我国煤矿智能化建设的背景下，从海量冲击地压监测数据提取出预警指标和建立有效的时空预警方法是目前研究的重点。论文以宽沟煤矿冲击地压工作面为工程背景，采用岩石力学实验、理论分析、机器学习和现场应用相结合，研究了煤岩组合体失稳声发射监测指标的响应特征及其破坏规律，计算分析了冲击地压时空预警指标前兆特征，基于多参量预警指标构建了冲击地压时空预警方法。

       针对宽沟煤矿I010203缓倾斜特厚综放面冲击显现分析，发现煤岩冲击倾向性、区段煤柱承载、坚硬顶板等为冲击地压主要影响因素。采用岩石力学单轴实验机和声发射监测系统开展煤岩组合体实验，发现试件临近失稳阶段b值处于低值波动、熵值在较高水平波动，试件在处于极值附近发生破坏，加卸载幅度越大随循环次数的增加对试件造成损伤程度更大，加卸载幅度小随循环次数增大煤岩体储能逐渐增大，试件易达到储能极限而破坏。基于宽沟煤矿I010203工作面的微震数据，运用冲击地压预警的力学基础，通过综合理论分析、现场观测和验证等方法，数值计算得到了时间及空间预警指标，优选b值、a值、A(b)值、缺震、△F值、S值和熵值H(t)为时间预警指标，指标曲线演化至极值处、接近极值的下降或回升阶段、指标曲线突增至较高值是孕育大能量事件发生的前兆。优选缺震、微震聚集度D_p、能量密度M_e、S值、AC值、△F值为空间预警指标，确定出阈值进而划分了工作面的风险区域。根据预警准确率分别确定了最终时间与空间的预警指标组合，并结合时间和空间数据特征构建了时空数据集，运用贝叶斯算法（Bayes）优化长短期记忆网络（LSTM）与极限梯度提升树（XGBoost）超参数建立了冲击地压时间和空间预警模型，基于时间和空间预警数据集对构建的模型进行了训练及测试，时间预警模型（Bayes-LSTM）在测试集上的准确率为84.8%；对建立的空间预警模型（Bayes-XGBoost）在设定的不同预警能级和提前预警时间进行了测试，得出了Bayes-XGBoost模型在提前预警能级为4-5且提前预警时间为24-48h范围内，测试集的准确率在76.5%-91.3%范围，预警模型在工程检验上同样有着良好的预警效果。

       本研究对冲击地压时空预警研究提供了新的思路，具有科学意义和参考价值。

       In the context of intelligent construction of coal mines in China, extracting early warning indicators from massive rockburst monitoring data and establishing effective spatio-temporal early warning methods are current research priorities. This thesis focuses on the rockburst working face of Kuangou Coal Mine, employing rock mechanics experiments, theoretical analysis, machine learning, and field applications to study the response characteristics and failure patterns of acoustic emission monitoring indicators in coal-rock combinations. It analyzes precursor features of spatio-temporal early warning indicators for rockbursts and constructs a method based on multiple early warning indicators.

       In the analysis of rockburst occurrences in the gently inclined extra-thick fully mechanized top-coal caving face of Kuangou Coal Mine I010203, coal-rock burst propensity, section pillar bearing, and hard roof are identified as the main influencing factors of rockburst. Experiments using a rock mechanics uniaxial testing machine and an acoustic emission monitoring system reveal that specimen instability is preceded by low b values, high entropy levels, and destruction near extreme values. Greater load-unload amplitudes cause increased damage with cycle repetition, while smaller amplitudes lead to a gradual increase in energy storage in the coal-rock body, making the specimens more likely to reach energy storage limits and fail. Based on microseismic data from the Kuangou Coal Mine I010203 working face and the mechanical basis of rockburst early warning, temporal and spatial early warning indicators were numerically calculated through comprehensive theoretical analysis, field observations, and validations. Optimal temporal indicators such as b value, a value, A(b) value, seismic quiescence, △F value, and entropy H(t) indicate the onset of significant energy events when curves evolve to extremes or when there is a sharp increase to high values. Optimal spatial indicators include seismic quiescence, microseismic clustering D_p, energy density M_e, S value, AC value, and △F value. Thresholds are established to delineate risk areas on the working face. Based on the accuracy of the warnings, the final combinations of temporal and spatial early warning indicators were determined. These were combined with temporal and spatial data characteristics to construct a spatio-temporal dataset. Bayesian algorithms optimized the hyperparameters of the Long Short-Term Memory network (LSTM) and the Extreme Gradient Boosting tree (XGBoost) to establish temporal and spatial rockburst early warning models. These models were trained and tested using the temporal and spatial early warning datasets. The temporal warning model (Bayes-LSTM) achieved an accuracy of 84.8% on the test set. The spatial warning model (Bayes-XGBoost) was tested across different early warning levels and times, achieving accuracies ranging from 76.5% to 91.3% for early warnings between 24-48 hours, demonstrating effective warning performance in engineering validation.

       This study provides new insights into the research of spatio-temporal early warning for rockburst, having scientific significance and reference value.

[1] 谢和平,任世华,谢亚辰,等. 碳中和目标下煤炭行业发展机遇[J]. 煤炭学报,2021,46(7):2197-2211.

[2] 岳福斌. 2019-2020年中国煤炭产业经济形势研究报告[J]. 中国煤炭,2020,46(3):5-12.

[3] 翟明国,胡波. 矿产资源国家安全、国际争夺与国家战略之思考[J]. 地球科学与环境学报,2021,43(1):1-11.

[4] 齐庆新,马世志,孙希奎,等. 煤矿冲击地压源头防治理论与技术架构[J]. 煤炭学报,2023,48(5):1861-1874.

[5] 王国法,潘一山,赵善坤,等. 冲击地压煤层如何实现安全高效智能开采[J]. 煤炭科学技术,2024,52(1):1-14.

[6] GA Zhu,LM Dou,AY Cao,et al. Assessment and analysis of strata movement with special reference to rock burst mechanism in island longwall panel[J]. Journal of Central South University,2017,24(12):2951-2960.

[7] F Cui,S Dong,XP Lai,et al. Study on the Fracture Law of Inclined Hard Roof and Surrounding Rock Control of Mining Roadway in Longwall Mining Face[J]. Energies,2020,13(20):5344.

[8] 齐庆新,彭永伟,李宏艳,等. 煤岩冲击倾向性研究[J]. 岩石力学与工程学报,2011,30(S1):2736-2742.

[9] 姚精明,闫永业,李生舟,等. 煤层冲击倾向性评价损伤指标[J]. 煤炭学报,2011,36(S2):353-357.

[10] 国家煤炭工业局.煤层冲击倾向性分类及指数的测定方法:MT/T174-2000[S]北京:中国标准出版社,2000.

[11] 夏方迁.多层厚硬岩层工作面冲击地压发生机理及监测技术研究[D]. 中国矿业大学(北京),2021.

[12] 赵本均. 冲击地压及防治[M]. 北京:煤炭工业出版社,1995:189-204.

[13] 潘立友,钟亚平.深井冲击地压及其防治[M]. 北京:煤炭工业出版社,997:89-92.

[14] 窦林名,何江,曹安业等.煤矿冲击矿压动静载叠加原理及其防治[J].煤炭学报,2015,40(7):1469-1476.

[15] 窦林名,陆菜平,牟宗龙,等. 冲击矿压的强度弱化减冲理论及其应用[J]. 煤炭学报,2005,(6):690-694.

[16] 潘一山,李忠华,章梦涛. 我国冲击地压分布、类型、机理及防治研究[J]. 岩石力学与工程学报,2003,22(11):1844-1851.

[17] 潘俊锋,宁宇,毛德兵,等. 煤矿开采冲击地压启动理论[J]. 岩石力学与工程学报,2012,31(3):586-596.

[18] W Cook N G. The failure of rock[J]. International Journal of Rock Mechanics and Mining Sciences&Geomechanics Abstracts,1965,15(4):389-403.

[19] 李玉生.冲击地压机理及其初步应用[J]. 中国矿业学院学报,1985,10(3):42-48.

[20] 李玉生.冲击地压机理探讨[J]. 煤炭学报,1984(3):1-10.

[21] 齐庆新,史元伟,刘天泉.冲击地压粘滑失稳机理的实验研究. [J]煤炭学报,1 997,22(2):34-38.

[22] 齐庆新,刘天泉,史元伟,等. 冲击地压的摩擦滑动失稳机理[J]. 矿山压力与顶板管理,1995,23(Z1):174-177,200.

[23] 齐庆新,毛德兵,王永秀.冲击地压的非线性非连续特征[J]. 岩土力学,2003,24(S2):575-579.

[24] Cook N G W. A note on coal bumps considered as a problem of stability[J]. Journal of the Southern African Institute of Mining and Metallurgy,1965,65(8):437-446.

[25] Cook N G W. The basic mechanics of coal bumps[J]. Journal of the Southern African Institute of Mining and Metal1urgy,1963,64(3):71-81.

[26] 潘一山,徐连满. 钻屑温度法预测冲击地压的试验研究[J]. 岩土工程学报,2012,34(12):2228-2232.

[27] 章梦涛,赵本钧,徐曾和.钻屑法的实验研究[C]. 地下工程经验交流会论文选集，北京:地质出版社,1982:369.

[28] Yaodong Jiang,Yixin Zhao,Hongwei Wang,et al. A review of mechanism and prevention technologies of coal bumps in China[J]. Journal of Rock Mechanics and Geotechnical Engineering,2017,9(1):180-19.

[29] Shan Qiyuan,Qin Tao. The Improved Drilling Cutting Method and Its Engineering Applications[J]. Springer International Publishin g,2019,37(5):3715-3726.

[30] 刘伟建,刘晔,潘贵豪. 基于应力监测系统的工作面终采线确定方法研究[J]. 煤炭工程,2019,51(6):112-115.

[31] Petr Konicek,Jan Schreiber. Heavy rockbursts due to longwall mining near protective pillars:A case study[J]. International Journal of Mining Science and Technology,2018,28(5):799-805.

[32] WH Yang, XP Lai,JT Cao,et al. Study on evolution characteristics of overburden caving and void during multi-horizontal sectional mining in steeply inclined coal seams[J]. Thermal Science,2020,24(6):3915-392.

[33] 谢广祥,杨科. 采场围岩宏观应力壳演化特征[J]. 岩石力学与工程学报,2010,29(S1):2676-2680.

[34] 吴健波. 冲击地压电磁辐射实时监测及自动预警研究[D]. 徐州:中国矿业大学,2018.

[35] 贺虎,孙昊,王茜. 冲击矿压危险的电磁-震动耦合评价[J]. 煤炭学报,2018,43(2): 364-370.

[36] Li Xuelong，Wang Enyuan,Li Zhonghui. Rock Burst Mon itoring by IntegratedMicroseismic and Electromagnetic Radiation Methods[J]. Springer Vienna,2016,49(11):4393-4406.

[37] 王永,刘金海,王颜亮,等. 煤矿冲击地压多参量监测预警平台研究[J]. 煤炭工程,2018,50(4):19-21.

[38] 姜耀东,吕玉凯,赵毅鑫,等. 综采工作面过断层巷道稳定性多参量监测[J]. 煤炭学报,2011,36(10):1601-1606.

[39] 赵善坤,李宏艳,刘军,等. 深部冲击危险矿井多参量预测预报及解危技术研究[J]. 煤炭学报,2011,36(S2):339-345.

[40] 贾瑞生,孙红梅,樊建聪,等. 一种冲击地压多参量前兆信息辨识模型及方法[J]. 岩石力学与工程学报,2014,33(8):1513-1519.

[41] Gutenberg B,Richter C.Frequency of earthquakes in California[J]. Bulletinof the Seismological Society of America,1994,34(4):185-188.

[42] 夏永学,康立军,齐庆新,等. 基于微震监测的5个指标及其在冲击地压预测中的应用[J]. 煤炭学报,2010,35(12):2011-2016.

[43] 刘建坡. 深井矿山地压活动与微震时空演化关系研究[D]. 沈阳:东北大学,2011.

[44] 谷继成,魏富胜. 论地震活动性的定量化:地震活动度[J]. 中国地震,1987,15(S1):14-24.

[45] 刘辉,陆菜平,窦林名,等. 微震法在煤与瓦斯突出监测与预报中的应用[J]. 煤矿安全,2012,43(4):82-85.

[46] 蔡武. 断层型冲击矿压的动静载叠加诱发原理及其监测预警研究[D]. 徐州:中国矿业大学,2015.

[47] 王盛川. 褶皱区顶板型冲击矿压“三场”监测原理及其应用[D]. 徐州:中国矿业大学,2021.

[48] 耿越. 基于深度卷积模型的煤矿顶板危险性预测与评价研究[D]. 北京:中国矿业大学(北京),2019.

[49] 贾义鹏,吕庆,尚岳全. 基于粒子群算法和广义回归神经网络的岩爆预测[J]. 岩石力学与工程学报,2013,32(2):343–348.

[50] Liang W,Sari A,Zhao G,et al. Short-term rockburst risk pre-diction using ensemble learning methods[J]. Natural Haz-ards,2020,104(2):1923-1946.

[51] Ahmad M,Hu J L,Hadzima-Nyarko M,et al. Rockburst Hazard Prediction in Underground Projects Using Two Intel-ligent Classification Techniques:A Comparative Study[J]. Symmetry-Basel,2021,13(4):632.

[52] 周超. 梯度载荷下煤层开采压转耦合诱冲机理与防治研究[D]. 北京科技大学,2023.

[53] 谭云亮,张修峰,肖自义,等. 冲击地压主控因素及孕灾机制[J]. 煤炭学报,2024,49(1):367-379.

[54] 左建平,谢和平,孟冰冰,等. 煤岩组合体分级加卸载特性的试验研究[J]. 岩土力学,2011,32(5):1287-96.

[55] 李欣慰,姚直书,黄献文,等. 循环加卸载下砂岩变形破坏特征与能量演化研究[J]. 岩土力学,2021,42(6):1693-704.

[56] 秦四清,李造鼎. 岩石声发射事件在空间上的分形分布研究[J]. 应用声学,1992,11(4):19－21.

[57] 刘希灵,潘梦成,李夕兵,等. 动静加载条件下花岗岩声发射b值特征的研究[J]. 岩石力学与工程学报,2017,36(S1):3148-3155.

[58] 牛勇. 裂隙岩石破裂过程声发射表征及失稳预警研究[D]. 重庆大学,2021.

[59] 魏嘉磊,刘善军,吴立新,等. 含孔岩石双轴加载过程声发射多参数特征对比分析[J]. 采矿与安全工程学报,2015,32(6):1017-1025.

[60] 吴再海,宋朝阳,谭杰等. 不同分级循环加卸载模式下岩石能量演化规律研究[J]. 采矿与安全工程学报,2020,37(4):836-844+851.

[61] 张如九,张延杰,高仝,等. 基于最大能量耗散率的岩爆倾向性指标研究[J]. 岩石力学与工程学报,2023,42(12):2993-3009.

[62] 贾冲. 综放面推进速度对采动应力与煤样力学特性的影响研究[D]. 西安科技大学,2021.

[63] 张子涵. 大屯矿区冲击地压多参量监测预警模型研究[D]. 中国矿业大学,2022.

[64] Holub K. Space-time variations of the frequency-energy relation for mining-induced seismicity in the Ostrava-Karviná mining district[J]. Pure Appl Geophys, 1996, 146: 265-280.

[65] 吴佳翼, 曹学锋. 地震活动性的定量化问题[J]. 地震, 1983(6): 13-16+22.

[66] Tokuji U,Yosihiko O,Ristuko S. The centenary of the Omori formula for a decay law of after shock activity[J]. Earth Planets and Space,1995,43(1):1-33.

[67] 吕悦军, 陆远忠. 用算法复杂性分析时间序列[J]. 中国地震, 1993(3): 39-44.

[68] XUE YR,SONG DZ,CHEN JQ,et al. Integrated rockburst hazard estimation methodolo-

gy based on spatially smoothed seismicity model and Mann-Kendall trend test[J]. International Journal of Rock Mechanics and Mining Sciences,2023,163:105329.

[69] CAI W,DOU L M,GONG S Y,et al. Quantitative analysis of seismic velocity tomography in rock burst hazard assessment[J]. Natural Hazards,2015,75(3):2453–2465.

[70] 何生全,何学秋,宋大钊,等. 冲击地压多参量集成预警模型及智能判识云平台[J]. 中国矿业大学学报,2022,51(5):850-862.

[71] 何生全. 近直立煤层群综放开采冲击地压机理及预警技术研究[D]. 北京科技大学,2021.

[72] 曹安业,刘耀琪,杨旭等.物理指标与数据特征融合驱动的冲击地压时序预测方法[J].煤炭学报,2023,48(10):3659-3673.

[73] 杨旭,刘亚鹏,曹安业,等. 基于微震多维信息融合的冲击地压全时空预测方法[J/OL].采矿与安全工程学报.

[74] 崔峰,何仕凤,来兴平,等. 基于相空间重构与深度学习的冲击地压矿井时间序列b值趋势[J]. 煤炭学报,2023,48(5):2022-2034.

[75] 李伟山,王琳,卫晨. LSTM在煤矿瓦斯预测预警系统中的应用与设计[J].西安科技大学学报,2018,38(6):1027–1035.

[76] 刘慧敏,徐方远,刘宝举,等. 基于CNN–LSTM的岩爆危险等级时序预测方法[J]. 中南大学学报(自然科学版),2021,52(3):659–670.

[77] 邓新国,游纬豪,徐海威. 贝叶斯极限梯度提升机结合粒子群算法的电阻点焊参数预测[J]. 电子与信息学报,2021,43(4):1042-1049.

[78] 高永涛,朱强,吴顺川,等. 基于AVOA-XGBoost模型的岩爆预测研究[J]. 华中科技大学学报(自然科学版),2023,51(12):151-157.

[79] 丁昌伟,王新,陈同俊,等. 贝叶斯优化的XGBoost在小断层地震解释中的应用[J]. 煤炭学报,2023,48(6):2530-2539.