Rockburst is a severe dynamic hazard that poses a critical threat to safe and efficient deep coal mining. Conventional rockburst prediction approaches commonly depend on single-source information, resulting in incomplete characterization of the spatiotemporal evolution of precursory signals. In this study, a spatiotemporal feature fusion and multimodal-driven framework, termed MDL-RBP, was developed for rockburst hazard prediction. Specifically, microseismic waveforms were transformed into time‑frequency representations using the Short-Time Fourier Transform (STFT), Hilbert‑Huang Transform (HHT), and Continuous Wavelet Transform (CWT), while eleven geophysical‑statistical indicators (e.g., b-value, a-value, event frequency, and energy extrema) were extracted to establish a complementary dataset integrating spatial, time‑frequency, and physical‑statistical information. To mitigate pronounced class imbalance, a unified scheme was implemented across data-, algorithm-, and evaluation-level settings to enhance recognition of hazardous categories. Furthermore, a transformer coupled with a multilayer perceptron (MLP) was employed to model spatiotemporal dependencies, and bidirectional cross-attention was introduced to facilitate deep coupling among heterogeneous feature sources. An adaptive gating fusion module was finally incorporated to learn modality contributions in a data-driven manner, thereby improving generalization. Experimental results show that MDL-RBP achieves an accuracy of 90.9% outperforming the best single-modal baseline by 15 percentage points, and attains 100% recall for the most hazardous class (Level Ⅳ), eliminating missed alarms. Across multi-horizon forecasts from 3 to 30 days, Level Ⅳ recall remains 100% for short-term horizons (3—7 days), and the 30-day horizon maintains an accuracy exceeding 86%. The proposed method advances multimodal spatiotemporal modeling for rockburst prediction and offers a practical technical solution for accurate hazard forecasting in deep coal mines.
式中: u 为物理指标、时序流与空间流特征拼接后的联合向量; g 为门控向量;Sigmoid(·)为Sigmoid激活函数;为门控层的可学习权重矩阵;为门控层的偏置向量; hfused为经门控加权后的最终融合特征向量;为逐元素乘法(Hadmard积). g 对每个特征维度进行加权,相当于“软性模态选择”.训练过程中,模型会自动学习哪些模态对当前样本更重要.例如,对于高危样本,能量相关特征可能获得更高权重;对于空间异常主导的样本,空间流权重会升高.这种自适应性提升了模型对不同冲击类型的泛化能力,同时增强了对噪声模态和模态缺失的鲁棒性.
CHENGXianggang, QIAOWei, DOULinming, et al.Distribution characteristics and complexity evaluation method of in‑situ stress field in the Ningzheng-Binchang coal mining area distribution law of in-situ stress field and complexity evaluation method in Ningzheng-Binchang coal mining area[J]. Journal of China University of Mining & Technology, 2026,55(2): 499-513.
CAOAnye, LIANGYingxin, YANGXu, et al. Research progress on microseismic early warning methods of rockburst in coal mines empowered by artificial intelligence[J/OL]. Journal of China University of Mining & Technology, 1-22[2026-03-16].
PANYishan, SONGYimin, LIUJun. Pattern,change and new situation of coal mine rockburst prevention and control in China [J]. Chinese Journal of Rock Mechanics and Engineering, 2023, 42(9): 2081-2095.
CUIFeng, HEShifeng, LUOZhong, et al. Research on multi-index early warning of rock burst based on bayesian optimization algorithm and machine learning[J]. Journal of China Coal Society, 2025, 50(Sup 1):297-313.
CAOAnye, WANGChangbin, YANGXu,et al. Fractures characterization in mining field considering seismic location accuracy and its application on pre-warning coal burst hazards[J]. Coal Science and Technology, 2024, 52(2): 1-9.
ZHUGuang'an, WANGYi, XUZihao,et al. Inversion of equivalent drill cuttings volume of water-bearing coal mass basedon acoustic emission signal processing in mining engineering[J]. Chinese Journal of Rock Mechanics and Engineering, 2025, 44(8): 2040-2054.
[13]
CHENGX G, QIAOW, HEH. Study on deep learning methods for coal burst risk prediction based on mining-induced seismicity quantification[J]. Geomechanics and Geophysics for Geo-Energy and Geo-Resources, 2023, 9(1): 145.
ZHUSitao, JIANGFuxing, LIUJinhai,et al. Mechanism and monitoring and early warning technology of rock burst inthe heading face of compound thick coal seam[J]. Journal of China Coal Society, 2020, 45(5): 1659-1670.
[16]
BASNETP M S, MAHTABS, JINA B, et al. A comprehensive review of intelligent machine learning based predicting methods in long-term and short-term rock burst prediction[J]. Tunnelling and Underground Space Technology, 2023, 142: 105434.
[17]
QIUY G, ZHOUJ. Short-term rockburst prediction in underground project: Insights from an explainable and interpretable ensemble learning model[J]. Acta Geotechnica, 2023, 18(12): 6655-6685.
LIUH M, XUF Y, LIUB J, et al. Time-series prediction method for risk level of rockburst disaster based on CNN-LSTM[J]. Journal of Central South University (Science and Technology), 2021, 52(3): 659-670.
[20]
ZHANGS C, MUC M, FENGX H, et al. Intelligent dynamic warning method of rockburst risk and level based on recurrent neural network[J]. Rock Mechanics and Rock Engineering, 2024, 57(5): 3509-3529.
WANGShaochen, WANGChao, LIShidong, et al. Time series prediction model of microseismicity in coal mine based on adaptive ensemble learning[J]. Journal of Mining and Strata Control Engineering, 2025, 7(4): 247-262.
WANGAiwen, LIChao, PANYishan,et al. Dynamic identification method for rockburst hazard areas based on multivariate geophysical indicators and its application[J]. Journal of China Coal Society, 2024, 49(6): 2573-2588.
[25]
CAOA Y, LIUY Q, YANGX, et al. FDNet: Knowledge and data fusion-driven deep neural network for coal burst prediction[J]. Sensors, 2022, 22(8): 3088.
[26]
LIANGR Y, ZHANGC G, HUANGC R, et al. Multimodal data fusion for geo-hazard prediction in underground mining operation[J]. Computers & Industrial Engineering, 2024, 193: 110268.
[27]
BAIJ, WANGS, XUQ, et al. Adaptive threshold multimodal fusion for rock prediction in complex geological environments while drilling[J]. Bulletin of Engineering Geology and the Environment, 2025, 84(4): 178.
CUIFeng, ZONGCheng, LAIXingping,et al. Intelligent prediction of time series and grade of rock burst in steeply inclined ultra-thick coal seam excavation roadway[J]. Journal of China Coal Society, 2025, 50(2): 845-861.
WANGShuwen .Study on calculation sample and parameter selection of mine microseism signal b value[J]. Coal Science and Technology, 2016, 44(12) : 51-56
[32]
GUTENBERGB, RICHERC F. Frequency of earthquakes in California[J]. Bulletin of the Seismological Society of America,1944, 34(4): 185-188.
[33]
吴佳翼,曹学锋. 地震活动性的定量化问题[J]. 地震,1983(6): 13-16.
[34]
WUJiayi, CAOXuefeng. On the quantification of seismicity[J]. Earthquake, 1983(6): 13-16.
[35]
卫鹏飞. 华北缺震曲线[J]. 地震研究,1980(4): 144-148.
[36]
WEIPengfei. Seismic gap curve in North China[J]. Journal of Seismological Research, 1980(4): 144-148.
CUIFeng, ZHANGSuilin, LAIXingping,et al. Mechanism of rock burst induced by mine earthquake and the time⁃space characteristics of in steep and extremely thick coal seam[J]. Journal of China Coal Society, 2023, 48(Sup 2): 449-463.
GUJicheng, WEIFusheng. On the quantification of seismicity: Seismic activity[J]. Earthquake Research in China, 1987, 3(Sup 1): 14-24.
[41]
CHENS H, ZHANGB H, PENGC B, et al. Double dominant frequency of microseismic signal and micro-fracture mechanism of granite rockmass[J]. Geomechanics and Geophysics for Geo-Energy and Geo-Resources, 2025,11(1): 115.
[42]
LIANGZ Z, XUER X, XUN W, et al. Characterizing rockbursts and analysis on frequency-spectrum evolutionary law of rockburst precursor based on microseismic monitoring[J]. Tunnelling and Underground Space Technology,2020,105:103564.
[43]
MAT H, TANGC A, TANGS B, et al. Rockburst mechanism and prediction based on microseismic monitoring[J]. International Journal of Rock Mechanics and Mining Sciences, 2018, 110: 177-188.
[44]
XUH C, LAIX P, ZHANGS, et al. Precursor information recognition of rockburst in the coal-rock mass of meizoseismal area based on multiplex microseismic information fusion and its application:A case study of Wudong coal mine[J]. Lithosphere, 2022, 2022(11): 7349759.
[45]
ZHOUJ, LIX B, MITRIH S. Strength of stacking technique of ensemble learning in rockburst prediction with imbalanced data: Comparison of eight single and ensemble models[J]. Natural Resources Research, 2021, 30(2): 1795-1815.
[46]
HEK M, ZHANGX Y, RENS Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016: 770-778.
[47]
HUJ, SHENL, SUNG. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2018: 7132-7141.
[48]
DAIJ, QIH, XIONGY, et al. Deformable convolutional networks[C]//Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2017: 764-773.
[49]
WANGY, SUNY B, LIUZ W, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 146.