Combined with Decomposition Algorithm and Generative Adversarial Networks on Landslide Displacement Prediction

Abstract

Landslide displacement prediction is essential to establishing the early warning system (EWS). To better grasp the landslide evolution process, this paper proposes a novel architecture of variational mode decomposition-Generative Adversarial Network (VMD-GAN) for forecasting the landslide displacement. Firstly, VMD was used to decompose the time series into multiple intrinsic mode functions (IMFs) to extract the internal hidden information of the original series and remove the interference of noise to improve the prediction accuracy of the model. Then, GAN predicts each IMFs. Finally, the predicted results for each IMFs component are added to get the final prediction result. The Baishuihe in the Three Gorges Reservoir was made as an example and displacement data from August 2003 to December 2011 were selected for analysis. Compared with empirical mode decomposition-Generative Adversarial Network(EMDGAN), long short-term memory (LSTM), and temporal convolutional networks (TCN) models, the result has shown that the root means square errors (RMSE) of VMD-GAN in landslide prediction was 3.33 and the correlation coefficient R-square was 0.99, which demonstrated the best prediction accuracy and fitting ability.