This invention discloses a method for predicting seepage pressure in concrete dams based on multi-scale preprocessing and dual-model fusion. First, the
monitoring data is cleaned using the 3σ method to construct multi-
hysteresis water pressure, rainfall, temperature, and time-dependent driving factors. Then, the seepage pressure
time series is decomposed using EEMD, and effective IMFs are screened using a four-threshold method and divided into high, medium, and low-frequency components.
Ridge regression and
Transformer-BiLSTM models are trained in parallel for each component, with R²... 2 The normalized comprehensive selection index COI, based on RMSE, MAE, and MAPE, automatically selects the optimal component for prediction. The optimal sequence of each component is input into a fully connected network, and the global seepage pressure sequence is reconstructed through
adaptive learning of multi-scale
nonlinear coupling relationships. The model is validated using four indices, and seepage anomalies of various types, such as rainfall,
water level, and
curtain wall degradation, are traced hierarchically based on deviations in each
frequency band. This invention overcomes the shortcomings of single models and
linear reconstruction, balancing prediction accuracy, generalization ability, and
engineering interpretability, and is suitable for early warning of seepage safety in concrete dams.