|本期目录/Table of Contents|

[1]钟兴润,周明宇,赵江平.基于GTCN-Transformer的地铁深基坑地表沉降预测方法研究*[J].中国安全生产科学技术,2026,22(5):100-109.[doi:10.11731/j.issn.1673-193x.2026.05.012]
　Zhong Xingrun,Zhou Mingyu,Zhao Jiangping.Research on a GTCN-Transformer-based method for predicting ground surface settlement induced by metro deep foundation pits[J].Journal of Safety Science and Technology,2026,22(5):100-109.[doi:10.11731/j.issn.1673-193x.2026.05.012]
点击复制

基于GTCN-Transformer的地铁深基坑地表沉降预测方法研究*

《中国安全生产科学技术》[ISSN:1673-193X/CN:11-5335/TB]

卷:
22
期数:
2026年5期
页码:
100-109
栏目:
安全工程技术
出版日期:
2026-05-30

文章信息/Info

Title:
Research on a GTCN-Transformer-based method for predicting ground surface settlement induced by metro deep foundation pits
文章编号:
1673-193X(2026)-05-0100-10
作者:
钟兴润; 周明宇; 赵江平
（西安建筑科技大学 资源工程学院，陕西 西安 710055）
Author(s):
Zhong Xingrun;  Zhou Mingyu;  Zhao Jiangping
(School of Resources Engineering,Xi’an University of Architecture and Technology,Xi’an Shaanxi 710055,China)
关键词:
地铁深基坑; 地表沉降预测; 图卷积网络; 时序卷积网络; Transformer
Keywords:
metro deep foundation pit;  ground surface settlement prediction;  graph convolutional network;  temporal convolutional network;  Transformer
分类号:
X931
DOI:
10.11731/j.issn.1673-193x.2026.05.012
文献标志码:
A
摘要:
针对地铁深基坑施工环境复杂、周期长及影响基坑安全因素众多等问题，以地表沉降这一反映基坑安全状态的重要指标为研究对象，旨在实现其高精度预测。同时针对现有深基坑地表沉降预测方法未充分考虑监测点空间相关性、预测精度不足的局限，提出1种基于自注意力机制与深度学习技术的时空协同预测模型（graph convolutional network-temporal convolutional network-transformer, GCN-TCN-Transformer），以下简称GTCN-Transformer。该模型构建空间、时间与自相关3个模块协同作用的端到端联合框架，采用图卷积网络提取监测点间动态空间相关信息，运用时序卷积网络提取沉降数据时间特征，并引入Transformer学习沉降序列内部自相关性，实现时空特征协同融合预测；以西安市某地铁深基坑开挖工程地表沉降监测数据为例，开展多监测点联合预测验证。研究结果表明：与GTCN-Transformer在各监测点均取得最优预测性能，MAE与RMSE均低于0.112和0.091,R2均高于0.994，模型具有更高的准确度与稳定性。研究结果可为深基坑地表沉降预测提供准确、稳定的时空联合方法，为地铁深基坑施工安全控制提供参考。
Abstract:
Aiming at the problems of complex construction environment,long period and many factors affecting the safety of metro deep foundation pits,the surface settlement,which is an important index reflecting the safety state of foundation pit,is designed to achieve high-precision prediction.At the same time,in view of the limitations of the existing deep foundation pit surface settlement prediction method that does not fully consider the spatial correlation of monitoring points and the lack of prediction accuracy,a spatio-temporal collaborative prediction model based on self-attention mechanism and deep learning technology (graph convolutional network-temporal convolutional network-transformer,GCN-TCN-Transformer),hereinafter referred to as GTCN-Transformer,is proposed.The model constructs an end-to-end joint framework for the synergy of three modules of space,time and autocorrelation.The graph convolution network is used to extract the dynamic spatial correlation information between monitoring points,and the temporal convolution network is used to extract the temporal characteristics of settlement data.The transformer is introduced to model the internal autocorrelation of settlement sequence,so as to realize the collaborative fusion prediction of spatial-temporal characteristics;Taking the surface settlement monitoring data of a metro deep foundation pit excavation project in Xi’an as an example,the joint prediction and verification of multiple monitoring points were carried out.The results show that: the GTCN-Transformer achieved the best predictive performance across all monitoring points, with MAE and RMSE below 0.112 and 0.091,respectively,and R2 exceeding 0.994.The model has higher prediction accuracy and stability.The research results can provide an accurate and stable time-space joint method for the prediction of surface settlement of deep foundation pit,and provide a reference for safety control of subway deep foundation pit construction.

参考文献/References:

［1］张宇航,戎思达,李慧,等.基于动态监测与事故树分析的超大深基坑风险分析方法研究［J］.中国安全生产科学技术,2023,19(9):89-95. Zhang Yuhan,Rong Sida,Li Hui,et al.Research on risk analysis methods for super large deep foundation pits based on dynamic monitoring and fault tree analysis［J］.Journal of Safety Science and Technology,2023,19(9):89-95.
［2］蒙国往,农忠建,吴波,等.地铁车站深基坑开挖变形及数值模拟分析［J］.中国安全生产科学技术,2020,16(7):145-151. Meng Guowang,Nong Zhongjian,Wu Bo,et al.Deformation and numerical simulation analysis of deep foundation pit excavation in metro stations［J］.Journal of Safety Science and Technology,2020,16(7):145-151.
［3］石杰红,史聪灵,刘晶晶.双线地铁隧道下穿管道安全性对比研究［J］.中国安全生产科学技术,2019,15(8):113-117. Shi Jiehong,Shi Congling,Liu Jingjing.Comparative study on the safety of underpass pipelines in double line subway tunnels［J］.Journal of Safety Science and Technology,2019,15(8):113-117.
［4］赵亚红,王伟娜,江培华,等.马尔可夫链改进的MMF沉降预测模型及应用［J］.测绘通报,2022(1):79-83. Zhao Yahong,Wang Weina,Jiang Peihua,et al.Markov chain improved MMF settlement prediction model and its application［J］.Bulletin of Surveying and Mapping,2022(1):79-83.
［5］Wang Changyu,Ding Zude,Shen Yuhang,et al.Prediction of surface settlement caused by tunneling with ARMA based time-series decomposition［J］.Tunnelling and Underground Space Technology,2025,164:106770.
［6］田双,骆寅,李雨株,等.新建深基坑工程施工及降水过程对既有地铁结构的影响［J］.中国安全生产科学技术,2023,19(增刊1):130-135. Tian Shuang,Luo Yin,Li Yuzhu,et al.Impact of new deep foundation pit construction and dewatering on existing metro structures［J］.Journal of Safety Science and Technology,2023,19(Supplement 1):130-135.
［7］张亮亮,程桦,姚直书,等.基于改进Knothe时间模型的地表最大沉降速度预测［J］.岩土力学,2023,44(4):1111-1119. Zhang Liangliang,Cheng Hua,Yao Zhishu,et al.Prediction of maximum surface subsidence velocity based on improved Knothe time model［J］.Rock and Soil Mechanics,2023,44(4):1111-1119.
［8］Yang Peixi,Yong Weixun,Li Chuanqi,et al.Hybrid random forest-based models for earth pressure balance tunneling-induced ground settlement prediction［J］.Applied Sciences,2023,13(4):2574.
［9］崔铁军,马云东.基于差异进化支持向量机的坑外土体沉降预测［J］.中国安全科学学报,2013,23(1):83-89. Cui Tiejun,Ma Yundong.Prediction of soil settlement outside the pit based on differential evolution support vector machine［J］.Journal of Safety Science and Technology,2013,23(1):83-89.
［10］Bengio Y,Simard P,Frasconi P.Learning long-term dependencies with gradient descent is difficult［J］.IEEE Transactions on Neural Networks,1994,5(2):157-166.
［11］Hochreiter S,Schmidhuber J.Long short-term memory［J］.Neural Comput,1997,9(8):1735-1780.
［12］Cho K,Van Merrienboer B,Gulcehre C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation［C］//Proceedings of EMNLP 2014.Doha:Association for Computational Linguistics,2014:1724-1734.
［13］Frassini E,Vijfvinkel T S,Butler R M,et al.Deep learning methods for clinical workflow phase-based prediction of procedure duration:a benchmark study［J］.Computer Assisted Surgery,2025,30(1):2466426.
［14］Lu Yao,Vijayananth V,Perumal T.Smart home energy prediction framework using temporal Kolmogorov-Arnold transformer［J］.Energy and Buildings,2025,335:115529.
［15］Wang Yongdong,Zhai Haonan,Cao Xianghong,et al.A novel accident duration prediction method based on a conditional table generative adversarial network and transformer［J］.Sustainability,2024,16(16):6821.
［16］Bai Shaojie,Kolter J Z,Koltun V.Convolutional sequence modeling revisited［C］//ICLR 2018 Workshop.Vancouver:OpenReview.net,2018.
［17］Ren Jiansi,Wu Wei,Liu Gang,et al.Bidirectional gated temporal convolution with attention for text classification［J］.Neurocomputing,2021,455:265-273.
［18］Hou Jingyong,Xie Lei,Zhang Shilei.Two-stage streaming keyword detection and localization with multi-scale depthwise temporal convolution［J］.Neural Networks,2022,150:28-42.
［19］Yan Xiongfeng,Ai Tinghua,Yang Min,et al.A graph convolutional neural network for classification of building patterns using spatial vector data［J］.ISPRS Journal of Photogrammetry and Remote Sensing,2019,150:259-273.
［20］Zheng Jin,Wang Yang,Xu Wanjun,et al.GSSA:pay attention to graph feature importance for GCN via statistical self-attention［J］.Neurocomputing,2020,417:458-470.
［21］Li Shu,Li Wentao,Wang Wei.Co-GCN for multi-view semi-supervised learning［C］//Proceedings of the AAAI Conference on Artificial Intelligence.New York:AAAI Press,2020:4691-4698.
［22］Vaswani A,Shazeer N,Parmar N,et al.Attention is all you need［C］//Advances in Neural Information Processing Systems 30.New York:Curran Associates,Inc.,2017:5998-6008.
［23］Ge Qi,Li Jin,Wang Xiaohong,et al.LiteTransNet:An interpretable approach for landslide displacement prediction using transformer model with attention mechanism［J］.Engineering Geology,2024,331:107446.
［24］Graves A.Supervised sequence labelling with recurrent neural networks［M］.Springer Berlin Heidelberg,2012.
［25］Lujan-Moreno G A,Howard P R,Rojas O G,et al.Design of experiments and response surface methodology to tune machine learning hyperparameters,with a random forest case-study［J］.Expert Systems with Applications,2018,109:195-205.

相似文献/References:

[1]叶万军,成炜康,陈笑楠,等.砂卵石地层大直径盾构工程地表沉降深度学习预测*[J].中国安全生产科学技术,2023,19(8):124.[doi:10.11731/j.issn.1673-193x.2023.08.018]
　YE Wanjun,CHENG Weikang,CHEN Xiaonan,et al.Deep learning and prediction on surface subsidence of large-diameter shield project in sandy cobble stratum[J].Journal of Safety Science and Technology,2023,19(5):124.[doi:10.11731/j.issn.1673-193x.2023.08.018]
[2]吴波,农宇,蒙国往,等.基于FBN地铁深基坑施工渗漏风险评估模型及应用*[J].中国安全生产科学技术,2022,18(5):178.[doi:10.11731/j.issn.1673-193x.2022.05.027]
　WU Bo,NONG Yu,MENG Guowang,et al.Risk assessment model of leakage in deep foundation pit construction of subway based on FBN and its application[J].Journal of Safety Science and Technology,2022,18(5):178.[doi:10.11731/j.issn.1673-193x.2022.05.027]

备注/Memo

备注/Memo:
收稿日期： 2026-01-30；修回日期：2026-02-27；网络首发日期： 2026-04-14
* 基金项目： 陕西省住房城乡建设科技计划项目（2020-K32）；西安建科大工程技术项目（XAJD-YF23N010）
作者简介： 钟兴润，博士研究生，讲师，主要研究方向为建筑安全工程与结构安全检测与鉴定。
通信作者： 周明宇，硕士研究生，主要研究方向为基坑变形监测。
更新日期/Last Update: 2026-06-03