基于时空特征的隧道拱顶沉降预测研究  

Tunnel Vault Settlement Prediction Based on Spatiotemporal Characteristics

作　　者：刘健 孙兴凯 陈广库 李尚宣 贾鹏蛟 陈城 LIU Jian;SUN Xingkai;CHEN Guangku;LI Shangxuan;JIA Pengjiao;CHEN Cheng(China Railway 18th Bureau Group First Engineering Co.,Ltd.,Baoding 072750,China;School of Rail Transportation,Soochow University,Suzhou 215000,China;Institute of Intelligent Manufacturing and Smart Transportation,Suzhou City University,Suzhou 215104,China)

机构地区：[1]中铁十八局集团第一工程有限公司,河北保定072750 [2]苏州大学轨道交通学院,江苏苏州215006 [3]苏州城市学院智能制造与智慧交通学院,江苏苏州215104

出　　处：《徐州工程学院学报（自然科学版）》2024年第4期11-17,共7页Journal of Xuzhou Institute of Technology(Natural Sciences Edition)

基　　金：国家自然科学基金项目(52108380)。

摘　　要：针对隧道拱顶沉降问题,基于深圳某暗挖隧道工程,提出一种结合卷积神经网络(CNN)、时间卷积神经网络(TCN)和注意力机制(Attention Mechanism)的新型拱顶沉降预测模型(CTA).首先,用CNN模型提取数据的空间特征,以挖掘不同特征对沉降的影响;然后,利用TCN模型捕捉沉降数据的时序特性,提升模型计算效率;最后,利用注意力机制捕捉重要时间节点信息,得出预测结果.CTA模型得到的MAE、RMSE值最小、R2值最大的结果表明CTA模型的预测效果最优,可以准确预测隧道拱顶沉降.To address the issue of tunnel vault settlement,a novel settlement prediction model(CTA),combining convolutional neural network(CNN),temporal convolutional network(TCN),and Attention Mechanism,is proposed based on a certain underground excavation tunnel project in Shenzhen.First,the CNN model is employed to extract spatial characteristics of the data,with a view to investigating the impact of different characteristics on settlement.Then,the TCN model captures the temporal characteristics of settlement data to enhance computational efficiency.Finally,the Attention Mechanism captures important temporal node information,thus obtaining prediction results.The CTA model obtains the smallest MAE and RMSE values and the highest R2 value,indicating that the CTA model provides optimal prediction performance and can accurately predict tunnel vault settlement.

关 键 词：暗挖隧道 拱顶沉降 深度学习 注意力机制 时空特征 

分 类 号：TU472[建筑科学—结构工程]