Landslide displacement prediction based on time series and GRU-ATTENTION neural network

Landslide displacement prediction is an important component of landslide monitoring and warning. Neural networks are gradually being applied to landslide displacement prediction. In order to enhance the extraction and capture of historical information to make predictions more accurate, this paper proposes a Gated Recurrent Unit neural network with attention mechanism (GRU-ATTENTION) for predicting landslide displacement. Firstly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm is used to decompose the landslide displacement into trend and periodic terms. Then, the GRU-ATTENTION is used to fit the trend and period terms of landslide displacement with noise. Finally, considering the influence of rainfall and reservoir water level factors, the periodic and trend terms of the landslide displacement are predicted in Bazimen landslide. Compared with the prediction results of other traditional neural networks, the results indicate the GRU-ATTENTION can better capture historical information features, and achieve better prediction results, which provide a new technical method for landslide prediction.