The Study on Settlement Prediction Model of Earth-Rock Dams Based on EMD-LSTM

In addressing the difficulties encountered by settlement prediction models for earth-rock dams—such as the susceptibility of regression models to multicollinearity, as well as challenges like overfitting, local minima traps, and the complexity of tuning hyperparameters in neural network models—this study introduces an enhanced approach that incorporates Empirical Mode Decomposition (EMD) and Long Short-Term Memory (LSTM) neural networks. To begin with, EMD is applied to decompose the time series data obtained from Global Navigation Satellite System (GNSS) measurement points at multiple scales, allowing for the extraction of both trend and periodic components. After that, Principal Component Analysis (PCA) is conducted to select key factors influencing the model, thus reducing the data’s dimensionality and improving its ability to generalize. Lastly, an LSTM-based time series model is developed, and its hyperparameters are optimized through the Whale Optimization Algorithm (WOA), which enhances both the model’s accuracy in prediction and its convergence speed. The experimental outcomes demonstrate that the proposed model significantly outperforms traditional settlement prediction methods for earth-rock dams, with a Mean Squared Error (MSE) of 7.0701 and a Mean Absolute Error (MAE) of 1.8859, alongside a coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) of 99.83%. Compared to existing techniques, this model exhibits considerable improvements in noise reduction, feature extraction, and hyperparameter tuning, offering a highly accurate and dependable approach for predicting settlement in earth-rock dams.