Experimental Analysis, Numerical Simulation and Prediction Model of Tunnel Lining Internal Forces under the Coupled Effect of Water-Rich Karst and Goaf

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Tunnels traversing strata with water-rich karst and coal seam goafs face prominent risks of multi-field coupling disasters. However, systematic experimental data and high-precision prediction methods for lining internal force responses under such coupled conditions remain scarce. This study, based on the Changliangzi Tunnel in China, conducted seven sets of 1:50 scale physical model tests covering gravity-only, water-rich karst, goaf and multi-field coupling conditions. An experimentally validated FLAC3D model was established. To overcome the limitation of small samples, the Auxiliary Classifier Generative Adversarial Network (ACGAN) was introduced for data augmentation. The prediction model was developed integrating physical model tests, numerical simulation, generative adversarial network-based data augmentation and explainable machine learning. The research results indicate that in the experimental simulations, water-rich karst and goafs lead to internal force redistribution and asymmetric deformation; through multi-parameter combinations, 50 numerical simulation cases were expanded, and data augmentation generated 800 high-quality synthetic samples, constructing a fused dataset containing 857 samples; The GA-LightGBM prediction model achieved an average coefficient of determination (R²) of 0.9057 on the test set, outperforming comparative models such as XGBoost, CatBoost and deep neural networks in both accuracy and efficiency; The SHAP framework was employed to quantitatively reveal the contribution mechanism of disaster-causing factors with tunnel depth as the globally dominant factor, followed by goaf thickness, karst location, distance to goaf, karst water pressure and secondary lining elastic modulus, which exhibits a "global dominance with local targeting" characteristic, wherein tunnel depth establishes the baseline structural stress level, while goaf and karst parameters significantly intensify local internal force concentration and damage risk at specific locations through local effects. The developed prediction model can provide a basis for intelligent design and risk assessment of tunnels under complex geological conditions.

This publication has 10 references indexed in Scilit: