Landslide displacement prediction based on the VMD-MIC-TCN-CBAM-LSTM hybrid model

Abstract

Landslides are one of the most common types of geological disasters. To address the complex process of landslide displacement in the landslide and achieve a more accurate and effective prediction of landslide displacement, this study proposes a prediction method based on VMD (Variational Mode Decomposition), MIC (Maximal Information Coefficient), and TCN-CBAM LSTM. VMD was used to decompose the cumulative landslide displacement into trend, periodic, and random terms. The time lag of external triggering factors is considered and the MIC was used to identify the main correlated triggering factors. The CBAM (Convolutional Block Attention Module) is embedded into the TCN-LSTM (Temporal Convolutional Network and Long Short-Term Memory) to focus the model on the more important parts of the features, reducing attention to irrelevant redundant features. The proposed method was evaluated and validated using rainfall, reservoir water level, and cumulative displacement of Baishuihe landslide over a period of 13 years. Based on the obtained results, the proposed method achieved the highest accuracy compared with the other five models, with the average MAE of 4.1637 mm, average RMSE of 5.2051 mm, and goodness-of-fit R2 of 0.9766. This study provides new insights that can be used to develop deep data-mining approaches for landslide displacement prediction.