Abstract: Objectives: The distinct step-like deformation characteristics and complex creep-mutation state transitions exhibited by landslides along the Three Gorges Reservoir banks, resulting from the dynamic coupling effects of rainfall and reservoir water level fluctuations, pose significant challenges for existing displacement prediction methods. Specifically, these challenges include difficulties in effectively characterizing the lag decay effects of rainfall-reservoir level interactions and insufficient cross-scale model adaptability, which severely constrain prediction accuracy. Methods: To address these issues, this study proposes a novel dynamic displacement prediction method for reservoir bank landslides. This method integrates a new lag-decayeffect-based dominant factor screening strategy with a combined model utilizing grey wolf optimizer (GWO), bidirectional temporal convolutional neural network (BITCN), bidirectional gated recurrent unit (BIGRU), and attention mechanism. The model aims to simultaneously capture time-series features and high-dimensional dynamic spatial features. Using the Xinpu and Baishuihe landslides in the Three Gorges Reservoir area as case studies at different temporal scales, we first decomposed cumulative landslide displacement into trend and periodic components using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Trend displacement was predicted using the double exponential smoothing method. Subsequently, we conducted an in-depth analysis of the dynamic response of landslide deformation to rainfall and reservoir level variations, explicitly considering their lag decay effects on displacement evolution. Building on this, we employed a comprehensive approach combining recursive feature elimination with cross-validation and extreme gradient boosting (RFECVXGBoost), classification and regression trees (CART), and the maximal information coefficient (MIC) to screen for the dominant factors controlling displacement evolution, identifying the antecedent precipitation index (API) as the core factor. Finally, the GWO-BITCN-BIGRU-Attention model was applied to predict the periodic displacement component. Results: The analysis of engineering examples shows that the method proposed in this paper is significantly superior to the existing methods in terms of correlation coefficient, root mean square error and other indicators. Conclusions: Model evaluation and ablation experiments demonstrate that the proposed method effectively characterizes the time-varying dynamic influence patterns of the lag decay effects associated with rainfall and reservoir water level fluctuations. Furthermore, these experiments confirm the method's capability to capture and learn the long-term dynamic evolutionary characteristics of reservoir bank landslide displacement across diverse temporal scales, leading to a significant improvement in displacement prediction accuracy.