A model for predicting landslide displacement based on time series and long and short term memory neural network

To address the transient characteristics of landslide processes and to overcome the deficiency of static forecasting models, a model for predicting the transient landslide displacement was proposed based on time series theory and long and short term memory neural network(LSTM). In the model, the moving average method was applied to decompose the cumulative displacement into the trend term and periodic term. Subsequently, the trend displacement was predicted by a polynomial model. A LSTM model, based on the response of inducing factors, was established to predict the periodic displacement. Finally, the trend displacement and periodic displacement were superposed to achieve the cumulative displacement. Baishuihe landslide, a typical stepped landslide in Three Gorges Reservoir area, was taken as an example to test the prediction performance of the proposed model and the support vector machine(SVM) was used for comparison. The results demonstrate that the transient model(LSTM) achieves higher prediction accuracy than the static model(SVM), especially during the period of stepped deformation. Furthermore, the prediction accuracy of the LSTM model is not limited by the timeliness analysis of training sets. © 2018, Science Press. All right reserved.