Wang Jing, Geng Dajiang, Luo Zongli, Zhao Jun, Liu Xin, Zong Hui, Long Minjian
School of Civil Engineering, Chongqing Industry Polytechnic College, Chongqing, China.
China Construction 4th Engineering Bureau 6th Corp. Ltd., Hefei, China.
Sci Prog. 2025 Apr-Jun;108(2):368504251349975. doi: 10.1177/00368504251349975. Epub 2025 Jun 18.
To enable real-time prediction of tunnel deformation induced by foundation pit excavation, this study develops a predictive framework that integrates an analytical solution based on a two-stage unloading method with a back-propagation neural network for parameter identification. The analytical model accounts for excavation-induced stress redistribution using Mindlin's solution, while the tunnel is modeled as a beam resting on a double-sided elastic foundation to represent its deformation behavior. Five key mechanical parameters are considered for inversion, including the initial residual stress coefficient, unloading reduction coefficient, longitudinal stiffness reduction factor, and the vertical and horizontal subgrade reaction moduli. These parameters are calibrated using field monitoring data through a trained back-propagation neural network. The proposed framework is embedded into a real-time prediction system and applied to the Gubei Road Station of Shanghai Metro Line 15, which is located adjacent to Line 10. The predicted tunnel displacements exhibit strong agreement with field measurements, with maximum relative errors of 21.82% in the vertical direction and 1.95% in the horizontal direction. Except during phases of displacement trend reversal, the system consistently maintains high predictive accuracy. These results verify the reliability of the proposed method for forecasting excavation-induced tunnel deformation and underscore its applicability in proactive risk management for urban tunneling projects.
为实现对基坑开挖引起的隧道变形的实时预测，本研究开发了一个预测框架，该框架将基于两阶段卸载法的解析解与用于参数识别的反向传播神经网络相结合。解析模型采用明德林解来考虑开挖引起的应力重分布，而隧道则被建模为置于双面弹性地基上的梁，以表征其变形行为。反演考虑了五个关键力学参数，包括初始残余应力系数、卸载折减系数、纵向刚度折减系数以及竖向和水平地基反力模量。通过训练后的反向传播神经网络，利用现场监测数据对这些参数进行校准。所提出的框架被嵌入到一个实时预测系统中，并应用于上海地铁15号线古北路站，该站毗邻10号线。预测的隧道位移与现场测量结果高度吻合，竖向最大相对误差为21.82%，水平向为1.95%。除位移趋势反转阶段外，该系统始终保持较高的预测精度。这些结果验证了所提方法在预测开挖引起的隧道变形方面的可靠性，并突出了其在城市隧道工程主动风险管理中的适用性。
