Gas Outburst Warning Method in Driving Faces: Enhanced Methodology through Optuna Optimization, Adaptive...

Area of Science:

• Mine safety engineering
• Data science
• Artificial intelligence

Background:

• Gas pre-warning systems face challenges like limited indicators, poor adaptability, and imprecise modeling.
• Broad application across diverse mines and insufficient data impact warning accuracy.

Purpose of the Study:

• To develop a hybrid predictive and pre-warning model for gas detection in mines.
• To address limitations in current gas pre-warning systems, enhancing accuracy and adaptability.

Main Methods:

• Adaptive Normalization (AN) for time-series data standardization, coupled with Gated Recurrent Unit (GRU).
• Ensemble Empirical Mode Decomposition (EEMD) for feature extraction and Variational Mode Decomposition (VMD) order selection.
• Enhanced transformer framework for non-linearities and long time-series analysis, optimized using Optuna and xgbRegressor.

Main Results:

• The hybrid model achieved an R-squared of 0.980975 and MAE of 0.000149, outperforming RNN, GRU, LSTM, and BiLSTM.
• Bootstrapping was used to estimate confidence intervals for the hybrid model, addressing data scarcity.
• Dimensional analysis created real-time gas emission metrics; anomaly detection enabled unsupervised early alerts for gas bursts.

Conclusions:

• The proposed hybrid model demonstrates strong predictive and pre-warning capabilities for gas hazards in mines.
• This approach offers technological reinforcement for advancing intelligent coal mine operations.
• The model effectively handles data scarcity and improves versatility across different mining scenarios.