In regions prone to tropical typhoons and heavy rainfall, reservoir bank slope deformation displays pronounced nonlinear and time-varying characteristics, posing challenges for traditional methods in capturing complex evolutionary patterns. To overcome this, this study introduces a deep learning model that integrates LSTM and Transformer architectures. The LSTM network hierarchically extracts temporal rainfall features, while the Transformer's multi-head self-attention mechanism captures global correlations in deformation fields. Hyperparameters are adaptively optimized using a Bayesian algorithm to build a dynamic prediction framework. Applied to a slope in a Hainan water conservancy project, experimental results demonstrate that the LSTM-Transformer model significantly outperforms standalone LSTM and Transformer models in prediction accuracy during deformation phases. This research delivers a high-precision, interpretable tool for intelligent reservoir bank slope monitoring, enabling proactive decision-making for geological disaster risk early warning.

暂无