Landslide displacement prediction based on Variational mode decomposition and MIC-GWO-LSTM model

Abstract

Landslide displacement prediction is essential to establish the early warning system (EWS). According to the dynamic characteristics of landslide evolution and the shortcomings of the traditional static prediction model, a dynamic prediction model of landslide displacement based on long short-term memory (LSTM) neural networks was proposed. Meanwhile, the Variational mode decomposition (VMD) theory was used to decompose the cumulative displacement and triggering factors, which not only give clear physical meaning to each displacement subsequence, but also closely connect the rock and soil conditions with the influence of external factors. Besides, the maximum information coefficient (MIC) was used to sort the redundant features. The LSTM is a dynamic model that can remember historical information and apply it to the current output. The hyperparameters of the LSTM model was optimized by the Grey wolf optimizer (GWO), and the dynamic one-step prediction was carried out for each displacement. All the predicted values were superimposed to complete the displacement prediction based on the time series model. The Tangjiao landslide in the Three Gorges Reservoir area (TGRA), China, was taken as a case study. The displacement data of monitoring sites GPS03 and GPS06 had step-like characteristics. Measured data from March 2007 to December 2016 were selected for analysis. The results indicate that the displacement prediction model based on MIC-GWO-LSTM model effectively improves the prediction accuracy and generalization ability, and is better than other prediction models. This model provides a new idea and exploration for the displacement prediction of step-like characteristics landslide in the Three Gorges Reservoir area.