Novel ensemble intelligence methodologies for rockburst assessment in complex and variable environments.

Area of Science:

• Geological Engineering
• Machine Learning
• Data Science

Background:

• Rockbursts pose significant risks to deep mining and tunnel construction.
• Existing prediction algorithms have limitations.
• Advanced machine learning offers potential for improved rockburst prediction.

Purpose of the Study:

• To investigate the efficacy of various ensemble tree algorithms for rockburst prediction.
• To optimize ensemble tree models using Bayesian optimization.
• To enhance prediction accuracy through model combination strategies.

Main Methods:

• Evaluated seven ensemble tree algorithms: Random Forest (RF), Extremely Randomized Tree (ET), Adaptive Boosting Machine (AdaBoost), Gradient Boosting Machine, Extreme Gradient Boosting Machine (XGBoost), Light Gradient Boosting Machine, and Category Gradient Boosting Machine.
• Applied Bayesian optimization for hyperparameter tuning.
• Implemented voting, bagging, and stacking strategies to combine multiple models.
• Validated models on 314 real rockburst cases.

Main Results:

• Extremely Randomized Tree (ET) and Extreme Gradient Boosting Machine (XGBoost) showed the highest single-model testing accuracy (85.71%).
• Ensemble combination models outperformed single models.
• The 'voting 2' model, combining XGBoost, ET, and RF, achieved the highest testing accuracy (88.89%).
• Sensitivity analysis demonstrated superior robustness and adaptability of the 'voting 2' model.

Conclusions:

• Ensemble tree algorithms, particularly combined models, significantly improve rockburst prediction accuracy.
• The 'voting 2' model offers a robust and adaptable solution for complex engineering environments.
• The findings are validated by field data from Sanshandao Gold Mine, confirming practical applicability.