Integrating Signal Decomposition and Intelligent Algorithms for Landslide Displacement Prediction through Multiple Decomposition modal and Clustering

Landslide displacement prediction is an important part of geologic disaster prevention and control. Landslide displacement in the Three Gorges area is affected by short-term factors such as rainfall and reservoir level fluctuation, showing obvious step-abrupt characteristics. Aiming at the problem that the single decomposition of GNSS high-frequency displacement signal easily leads to the mixing of the periodic term and noise, this paper innovatively proposes the CVSCB (CEEMDAN-VMD-SCSSA-CNN-BILSTM) hybrid prediction model. Firstly, the cumulative displacement is adaptively decomposed by CEEMDAN, and the periodic component is recognized based on Sample Entropy (SampEn) and K-means clustering; following that, the VMD is used to decompose the periodic component in a secondary way, which effectively separates the noise interference. In the prediction stage, the trend term is modeled by ARIMA, while the periodic and stochastic terms are combined with the influence factors screened by Spearman's correlation, and input into the CNN-BILSTM network optimized by SCSSA for prediction. Among them, the improved sparrow algorithm (SCSSA) significantly improves the efficiency of hyperparameter optimization. Finally, the predicted period term, trend term and random term are superimposed to obtain the cumulative predicted displacement. The validation is based on the daily data from the monitoring point ZD3 in Baijiabao, and the results show that the RMSE of the periodic term is reduced from 2.5401 mm to 0.3899 mm after multiple decomposition, and the prediction effect is significantly improved. In order to verify the generalization ability of the model, the daily monitoring data were extended to the data of Baijiabao ZD2 monitoring point for validation, and the monthly monitoring data were adopted from Xintan DTX1 monitoring point for the validation of the same model. The experimental results show that the method has good generalization, and it can provide a reliable reference for the prediction of landslide displacements in the trend of the step change.