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
刘健 1孙兴凯 1陈广库 1李尚宣 2贾鹏蛟 2陈城3
作者信息
- 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
- 折叠
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.
Abstract
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Key words
underground excavation tunnel/vault settlement/deep learning/Attention Mechanism/spatial-temporal characteristics引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据