Research on Displacement Prediction Model of Reservoir Slope Based on LRMA , LSTM, and GA ‐ BP

Accurately predicting slope displacement is critical for slope stability assessment and geological disaster early‐warning. Most existing studies focus on single‐model improvement for slope displacement forecasting, lacking fair benchmark comparison between different models under identical engineering datasets, which restricts clear definition of model application boundaries. Taking nearly 500‐day monitoring displacement data of a reservoir bank slope as input, this paper comparatively investigated one statistical‐decomposition‐based data‐driven LRMA (Local Regression Moving Average) model and two other data‐driven models (LSTM, GA‐BP neural network) under a unified input–output framework and evaluation metrics. The LRMA model decomposes the displacement sequence into trend and periodic components, which are extracted by LOESS and adaptive moving average respectively. The LSTM and GA‐BP take the displacement of consecutive 5 days as input to predict the 6th‐day displacement. The test results indicate that the optimal LRMA, LSTM, and GA‐BP models have R 2 values of 0.5238, 0.8388, and 0.9628 in the test set, respectively. The core innovation lies in quantitatively revealing the respective applicable scenarios of two types of data‐driven models (statistical‐decomposition‐based and neural‐network‐based) for reservoir‐slope displacement prediction via parallel comparison. The LRMA with clear physical interpretation fits mechanism‐analysis scenarios; GA‐BP achieves the highest prediction accuracy and outperforms LSTM in prediction precision and stability for strong‐nonlinear displacement prediction, while LSTM is competent for nonlinear prediction tasks as well. This work provides a quantitative reference for reservoir bank slope stability evaluation and landslide disaster warning.