摘要
为了减少盾构掘进对周围地层及管片姿态的影响,盾构姿态需要尽可能与设计轴线吻合,以减少盾构实时位置与设计值间的偏差。开发了一种基于混合深度学习模型的预测方法,提出了一种盾构隧道中实时姿态和位置的动态预测框架,对盾构姿态和位置的4个关键参数进行预测。该框架包含Kriging插值、小波变换(wavelet transform, WT)噪声滤波器、带通道注意力机制的卷积神经网络(convolutional neural networks with channel attention, CNN-CA)以及用于确定盾构机未来姿态和位置的长-短期记忆网络(long short-term memory, LSTM)。基于上海市域铁路工程中盾构收集的数据集对该模型的性能进行验证,并研究了CNN-CA网络对模型的影响。研究表明,该模型在盾构的姿态预测方面效果较好,CNN-CA网络能够有效提取预测值所需的特征,对计算精度贡献较大。
In order to reduce the impact of the shield tunneling on the surrounding strata and segments attitude,the shield attitude needs to match the design axis as much as possible to reduce the correction and adjustment of the shield s real-time position.A prediction method based on a hybrid deep learning model was developed,and a dynamic prediction framework for real-time attitude and position in shield tunnel was proposed to predict four key parameters of shield attitude and position.The framework contains Kriging interpolation,wavelet transform(WT)noise filter,convolutional neural networks with channel attention(CNN-CA),and long short-term memory(LSTM)for determining the future attitude and position of the shield.The model performance was validated based on the dataset collected from shield structures in Shanghai metropolitan railroad project,and the contribution of CNN-CA to the model was investigated.Research demonstrates that the model is effective in predicting the attitude of shield,and the CNN-CA can effectively extract the features required for the predicted values,which contributes greatly to computational accuracy.
作者
李培楠
刘学
戴泽余
范杰
寇晓勇
翟一欣
LI Peinan;LIU Xue;DAI Zeyu;FAN Jie;KOU Xiaoyong;ZHAI Yixin(College of Environmental Science and Engineering,Donghua University,Shanghai 201620,China;Shanghai Tunnel Engineering Co.,Ltd.,Shanghai 200032,China)
出处
《东华大学学报（自然科学版）》
CAS
北大核心
2024年第3期145-152,共8页
Journal of Donghua University(Natural Science)
基金
上海市科技创新行动计划社发展科技攻关项目(21DZ1201105)
中央高校基本科研业务费专项资金(21D111320)
关键词
盾构施工
人工智能
姿态和位置
混合深度学习
动态预测
shield construction
artificial intelligence
attitude and position
hybrid deep learning
dynamic prediction
作者简介
通信作者:李培楠,男,副教授,主要从事研究方向为隧道及地下建筑工程等方面的教学与研究工作,E-mail:lipeinan@dhu.edu.cn。
