A Hybrid Framework Integrating Past Decomposable Mixing and Inverted Transformer for GNSS-Based Landslide...

Area of Science:

• Geosciences
• Geotechnical Engineering
• Data Science

Background:

• Landslide displacement prediction is crucial for geohazard early warning and infrastructure safety.
• Global Navigation Satellite System (GNSS) time series data present challenges due to nonstationary, nonlinear, and multiscale behaviors.
• Accurate modeling of these complex behaviors is essential for reliable predictions.

Purpose of the Study:

• To develop a hybrid prediction framework (PDM-iTransformer) for modeling complex landslide displacement.
• To effectively handle nonstationary, nonlinear, and multiscale characteristics in GNSS time series data.
• To improve the accuracy and robustness of landslide displacement predictions.

Main Methods:

• A hybrid framework integrating Past Decomposable Mixing (PDM) and an inverted Transformer (iTransformer) was proposed.
• The PDM module decomposes time series into multi-resolution components, enhancing feature representation.
• The iTransformer models individual time series with cross-variable self-attention for latent dependency capture and feed-forward networks for local feature extraction.

Main Results:

• The PDM-iTransformer framework demonstrated superior performance compared to traditional models.
• Coefficient of determination (R²) increased by 16.2-48.3%.
• Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were reduced by up to 1.33 mm and 1.08 mm, respectively.

Conclusions:

• The proposed PDM-iTransformer framework is effective and robust for predicting landslide displacement in complex terrain.
• The hybrid approach successfully models both long-term trends and short-term fluctuations in GNSS data.
• This advancement contributes significantly to geohazard early warning and infrastructure safety.