Landslide Displacement Prediction Based on Multivariate LSTM Model.

Area of Science:

• Geotechnical Engineering
• Earthquake Engineering
• Data Science

Background:

• Landslides in China cause significant loss of life and economic damage.
• Accurate landslide displacement prediction is crucial for mitigating these impacts.
• Traditional time series models like ARIMA and univariate LSTM have limitations in complex geological scenarios.

Purpose of the Study:

• To develop a more accurate landslide displacement prediction model.
• To integrate multiple influential factors like rainfall and reservoir water level into the prediction model.
• To evaluate the performance of the proposed model against traditional methods.

Main Methods:

• A multivariable Long Short-Term Memory (LSTM) model was developed.
• The model integrates time series data of landslide displacement, rainfall, and reservoir water level.
• Data from the Baijiabao landslide in the Three Gorges Reservoir area (2006-2012) was used for validation.

Main Results:

• The multivariable LSTM model demonstrated superior accuracy in predicting landslide displacement.
• Error metrics (MSE, RMSE, MAE) were significantly lower compared to ARIMA and univariate LSTM models.
• Specific error values achieved were MSE: 0.64223, RMSE: 0.8014, and MAE: 0.50453 mm.

Conclusions:

• The multivariable LSTM model offers higher accuracy and better practical application prospects for landslide displacement prediction.
• This approach provides a valuable reference for predicting similar landslides, especially in reservoir areas.
• Integrating environmental factors enhances the reliability of landslide prediction models.