Research on Deformation Prediction of Small Interval Tunnel Based on Machine Learning and Numerical Simulation.
Saved in:
| Title: | Research on Deformation Prediction of Small Interval Tunnel Based on Machine Learning and Numerical Simulation. |
|---|---|
| Authors: | Zhai, Wenjie1 (AUTHOR), Xue, Jia2 (AUTHOR), Fu, Yongjie1 (AUTHOR), Yu, Beichen3,4 (AUTHOR) yubc@cumt.edu.cn, Zhu, Yan5 (AUTHOR), Yuan, Man3,4 (AUTHOR), Chi, Mingbo5 (AUTHOR), Li, Zhentao5 (AUTHOR) |
| Source: | Energy Science & Engineering. May2026, Vol. 14 Issue 5, p2380-2392. 13p. |
| Subject Terms: | *Genetic programming, *Prediction models, *Artificial neural networks, *Machine learning, *Computer simulation, *Tunnel design & construction, *Simulation software, *Excavation (Civil engineering) |
| Abstract: | Constrained by complex geological conditions and specialized structural requirements, highway tunnel construction often employs the small interval tunnel configuration. As a core component of the New Austrian Tunneling Method (NATM), monitoring measurement is instrumental in guiding the construction process for such tunnels. However, conventional monitoring approaches are limited in their capacity to detect latent deformation trends or anticipate unforeseen events. This study employs machine learning algorithms—specifically artificial neural networks (ANN) and genetic programming (GP)—to predict and analyze variations in vault settlement during the excavation of small interval tunnels. A predictive model for settlement was developed and validated through comparative analysis with regression methods to evaluate the predictive efficacy of machine learning approaches. Integrated with this predictive framework, numerical simulations of tunnel excavation were conducted using FLAC3D to calculate vault settlements following the sequential excavation of twin tunnels. Results indicate that the excavation of the subsequent tunnel induces adverse effects on the stability of the primary tunnel. The predictive model exhibits exceptional agreement with field monitoring data, accurately capturing the temporal evolution of settlement throughout the excavation process. Compared with ANN, GP achieves superior predictive accuracy and stability, accurately reproducing the three‐stage deformation sequence. Asymptotic GP analysis predicts a final vault settlement of 37.102 mm under sustained excavation influence, with secondary tunnel excavation contributions quantified as 4.619 mm of incremental displacement. Numerical simulations further validate the settlement increase in the primary tunnel following the excavation of the subsequent tunnel, thereby affirming the reliability of FLAC3D in deformation prediction. The proposed settlement prediction model provides a robust theoretical and computational foundation for determining allowable deformation reserves in tunnel engineering applications. [ABSTRACT FROM AUTHOR] |
| Database: | Energy & Power Source |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | Constrained by complex geological conditions and specialized structural requirements, highway tunnel construction often employs the small interval tunnel configuration. As a core component of the New Austrian Tunneling Method (NATM), monitoring measurement is instrumental in guiding the construction process for such tunnels. However, conventional monitoring approaches are limited in their capacity to detect latent deformation trends or anticipate unforeseen events. This study employs machine learning algorithms—specifically artificial neural networks (ANN) and genetic programming (GP)—to predict and analyze variations in vault settlement during the excavation of small interval tunnels. A predictive model for settlement was developed and validated through comparative analysis with regression methods to evaluate the predictive efficacy of machine learning approaches. Integrated with this predictive framework, numerical simulations of tunnel excavation were conducted using FLAC3D to calculate vault settlements following the sequential excavation of twin tunnels. Results indicate that the excavation of the subsequent tunnel induces adverse effects on the stability of the primary tunnel. The predictive model exhibits exceptional agreement with field monitoring data, accurately capturing the temporal evolution of settlement throughout the excavation process. Compared with ANN, GP achieves superior predictive accuracy and stability, accurately reproducing the three‐stage deformation sequence. Asymptotic GP analysis predicts a final vault settlement of 37.102 mm under sustained excavation influence, with secondary tunnel excavation contributions quantified as 4.619 mm of incremental displacement. Numerical simulations further validate the settlement increase in the primary tunnel following the excavation of the subsequent tunnel, thereby affirming the reliability of FLAC3D in deformation prediction. The proposed settlement prediction model provides a robust theoretical and computational foundation for determining allowable deformation reserves in tunnel engineering applications. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 20500505 |
| DOI: | 10.1002/ese3.70402 |
