The Application of Long Short-Term Memory(LSTM) Method on Displacement Prediction of Multifactor-Induced Landslides

Abstract

Landslides induced by rainfall frequently happen in Southwestern China where steep slopes, loess plateau occur. Thus, it is empirical to build the early warning system to evaluate the potential of landslide hazards. However, current researches mostly focus the static model on displacement prediction. The landslide is a nonlinear hazard characterized by dynamic features. Therefore, the dynamic model should be investigated to more precisely predict the displacement associated with the landslide. In this paper, Laowuji Landslide is adopted to investigate the dynamic failure mode. The displacement of the Laowuji landslide contains the trend and periodic component. The trend component is predicted by the empirical mode decomposition and the periodic component is predicted by the long short-term memory (LSTM) method. Model's input includes multiple factors of geological conditions, rainfall intensity, and human activities. The measured data and the predicted data show good consistency. In addition, the predicted results of the periodic component show that the performance of the LSTM has good characteristics of dynamic feature. Compared with a traditional mechanical model, the hybrid model is more powerful to predict the landslide displacement triggered by multiplying factors. INDEX TERMS Landslide, rainfall, long-term stability, SouthWestern China. The associate editor coordinating the review of this manuscript and approving it for publication was Xin Luo.

Machine Learning: New Potential for Local and Regional Deep-Seated Landslide Nowcasting

“…Xie et al [104] Laowuji, China Rainfall, toe excavation Total Station LSTM Bossi and Marcato [105] Passo della Morte, Italy Rainfall, groundwater Inclinometer Linear regression Yang et al [106] Baishuihe & Bazimen, China Rainfall, reservoir level GNSS LSTM Miao et al [107] Baishuihe, China Rainfall, reservoir level GNSS, inclinometer GA-SVR, GS-SVR, PSO-SVR Li et al [37] Baishuihe, China Rainfall, reservoir level GNSS LASSO-ELM, Copula (ELM, SVM, RF, kNN) Logar et al [108] Ventor, United Kingdom Rainfall Crackmeter ANN Krkač et al [33] Kostanjek, Croatia Groundwater (change), season GNSS RF Zhou et al [109] Bazimen, China Rainfall, reservoir level GNSS PSO-SVM (GA-SVM, GS-SVM, BPNN) Cao et al [110] Baijiabao, China Rainfall, groundwater, reservoir level GNSS ELM (SVM) Lian et al [111] Baishuihe & Bazimen, China Rainfall, reservoir level GNSS LSSVM, ELM, combination Chen and Zeng [112] Baishuihe, China None GNSS BPNN Du et al [31] Baishuihe & Bazimen, China Rainfall, reservoir level GNSS, inclinometer BPNN Lian et al [113] Buishuihe, China None GNSS EEMD-ELM, M-EEMD-ELM (ANN, BPNN, RBFNN, SVR, ELM) Corominas et al [114] Vallcebre, Spain Groundwater Extensometers Physics Neaupane and Achet [115] Okharpauwa, Nepal…”

Machine Learning for Landslides Prevention: A Survey

“…Xie et al [164] adopted an LSTM model to predict dynamic landslide displacement by evaluating the dynamic characteristics with the time domain. The prediction result indicated that the rainfall intensity and the excavation-induced stress redistribution affected the periodic displacement.…”

70 years of machine learning in geoscience in review

“…A classic application of LSTMs is text analysis and natural language understanding, which has been applied to geological relation extraction from unstructured text documents ( Blondelle, Juneja, Micaelli, & Neri, 2017 ; Luo, Zhou, Wang, Zhu, & Deng, 2017 ). Due to the nature of LSTMs being suited for time series data, it is has been applied to seismological event classification of volcanic activity Titos, Bueno, García, Benítez, and Ibañez (2018) , multifactor landslide displacement prediction ( Xie, Zhou, & Chai, 2019 ), and hydrological modeling ( Kratzert et al, 2019 ). Talarico, Leäo, and Grana (2019) applied LSTM to model sedimentological sequences and compared the model to baseline hidden Markov model (HMM), concluding that RNNs outperform HMMs based on first-order Markov chains, while higher order Markov chains were too complex to calibrate satisfactorily.…”