基于EMD-多层LSTM的盾构隧道长期沉降预测  

Long-Term Settlement Prediction of Shield Tunnel Based on EMD-Multilayer LSTM

作　　者：沈一鸣 张冬梅[1,2] 黄忠凯[1] 朱美恒 Shen Yiming;Zhang Dongmei;Huang Zhongkai;Zhu Meiheng(College of Civil Engineering,Tongji University,Shanghai 200092,P.R.China;State Key Laboratory of Disaster Reduction in Civil Engineering,Tongji University,Shanghai 200092,P.R.China;CCCC Tunnel Engineering Bureau Co.,Ltd.,Shanghai 200332,P.R.China)

机构地区：[1]同济大学土木工程学院,上海200092 [2]同济大学土木工程防灾减灾全国重点实验室,上海200092 [3]中交隧道工程局有限公司,上海200332

出　　处：《地下空间与工程学报》2025年第2期684-694,共11页Chinese Journal of Underground Space and Engineering

基　　金：国家重点研发计划课题(2022YFC3800905);上海市“科技创新行动计划”优秀学术/技术带头人计划项目(22XD1430200);同济大学学科交叉联合攻关项目(2022-3-ZD-07)。

摘　　要：长期沉降是运营期盾构隧道纵向结构安全的重要影响因素之一。为了更精确地预测运营期盾构隧道长期沉降发展情况,本文采用深度学习算法提出了基于经验模态分解-多层长短期记忆人工神经网络(EMD-多层LSTM)的运营期盾构隧道长期沉降预测方法。建立的方法主要通过数据前处理、经验模态分解、预测模型构建和结果重构4个关键步骤实现对运营期盾构隧道长期沉降的预测,并成功应用于上海地铁10号线多个监测点的长期沉降预测分析中。结果表明:(1)相较于传统LSTM和支持向量机(SVM)等基准算法,本研究所提出的模型能够更有效的预测运营期盾构隧道长期沉降的动态发展情况,模型泛化能力显著提升,长期沉降预测结果精度良好,最大绝对预测误差可控制在1.3 mm以内;(2)通过模型适用性能分析表明,本文建立的EMD-多层LSTM预测模型适用性能好,能在不同情况下始终保持模型的预测结果真实可信。本文提出的预测方法为运营期盾构隧道长期沉降预测提供了一种新的解决途径。Long-term settlement is one of the significant factors affecting the longitudinal structure safety of shield tunnel during operation.In order to more accurately predict the long-term settlement development of shield tunnel during operation,relying on deep learning algorithms,this paper proposes a long-term settlement prediction method of shield tunnel based on empirical mode decomposition and multi-layer long short-term memory artificial neural network(EMD-Multi-layer LSTM).The proposed method is designed to achieve the long-term settlement prediction of shield tunnel during operation through four key steps,namely data pre-processing,empirical mode decomposition,prediction model construction and result reconstruction,and has been successfully applied to the long-term settlement prediction analysis of multiple monitoring points in Shanghai Metro Line 10.The results indicate that:(1)Compared with the benchmark algorithms like traditional LSTM and Support Vector Machine(SVM),the model proposed in this study can more effectively predict the dynamic development of the shield tunnel's long-term settlement during operation.The generalization ability of the model is significantly improved,and the prediction accuracy of the long-term settlement is excellent,of which the maximum absolute prediction error can be controlled within 1.3 mm.(2)Meanwhile,the applicability performance analysis shows that the EMD-Multi-layer LSTM prediction model established in this paper has excellent applicability and the prediction results are always reliable under different circumstances.In general,the prediction method proposed in this study provides a new solution for the long-term settlement prediction of shield tunnel during operation,and relevant research can provide early warning and guidance for the maintenance of shield tunnel during operation.

关 键 词：盾构隧道 长期沉降 经验模态分解 长短期记忆人工神经网络 

分 类 号：U121[交通运输工程]