Landslide Displacement Prediction Based on Time Series Analysis and Double-BiLSTM Model

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence
• Time Series Analysis

Background:

• Machine learning has improved landslide displacement prediction.
• Existing models often overlook temporal data, affecting prediction accuracy.
• Accurate landslide prediction is crucial for risk mitigation.

Purpose of the Study:

• To propose a dynamic landslide displacement prediction model.
• To improve prediction accuracy by considering temporal data dependencies.
• To analyze the influence of factors like rainfall and reservoir levels on landslide behavior.

Main Methods:

• Time series analysis using Exponentially Weighted Moving Average (EWMA) to decompose displacement into trend and periodic components.
• Application of a Bidirectional Long Short-Term Memory (BiLSTM) model for trend displacement prediction.
• Utilizing the Maximum Information Coefficient (MIC) to assess correlations between influencing factors (rainfall, reservoir level) and periodic displacement, followed by BiLSTM prediction.

Main Results:

• The proposed Double-BiLSTM model demonstrated superior prediction performance compared to classical methods.
• The model effectively captured both trend and periodic components of landslide displacement.
• Validation on the Baishuihe landslide data confirmed the model's effectiveness.

Conclusions:

• The dynamic landslide displacement prediction model offers enhanced accuracy.
• Integrating time series analysis with Double-BiLSTM provides a robust framework for landslide prediction.
• This approach effectively predicts landslide displacement, aiding in hazard assessment and management.