CNSPC-TabPFN-Based Structural Disease Prediction Model for River-Crossing Shield Tunnels

Shield tunnels, as key cross-river infrastructure, are prone to structural diseases such as leakage, segment misalignment, and lining cracks due to complex geological conditions, load effects, and hydrodynamic changes. These diseases exhibit spatiotemporal correlations and are closely linked to structural deformations (e.g., settlement, convergence) and environmental factors (e.g., riverbed erosion). However, traditional prediction methods struggle with high-dimensional data, strong spatiotemporal dependencies, and small sample limitations. This study proposes a disease prediction approach integrating Complex Network Spectral Clustering (CNSPC) and Tabular Prior-Data Fitted Networks (TabPFN). First, a similarity network of monitoring points is constructed using complex network theory to incorporate structural and environmental data. CNSPC then detects anomalous patterns, followed by TabPFN for disease level prediction, improving accuracy in small-sample scenarios. Using the Shanghai Yangtze River Tunnel as a case study, results show that the proposed method effectively identifies disease patterns and outperforms traditional models like XGBoost and Random Forest. This research provides scientific support for intelligent tunnel maintenance and disease monitoring.