School of Computer Science & Technology, Engineering Research Center of Digital Mine, Ministry of Education, China University of Mining and Technology, Xuzhou
School of Mines, China University of Mining and Technology, Xuzhou
State Key Laboratory for Fine Exploration and Intelligent Development of Coal Resources, China University of Mining and Technology, Xuzhou

为解决当前冲击地压时间与空间预测协同难、微震数据时空特征挖掘不充分的困境,结合深度学习相关理论与方法,提出了基于微震多维信息融合的冲击地压全时空预测方法,该方法主要包括微震时空特征指标、时间预测以及空间预测3个模块,设计了基于主成分分析和核密度估计的微震时空特征指标构建方法,在此基础之上,构建了基于深度循环神经网络的冲击地压时间预测模型,提出了基于长短期时间窗融合的冲击地压空间预测方法,从而实现了冲击地压时间-空间协同的全时空预测。此外,为了评估所提方法的有效性,在内蒙古鄂尔多斯矿区某冲击危险工作面进行了工程应用测试,测试时间段共出现13条大于105J的大能量微震事件,在时间预测方面,对于大能量事件的时间预测结果为10个强危险、3个中等危险,并且整个测试阶段模型误报率仅为0.133。在空间预测方面,对于大能量事件的空间预测结果的分布区域为6个强危险、3个中等危险、4个弱危险。实验表明该方法可满足工程应用的需求,研究成果可为冲击地压监测预警提供参考与借鉴。

It is difficult to cooperate temporal and spatial prediction of coal burst and spatio-temporal feature mining of massive microseismic data are insufficient. A spatio-temporal prediction method for coal burst is proposed based on the fusion of microseismic multidimensional information with the relevant theories and methods of deep learning. The process mainly includes three modules: microseismic spatio- temporal characteristic index, temporal prediction and spatial prediction. Microseismic spatio-temporal characteristic index method is based on principal component analysis and kernel density estimation, withwhich temporal prediction model of coal burst based on deep recurrent neural network is constructed, and spatial prediction method of coal burst based on the fusion of long and short time windows is pro- posed, thus realizing the spatio-temporal prediction of coal burst with spatio-temporal coordination. To e- valuate the effectiveness of the method, engineering application tests are conducted on a hazardous work- ing face in the Ordos mining area of the Inner Mongolia Autonomous Region. During the test, 13 large energy events (microseismic events with energy greater than 10 5 J) occur. In temporal prediction, the prediction results are 10 strong hazards and 3 medium hazards, and the false positive rate of the model is only 0.133. In temporal prediction, the distribution region of large energy events corresponds to 6 strong hazards, 3 medium hazards and 4 weak hazards. The test results show that the method can meet the spa- tio-temporal requirements of engineering application, and the research results can provide a paradigm for spatio-temporal prediction of coal burst source.

国家重点研发计划项目(2022YFC3004603);国家自然科学基金项目(52274098);江苏省自然科学基金项目(BK20221109);江苏省创新支撑计划国际科技合作/港澳台科技合作———重点国别产业技术研发合作项目(BZ2023050)