CHN Energy Xinjiang Energy Co., Ltd.
College of Energy Engineering, Xi'an University of Science and Technology

针对急倾斜巨厚煤层掘进巷道冲击地压前兆信息挖掘不充分的困境,以新疆乌东煤矿急倾斜巨厚煤层掘进巷道的冲击危险预警为背景,探究了多项物理指标的演化规律,采用BP神经网络、随机森林(RF)、支持向量机(SVM)、梯度提升决策树(GBDT)等4种机器学习算法,分别建立了急倾斜巨厚煤层掘进巷道冲击地压的预测模型,并对各模型预测性能进行了验证和实例分析,明确了掘进条件下神经网络算法对急倾斜巨厚煤层各指标的敏感性分布。研究结果表明:机器学习算法能够有效提取出各物理指标发展趋势中蕴含的冲击前兆特征,其中BP神经网络算法的建模精度最高,相较于其他模型平均绝对误差分别降低了58.44%,55.64%,26.37%;R2分数分别提高了11.25%,8.37%,3.66%;依据BP神经网络算法得到的各物理指标敏感性排序分别为A(b)值、b值、矿震活动度S值、等效能级参数EEM值、缺震值、算法复杂性AC值;应用R值评分法对乌东煤矿急倾斜巨厚煤层掘进巷道的冲击预警进行了效能验证,结果表明该模型预测准确率较好,总体预测准确率为90.65%,并成功预测了掘进期间的两次大能量事件,该方法适用于急倾斜巨厚煤层掘进巷道的防冲预警,建立的模型对于冲击地压危险的预测有着较好的潜力,研究结果可为类似条件下急倾斜煤层冲击危险的演化发展提供参考。

For the dilemma of insufficient mining of rock burst precursor information in excavation roadway of steeply inclined ultra-thick coal seam, the evolution law of multiple physical indicators was explored based on the early warning of the mining roadway of steeply inclined ultra-thick coal seam in Wudong Coal Mine, Xinjiang Province. Four machine learning algorithms, including BP neural network, random forest (RF), support vector machine (SVM) and gradient lifting decision Tree (GBDT), were adopted. The prediction models of rock burst in excavation roadway of steeply inclined ultra-thick coal seam are established respectively, the prediction perfor-mance of each model is verified and the case analysis is carried out, and the sensitivity distribution of each index under the excavation of steeply inclined ultra-thick coal seam based on neural network algorithm is clarified. The results show that the machine learning algorithm can effectively extract the impact precursor features contained in the development trend of each physical index, among which the modeling accuracy of BP neural network algo-rithm is the highest. When compared with other models, the average absolute error reduced by 58.44%, 55.64% and 26.37%, respectively, and the R2 increased by 11.25%, 8.37% and 3.66%, respectively. Under the geological conditions of the excavation roadway of steeply inclined ultra-thick coal seam, the sensitivity order of each physi-cal index obtained by BP neural network algorithm is A(b) value, b value, S value of mine seismic activity, EEM value of equivalent energy level parameter, earthquake absence value, and algorithm complexity AC value. The R-value scoring method is applied to verify the efficiency of the rockburst warning of the excavation roadway in Wudong Coal Mine, and the results show that the prediction accuracy of the model is good, the overall prediction accuracy is 90.65%, and two large energy events during the excavation are successfully predicted. The method used is suitable for the rockburst warning of the excavation roadway in excavation roadway of steeply inclined ultra-thick coal seam. The established model has a good potential for predicting rock burst risk, and the research results can provide reference for the evolution and development of rockburst risk in steeply inclined coal seam.

蒋新军,刘旭东,李军权,等.基于机器学习的急倾斜巨厚煤层掘进巷道冲击地压预测方法[J].采矿与岩层控制工程学报,2024,6(3):033037.

JIANG Xinjun, LIU Xudong, LI Junquan, et al. Prediction method of rock burst in drivage roadway in steeply inclined ultra-thick coal seam based on machine learning[J]. Journal of Mining and Strata Control Engineering, 2024, 6(3): 033037.