A Landslide Displacement Prediction Method Based on Convolutional Neural Networks_Bidirectional Long Short-Term Memory and Extreme Gradient Boosting

To address the limitations of traditional single deep learning network models in landslide displacement prediction, this paper proposes a combined prediction model based on Convolutional Neural Networks (CNN)_Bidirectional Long Short-Term Memory (BiLSTM)-Extreme Gradient Boosting (XGB). In this paper, the Sparrow Search Algorithm (SSA)- Variational Mode Decomposition (VMD) method is used to decompose the landslide displacement series into periodic, random, and trend components. Appropriate influencing factors are selected as input variables based on Grey Relation Analysis (GRA) Using ensemble learning method, the CNN_BiLSTM-XGB combination model is constructed to predict the periodic and random components. The trend component is fitted by double exponential smoothing method. Finally, the cumulative landslide predicted displacement is obtained by time series theory. The model is evaluated by Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R2), and is validated by the Baishuihe landslide, the performance is compared with CNN-Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Support Vector Machine (SVM) models. The experimental prediction results demonstrate that the CNN_BiLSTM-XGB combination model has higher prediction accuracy. This paper provides new insights and exploration for landslide displacement prediction.