Characteristics and identification methods of deep coal and gas outbursts

WANG Enyuan; ZHANG Guorui

doi:10.19956/j.cnki.ncist.2026.01.003

Abstract：With the increase in coal mining depth,influenced by the increase in ground stress and gas,as well as the complex coupling of multiple factors,"low-parameter and low-index"outburst accidents have occurred from time to time in recent years.The traditional"binary classification"method for identifying hazardous and non-hazardous conditions entails certain uncertainties;in particular,outburst risks near critical values are diffi-cult to determine,failing to meet on-site outburst prevention requirements.To address this issue,a"ternary classification"(danger,threat,no risk)intelligent identification method for outburst risk assessment was pro-posed and studied.Characteristics of typical historical deep outburst accidents were analyzed,and a database of 1,378 coal seam outburst risk samples was constructed based on 18 types of identification indices,including coal seam burial depth,gas pressure,and firmness coefficient.Feature information mining(categorical varia-bles,missing values,and index screening)was employed to preprocess the data,aiming to reduce the impact of high dimensionality and small sample sizes on the modeling of identification methods.Various machine learning identification models were established,combined with the optimization algorithm for hyperparameter tuning.The results indicate that the"ternary classification"performance of the ensemble learning BO-LightG-BM model is significantly superior to that of basic machine learning models.Case verification shows that the identification accuracy for"No Outburst Risk-I"is 100％,and the overall accuracy for"Outburst Threat-II"and"Outburst Danger-III"is 88.9％.This achieves reliable and accurate identification of outburst risk data,providing an effective intelligent identification method for outburst risk assessment.