Abstract:

To address issues related to excessive human influence and prolonged prediction times in rockburst prediction, we propose a rockburst intensity classification prediction model based on automatic machine learning. This model is trained using five automatic machine learning frameworks and evaluated using metrics such as accuracy, precision, recall, and F1-score. Subsequently, we compare the performance of this trained model with results from thirteen common machine learning models. The model developed with the Auto-Sklearn framework achieved a high accuracy of 0.969, while the model created with the Auto-Gluon framework, although having the lowest accuracy among the five frameworks, still achieved an accuracy of 0.927. Rockburst prediction models constructed using AutoML frameworks significantly outperformed traditional machine learning algorithms. The Auto-Sklearn-based model exhibited the highest accuracy. In summary, the optimized model was applied to predict rockburst events at the Shaiqi River phosphate mine, and the predictions were consistent with the actual observations on-site. This indicates that the automatic machine learning-based model for rockburst intensity classification prediction can accurately predict rockburst incidents in real-world engineering settings.

Key words: rockburst, intensity grading, forecast, automatic machine learning, algorithm