Landslide Displacement Prediction Based on Multi-source Time Series

Displacement prediction has an important position in the landslide early warning and forecasting system. In order to improve the accuracy of displacement prediction, a landslide displacement dynamic prediction method combining variational modal decomposition with double exponential smoothing method and neural network is proposed. Firstly, the variational modal decomposition is performed on the historical monitoring displacement data to generate multiple modal components, and then the double exponential smoothing method and the extreme learning machine (ELM) model are used for prediction. Genetic algorithm is used to optimize the ELM model and finally accumulate the predicted values of each mode to complete the prediction. Taking the Baijiabao landslide in the Three Gorges reservoir area as an example, the proposed model is compared with the periodic terms predicted by the least squares support vector machine (LSSVM) and convolutional neural network-gated recurrent unit (CNN-GRU). The results show that the DES-GA-ELM used can effectively predict landslide displacement changes with the smallest prediction errors of 1.039 mm, 0.808 mm, 0.001, and 0.9890 for RMSE, MAE, MAPE, and R
²
values, respectively. The research results can provide a technical basis for the landslide early warning monitoring system.