A Concrete Dam Deformation Prediction Method Based on Mode Decomposition and Self-Attention-Gated Recurrent Unit

Accurate prediction of dam deformation is crucial for structural safety monitoring. For enhancing the prediction accuracy of concrete dam deformation and addressing the issues of insufficient precision and poor stability in existing methods when modeling complex nonlinear time series, a concrete dam deformation prediction method based on mode decomposition and Self-Attention-Gated Recurrent Unit (SAGRU) was proposed. First, Variational Mode Decomposition (VMD) was employed to decompose the raw deformation data into several Intrinsic Mode Functions (IMFs). These IMFs were then classified by K-means algorithm into regular signals strongly correlated with water level, temperature, and aging factors and weakly correlated random signals. For the random signals, an Improved Wavelet Threshold Denoising (IWTD) method was specifically applied for noise suppression. Based on this, a Deep Learning (DL) model based on SAGRU was constructed to train and predict the decomposed regular signals and the denoised random signals, respectively. And finally, the sum of the calculation results of each signal can be output as the predicted deformation. Experimental results demonstrate that the proposed method outperforms existing models in both prediction accuracy and stability. Compared to LSTM, this method reduces the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by approximately 30.9% and 27.2%, respectively. This provides an effective tool for analyzing concrete dam deformation and offers valuable reference directions for future time series prediction research.