A novel displacement prediction method using gated recurrent unit model with time series analysis in the Erdaohe landslide

• Published in: Natural hazards (Dordrecht) 2021, Vol.105 (1), p.783-813
• Main Authors: Zhang, Yong-gang, Tang, Jun, He, Zheng-ying, Tan, Junkun, Li, Chao
• Format: Article
• Language: English
• Subjects: Canyons; Civil Engineering; Displacement; Earth and Environmental Science; Earth Sciences; Environmental Management; Geological hazards; Geophysics/Geodesy; Geotechnical Engineering & Applied Earth Sciences; Global positioning systems; GPS; Hydrogeology; Landslides; Landslides & mudslides; Learning algorithms; Machine learning; Model accuracy; Natural Hazards; Natural phenomena; Original Paper; Polynomials; Positioning systems; Predictions; Rain; Rainfall; Reservoirs; Socioeconomic factors; Support vector machines; Time series; Training; Warning systems
• ISSN: 0921-030X; 1573-0840
• DOI: 10.1007/s11069-020-04337-6

Landslides are natural phenomena, causing serious fatalities and negative impacts on socioeconomic. The Three Gorges Reservoir (TGR) area of China is characterized by more prone to landslides for the rainfall and variation of reservoir level. Prediction of landslide displacement is favorable for the establishment of early geohazard warning system. Conventional machine learning methods as forecasting models often suffer gradient disappearance and explosion, or training is slow. Hence, a dynamic method for displacement prediction of the step-wise landslide is provided, which is based on gated recurrent unit (GRU) model with time series analysis. The establishment process of this method is interpreted and applied to Erdaohe landslide induced by multi-factors in TGR area: the accumulative displacements of landslide are obtained by the global positioning system; the measured accumulative displacements is decomposed into the trend and periodic displacements by moving average method; the predictive trend displacement is fitted by a cubic polynomial; and the periodic displacement is obtained by the GRU model training. And the support vector machine (SVM) model and GRU model are used as comparisons. It is verified that the proposed method can quite accurately predict the displacement of the landslide, which benefits for effective early geological hazards warning system. Moreover, the proposed method has higher prediction accuracy than the SVM model.