LSTM-Transformer-Based Mine Pressure Prediction Using Hydraulic-Support Monitoring Data

Area of Science:

• Mining Engineering
• Geotechnical Engineering
• Artificial Intelligence

Background:

• Hydraulic support pressure data in longwall mining is vital for understanding roof-support interaction and enabling intelligent monitoring.
• Pressure sequences are often corrupted by noise, missing data, and complex patterns like periodic weighting, complicating accurate forecasting.
• Existing methods struggle to capture both local nonlinear dynamics and global temporal dependencies in support pressure data.

Purpose of the Study:

• To develop and validate a hybrid LSTM-Transformer model for accurate hydraulic support pressure forecasting in fully mechanized longwall mining.
• To address the challenges posed by noisy, non-stationary, and periodic features in underground support pressure data.
• To improve the reliability of mine pressure prediction for enhanced safety and operational efficiency.

Main Methods:

• An LSTM-Transformer hybrid model was designed, integrating LSTM for local feature extraction and Transformer for global sequence modeling.
• Support-wise experiments were conducted using independent processing of field monitoring data from Yili No. 1 Mine.
• The model was evaluated based on its ability to forecast future pressure sequences and capture periodic patterns.

Main Results:

• The proposed LSTM-Transformer model achieved a high coefficient of determination (R² = 0.971) and significantly reduced Mean Absolute Error (MAE = 0.471 MPa).
• The model demonstrated improved phase consistency in predicting pressure peaks, indicating better capture of dynamic events.
• Stable accuracy was maintained for short-term predictions (25-step forecasting), with sufficient historical data crucial for capturing complete pressure cycles.

Conclusions:

• The LSTM-Transformer hybrid model presents a feasible and effective approach for hydraulic support pressure prediction in challenging underground mining environments.
• The findings highlight the importance of hybrid deep learning architectures in handling complex time-series data from mining operations.
• Accurate mine pressure forecasting using this method can support intelligent monitoring systems and improve mine safety.