Study on unsaturated rainfall-induced slope stability and failure probability based on GA-LSTM-MC

As rainfall-induced landslide hazards increase, accurately predicting slope stability becomes essential. This study combines FLAC3D numerical simulations with unsaturated seepage theory to develop a GA-LSTM-MC model for predicting slope displacement, safety factors, and failure probability. K-fold cross-validation was used to enhance model robustness, especially in cases with small sample sizes. The results show that low-intensity, short-duration rainfall (e.g., ≤50 mm) maintains relatively stable safety factors, with small displacements and low failure probabilities. In contrast, heavy or prolonged rainfall (≥100 mm) significantly reduces safety factors, causing a large increase in displacement and a rapid rise in failure probability. Particularly, when rainfall exceeds 150 mm, safety factors drop to critical levels, displacements exceed 50 cm, and failure probability approaches 100 %. The GA-LSTM-MC model performs well under moderate to high-intensity rainfall conditions, accurately predicting dynamic changes in slope displacement and safety factors. Combined with a graphical user interface (GUI), the system allows real-time input and analysis of rainfall parameters, providing an efficient tool for slope risk assessment and early warning. However, under extreme rainfall conditions, displacement predictions show some bias, especially near failure. Future improvements could include optimizing the GUI interface, incorporating field data validation, and considering multi-factor interactions to further enhance the system's practicality and accuracy.