Saturation Line Forecasting in Tailings Dams with a Multivariate TCN-Transformer Model

As high-risk mining facilities, tailings dams are directly linked to personnel safety and ecological integrity. Accurate prediction of the internal saturation line is essential for assessing dam stability and preventing potential failures. This study proposes a hybrid deep learning model that integrates a Temporal Convolutional Network (TCN) with a Transformer encoder to effectively capture both local temporal features and long-range dependencies from complex, noisy, and nonlinear multisource monitoring data. The model is trained and validated using real-world sensor data collected from a tailings dam in Luoyang, incorporating variables such as rainfall, reservoir level, and displacement. Experimental results demonstrate that the proposed TCN-Transformer model achieves high predictive accuracy and generalization capability across multiple evaluation metrics, including MAE, RMSE, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, and MAPE. These results highlight the model's effectiveness and robustness, indicating its strong potential for deployment in intelligent early warning systems and safety monitoring frameworks for tailings dam management.