Landslide Displacement Prediction Based On EEMD-CNN-LSTM Model

The prediction of landslide displacement is an important part of landslide monitoring. Although time series methods based on deep learning paradigms have achieved some success in predicting landslide displacement, the non-stationary, periodic, and trending characteristics of landslide displacement data make it prone to overfitting in current time series models. In this paper, we propose a landslide displacement prediction model based on the combination of Isolation Forest (IF) anomaly detection, Ensemble Empirical Mode Decomposition (EEMD), Convolutional Neural Networks(CNN), and Long Short-Term Memory (LSTM) neural network, which addresses the volatility of landslide displacement data and the characteristics of displacement composed of periodic and trending components. The Baijiabao landslide in the Three Gorges Reservoir area, which is affected by rainfall, is selected as the research object. The IF algorithm is used to detect anomalies in the original landslide displacement data. The trend and periodic components of the landslide displacement are extracted based on the EEMD method, and the overall displacement is predicted using the LSTM model. The accuracy of the prediction is evaluated using four indicators: RMSE, MAE, MAPE, and R². The results show that the proposed EEMD-CNN-LSTM model is superior to traditional LSTM models, random forest methods and EEMD-LSTM methods under both external influences of rainfall and without rainfall, which can greatly reduce overfitting and improve the accuracy of prediction.