Prediction of Coal Burst Location and Risk Level in Roadway Using XGBoost with Multi-element Microseismic Information and Its Application in Steeply Inclined Ultra-Thick Coal Seam

The intelligent warning of coal burst is crucial for ensuring the safety of mine production. The present study proposes an intelligent method for classifying warnings to effectively identify danger zones prone to coal bursts based on microseismic (MS) data, with a specific focus on the excavation roadway in Wudong Coal Mine, Xinjiang Province. The area is divided into grids using a spatial scanning approach, and a dataset is constructed comprising multiple MS information indexes that are associated with regional coal burst danger levels. The temporal and spatial precursory characteristics and evolution law of a typical coal burst in a steeply inclined coal seam roadway are investigated. The data features of the samples are subjected to correlation analysis. The prediction model of regional coal burst risk level is developed by integrating the XGBoost machine learning algorithm, enabling early warning for varying intensities of coal burst in different regions (intense, moderate, slight, and none). The modeling using XGBoost outperforms GBDT, LightGBM, and Adaboosting in terms of prediction performance, exhibiting a superior comprehensive classification prediction accuracy of 0.9459 and F1-Score of 0.9443. The model is finally applied to conduct a regional coal burst risk assessment for two major energy events that occur during the excavation of the steeply inclined ultra-thick coal seam roadway. The prediction results align with the monitoring findings, demonstrating the feasibility and accuracy of the prediction method. This research approach can serve as a reference for intelligent early warning systems against coal bursts in steeply inclined ultra-thick coal seam roadways.