An improved GRU method for slope stress prediction.

Area of Science:

• Geotechnical Engineering
• Data Science
• Artificial Intelligence

Background:

• Open-pit mine slope stability is a complex nonlinear system influenced by stress variations.
• Accurate landslide prediction and early warning are crucial for risk assessment.
• Traditional methods struggle with nonlinear time series data, exhibiting low accuracy and robustness.

Purpose of the Study:

• To develop an intelligent prediction model for open-pit mine slope stress data.
• To improve the accuracy and robustness of landslide early warning systems.
• To address the limitations of traditional prediction methods in handling nonlinear time series data.

Main Methods:

• Variational Mode Decomposition (VMD) for preprocessing and feature extraction of stress data.
• Dung Beetle Optimization (DBO) to optimize Gated Recurrent Unit (GRU) parameters (hidden layers, learning rate).
• Integration of a self-attention mechanism within the GRU to enhance feature relevance capture.

Main Results:

• The VMD-DBO-GRU-A model demonstrated a significant reduction in root-mean-square error (77% vs. LSTM, 84% vs. SVM).
• Achieved a high coefficient of determination (0.9978), indicating excellent predictive performance.
• The model effectively extracts localized features and captures internal data relevance.

Conclusions:

• The VMD-DBO-GRU-A model offers superior performance and high prediction accuracy for open-pit mine slope stability.
• This intelligent model provides valuable advancements for practical landslide early warning applications.
• The optimized GRU with VMD preprocessing and self-attention is effective for complex nonlinear time series analysis.