Time Series of Landslide Displacement Prediction based on VMD-LSTM

This paper proposes a new model Variational Mode Decomposition-Long-Short Term Memory(VMD-LSTM) for landslide displacement prediction. Firstly, Variational Mode Decomposition(VMD) decomposes the landslide displacement time series into multiple intrinsic mode functions(IMFs) to reduce the interference of noise. At the same time, three normalization methods process the data to explore the applicability of normalization method to model. Then, each decomposed IMF is predicted by LSTM model. Finally, the prediction result of each IMF is added to obtain the final predicted value. In this paper, the landslide displacement sequence of Baishuihe in the Three Gorges Reservoir area is taken as one case. This experiment proved the feasibility of VMD-LSTM model. Comparing with the traditional LSTM moel, the root-mean-squareerror(RMSE) of VMD-LSTM is reduced to 9.75.