN模型对于两组参数的自适应能力均优于LSTM模型。
Shield construction parameters are closely related to the stability of the tunnel excavation process,in order to avoid relying on experience to set the excavation parameters,and at the same time to provide a reference basis for the dynamic adjustment of the construction parameters.In this paper,relying on Nantong Metro Line 1,for the setting of cutter thrust and cutter torque,the historical data of 7 parameters such as cutter rotational speed are used as the training set,and the prediction model is established based on two algorithms of recurrent neural network(RNN)and long short-term memory network(LSTM),respectively.The datasets used in both models are preprocessed based on sliding window interpolation.The NN model adopts the mean-squared error(MSELoss)as the loss function,while the LSTM model adopts the mean-squared error customized loss function from the PyTorch library.The results show that the learning prediction ability of both models for the total shield propulsion force is better than that for the blade torque,in which the RNN model has better adaptive ability than the LSTM model for both sets of parameters.
作者
魏良丰
陈大权
毛志超
汪优
WEI Liangfeng;CHEN Daquan;MAO Zhichao;WANG You(Nantong Rail Transit Co.,Ltd.,Nantong 226010,China;The Fifth Company of China Railway 2nd Bureau Group,Chengdu 610091,China;School of Civil Engineering,Central South University,Changsha 410075,China)
出处
《交通科技与经济》
2025年第4期72-77,共6页
Technology & Economy in Areas of Communications
基金
国家自然科学基金项目(51778633)
中铁二局集团有限公司科技开发项目(局2019-重点-04)。
关键词
隧道工程
掘进参数
机器学习
循环神经网络
长短期记忆网络
tunneling
excavation parameters
machine learning
recurrent neural networks
long short-term memory network
