Prediction of Concealed Water Body Ahead of Construction Tunnels Based on Temperature Patterns and Artificial Neural Networks

Concealed water bodies within surrounding rock formations pose a serious threat to tunnel construction. To address this risk, this study integrates physics-based heat conduction theory with deep learning, unlike existing methods that treat temperature as isolated data points or rely solely on empirical models. The approach introduces three key innovations: (a) analytical temperature–location relationships for water body characterization; (b) pseudo-temporal modeling of spatial sequences and (c) physics-guided neural architecture design. First, a steady-state heat conduction model is established to characterize axial temperature distribution patterns caused by concealed water bodies during excavation. From this, quantitative relationships between temperature anomalies and the location and size of the water bodies are derived. Next, a deep learning model, ST-HydraNet, is proposed to treat tunnel axial temperature data as a pseudo-time series for hazard prediction. Experimental results demonstrate that the model achieves high accuracy (91%) and perfect precision (1.0), significantly outperforming existing methods. These findings show that the proposed framework provides a non-invasive, interpretable, and robust solution for real-time hazard detection, with strong potential for integration into intelligent tunnel safety systems. By enabling earlier and more reliable detection, the model directly enhances construction safety, economic efficiency, and environmental sustainability.