煤炭作为我国能源的主要来源之一，对维持我国经济的稳定发展起着关键作用。随着我国矿山开采深度的加深，因矿压隐患造成的安全事故不断增加，极大的威胁着井下矿工的人身安全，影响矿井的正常生产。为准确捕捉工作面的矿压变化规律，提高矿压预测精度，减少煤矿顶板事故的发生，本文分别提出了基于周期分解Transformer的矿压预测方法和基于特征融合-Transformer的综采工作面矿压时空预测方法。其主要的研究内容如下：

（1）矿压周期性对矿压序列预测影响的研究。为了能够在复杂的时间序列中学习长期依赖信息，提出了基于周期分解Transformer的矿压预测方法。首先，对矿压原始数据进行预处理；其次，将周期分解模块嵌入到Transformer结构中作为内部操作算子构建周期分解Transformer预测模型，可以逐步从预测的隐藏变量中分离出长期趋势信息；最后，利用优化算法进行参数优化，并采用均方误差损失函数进行模型训练，从而实现综采工作面上矿压的预测。实验结果表明，相较于基于BP神经网络、GRU、LSTM以及Transformer模型的矿压预测方法，本文所提的基于周期分解Transformer模型的矿压预测方法在平均绝对误差(MAE)和均方根误差(RMSE)性能指标方面分别减少了17.8%和24.5%，因此该方法具有更高的准确性。

（2）多支架矿压时空特性对矿压序列预测影响的研究。在采煤机推进的过程中，岩层的空间结构不断变化，使得支架压力采集器监测到的多条矿压序列数据在空间上存在相关性。为了充分挖掘矿压序列的潜在特征，提升预测准确率，本文提出了一种基于特征融合-Transformer模型的综采工作面矿压时空预测方法。该方法对多个矿压序列之间的空间关联关系进行灰色关联度分析，选取强关联度序列组；然后，对工作面矿压序列的数值信息、时间信息和空间位置信息进行特征嵌入并融合得到特征向量，Transformer模型接收矿压序列的特征向量，并挖掘矿压序列的非线性时序依赖关系；最后，通过全连接网络构成的预测层，实现综采工作面矿压的时空预测。实验结果表明，相较于基于堆叠LSTM、堆叠GRU和Transformer模型的矿压预测方法，本文所提的基于特征融合-Transformer模型的综采工作面矿压时空预测方法在平均绝对误差(MAE)和均方根误差(RMSE)性能指标方面分别减少了22.3%和25.1%，可以有效提高矿压预测精度。

综上所述，基于周期分解Transformer的矿压预测方法通过对工作面矿压时间序列进行周期性分解，有效捕捉到序列中的长期趋势信息，从而提高了矿压预测精度。此外，基于特征融合-Transformer模型的综采工作面矿压时空预测方法充分挖掘多个矿压序列间的数值信息、时间信息和空间位置信息，可以提高对目标支架序列的矿压预测准确度。上述方法的研究可以有效缓解煤矿井下综采工作面矿压灾害的发生，为保障煤矿安全生产提供理论支撑。

As one of the main sources of energy in China, coal plays a key role in maintaining the stable development of China's economy. With the deepening of mining depth in China, safety accidents caused by potential mine pressure have been increasing, which greatly threaten the personal safety of underground miners and affect the normal production of mines. In order to accurately capture the mine pressure change pattern at the working face, improve the accuracy of mine pressure prediction, and reduce the occurrence of coal mine roof accidents, this thesis proposes the mine pressure prediction method based on Periodic Decomposition Transformer and the spatio-temporal prediction method of mine pressure at the comprehensive mining working face based on Feature Fusion-Transformer, respectively. Its main research contents are as follows:

(1) The study of the influence of mine pressure periodicity on the prediction of mine pressure sequences. In order to be able to learn long-term dependence information in complex time series, the method of mine pressure prediction based on the period decomposition Transformer is proposed. First, the raw mine pressure data is preprocessed; second, the period decomposition module is embedded into the Transformer structure as an internal operation operator to construct the period decomposition Transformer prediction model, which can gradually separate the long-term trend information from the predicted hidden variables; finally, the optimization algorithm is used to optimize the parameters and the mean square error loss function is used to train the model. Thus, the prediction of mine pressure on the comprehensive mining face is realized. The experimental results show that compared with the BP neural network, GRU, LSTM and Transformer model-based mining pressure prediction methods, the proposed period decomposition Transformer model-based mining pressure prediction method has a 17.8% and 24.5% reduction in the mean absolute error (MAE) and root mean square error (RMSE) performance indexes, respectively. The method has higher accuracy.

(2) Study on the influence of spatial and temporal characteristics of multi-stent mine pressure on the prediction of mine pressure sequences. In the process of coal miner advancing, the spatial structure of rock strata keeps changing, which makes the data of multiple mine pressure sequences monitored by the holder pressure collector correlate in space. In order to fully exploit the potential features of mine pressure sequences and improve the prediction accuracy, this thesis proposes a spatio-temporal prediction method of mine pressure at the comprehensive mining working face based on the feature fusion-Transformer model. The method performs gray correlation analysis on the spatial correlations among multiple mine pressure sequences and selects a strong correlation sequence group; then, the numerical information, temporal information and spatial location information of mine pressure sequences at the working face are feature-embedded and fused to obtain feature vectors, and the Transformer model receives the feature vectors of mine pressure sequences and mines the nonlinear temporal dependencies of mine pressure sequences; finally, the The spatio-temporal prediction of mine pressure at the heaving workface is achieved by the prediction layer composed of fully connected network. The experimental results show that compared with the stacked LSTM, stacked GRU and Transformer models, the proposed feature fusion-Transformer model-based spatio-temporal prediction method for mine pressure at the heaving workface reduces the mean absolute error (MAE) and root mean square error (RMSE) performance indexes by 22.3% and 25.1%, which can effectively improve the accuracy of mine pressure prediction.

In summary, the mine pressure prediction method based on periodical decomposition Transformer effectively captures the long-term trend information in the sequence by periodically decomposing the workface mine pressure time series, thus improving the mine pressure prediction accuracy. In addition, the spatio-temporal mining pressure prediction method based on feature fusion-Transformer model for comprehensive mining working face fully exploits the numerical information, temporal information and spatial location information among multiple mining pressure sequences, which can improve the accuracy of mining pressure prediction for the target bracket sequence. The research of the above method can effectively alleviate the occurrence of mine pressure disasters in underground coal mine comprehensive mining working face, and provide theoretical support to ensure the safe production of coal mines.

[1] 钱鸣高, 许家林, 王家臣. 再论煤炭的科学开采[J]. 煤炭学报, 2018, 43(1): 1–13.

[2] 丁慧霞. 5G技术在煤矿智能化建设中的应用展望[J]. 内蒙古煤炭经济, 2022, (17): 154–156.

[3] 田学春, 葛峰, 王正强. “十四五”中国煤矿智能化绿色开采前景展望[C]// 中国管理科学研究院商学院, 中国技术市场协会, 中国高科技产业化研究会. 第十八届中国科学家论坛论文集. 北京:《发现》杂志社, 2021: 8–10.

[4] 杨志平. 实测时间序列数据间隔提取处理方法及矿压超前预测[D]. 太原: 太原理工大学, 2022.

[5] Arslan O, Xing W, Inan F A, et al. Understanding topic duration in Twitter learning communities using data mining[J]. Journal of Computer Assisted Learning, 2022, 38(2): 513–525.

[6] 王海波, 贺耀宜. 深耕煤矿安全生产领域助力煤矿智能化建设——中煤科工集团常州研究院服务煤矿智能化建设之路[J]. 智能矿山, 2022, 3(2): 2–10.

[7] Chengqing Zong, Rui Xia, Jiajun Zhang. Text Data Mining[M]. 2021–06–07.

[8] 彭苏萍. 我国煤矿安全高效开采地质保障系统研究现状及展望[J]. 煤炭学报, 2020, 45(7): 2331–2345.

[9] 王家臣, 刘峰, 王蕾. 煤炭科学开采与开采科学[J]. 煤炭学报, 2016, 41(11): 2651–2660.

[10] Zhang B, Yang Z, Ji C, et al. Research on the Influence of the Key Stratum Position on the Support Working Resistance during Large Mining Height Top-Coal Caving Mining[J]. Advances in Civil Engineering, 2021: 1–9.

[11] 冀汶莉, 刘艺欣, 柴敬, 王斌. 基于随机森林的矿压预测方法[J]. 采矿与岩层控制工程学报, 2021, 3(3): 71–81.

[12] 王润沛. 基于机器学习的分布式光纤监测覆岩变形矿压预测研究[D]. 西安: 西安科技大学, 2020.

[13] 尹希文, 徐刚, 刘前进, 等. 基于支架载荷的矿压双周期分析预测方法[J]. 煤炭学报, 2021, 46(10): 3116–3126.

[14] 冯夏庭, 王泳嘉, 姚建国. 煤矿顶板矿压显现实时预报的自适应神经网络方法[J]. 煤炭学报, 1995, 20(5): 455–460.

[15] Feng X T, Wang Y J, Yao J G. A neural network model for real-time roof pressure prediction in coal mines[C]// International journal of rock mechanics and mining sciences & geomechanics abstracts. Pergamon, 1996, 33(6): 647–653.

[16] 贺超峰, 华心祝, 杨科, 马菁花. 基于BP神经网络的工作面周期来压预测[J]. 安徽理工大学学报(自然科学版), 2012, 32(1): 59–63.

[17] 柴敬, 王润沛, 杜文刚, 等. 基于XGBoost的光纤监测矿压时序预测研究[J]. 采矿与岩层控制工程学报, 2020, 2(4): 60–67.

[18] 程敬义, 万志军, PENG S S. 基于海量矿压监测数据的采场支架与顶板状态智能感知技术[J]. 煤炭学报, 2020, 45(6): 2090–2103.

[19] 徐刚, 范志忠, 张春会, 于健浩, 张震, 刘前进, 李正杰, 卢振龙. 宏观顶板活动支架增阻类型与预测模型[J]. 煤炭学报, 2021, 46(11): 3397–3407.

[20] 王伟强, 公维宽. 大埋深高应力巷道掘进期间矿压显现规律与冲击地压防治[J]. 煤矿现代化, 2020(1): 24–25.

[21] 朱涛, 常云博, 田春阳. 深埋坚硬煤岩大采高长壁工作面矿压观测及显现规律[J]. 中国矿业, 2020, 29(2): 124–128.

[22] 陆明心, 郝海金, 吴健. 综放开采上位岩层的平衡结构及其对采场矿压显现的影响[J]. 煤炭学报, 2002, 27(6): 591–595.

[23] 张通, 赵毅鑫, 朱广沛, 等. 神东浅埋工作面矿压显现规律的多因素耦合分析[J]. 煤炭学报, 2016(S2): 287–296.

[24] 闫吉太, 梁广锋, 安满林, 等. “孤岛”综采放顶煤工作面矿压预测预报[J]. 中国矿业大学学报, 1996, 25(4): 98–103.

[25] 窦林名, 何学秋. 煤矿冲击矿压的分级预测研究[J]. 中国矿业大学学报, 2007, 36(6): 717–722.

[26] 庞耀龙. 我国采煤沉陷区现状、综合治理模式及治理建议[J]. 西部探矿工程, 2020, 32(1): 153–155.

[27] 赵毅鑫, 杨志良, 马斌杰, 等. 基于深度学习的大采高工作面矿压预测分析及模型泛化[J]. 煤炭学报, 2020, 45(1): 54–65.

[28] 顾清华, 江松, 李学现, 等. 人工智能背景下采矿系统工程发展现状与展望[J]. 金属矿山, 2022, 51(5): 10–25.

[29] 梁沙平, 陆银龙, 郭鹏等. 特厚煤层坚硬顶板初次破断特征的力学分析[J]. 煤矿安全, 2020, 51(08): 245–250.

[30] 杨俊哲, 郑凯歌, 王振荣, 等. 坚硬顶板动力灾害超前弱化治理技术[J]. 煤炭学报, 2020, 45(10): 3371–3379.

[31] 乔懿麟, 王苏健, 吴学明, 等. 基于模糊综合评价的冲击矿压危险性信息平台研究[J]. 煤炭技术, 2016(11): 253–255.

[32] 惠艳荔. 电磁辐射技术在采煤工作面矿压预测中的应用[J]. 煤炭技术, 2013, 32(12): 73–75.

[33] 李静, 何江, 巩思园, 等. 上覆遗留煤柱作用下冲击矿压预测预警案例研究[J]. 煤炭学报, 2016, 41(S2): 305–310.

[34] 郭辉, 王新萍, 芦盛亮, 等. 基于微震监测的高强度开采工作面煤岩破坏规律研究[J]. 矿业研究与开发, 2018, 38(12): 52–56.

[35] 刘昆轮, 闫瑞兵. 基于地音监测的近直立煤层冲击地压前兆特征研究[J]. 煤炭工程, 2020, 52(4): 48–51.

[36] 刘卫东, 孟晓静, 丁恩杰, 等. 模糊逻辑在矿井冲击矿压预测中的应用[J]. 煤矿安全, 2008, 39(1): 66–68.

[37] Li Y, Shi X. Mine pressure prediction study based on fuzzy cognitive maps[J]. International Journal of Computational Intelligence and Applications, 2020, 19(3): 1–13.

[38] 乔懿麟, 黄克军, 耿耀强, 丁国峰. 采动影响下沿空留巷底板非对称性底鼓变形研究[J]. 煤炭技术, 2017, 36(7): 94–96.

[39] Qiao H, Xu H, Pang A, et al. A research on mine pressure monitoring data analysis and forecast expert system of fully mechanized coal face[C]// The 19th International Conference on Industrial Engineering and Engineering Management: Assistive Technology of Industrial Engineering. Springer Berlin Heidelberg, 2013: 925–934.

[40] 尹增德. 矿山压力监测与预报[M]. 北京: 煤炭工业出版社, 2011, 151–154.

[41] 屈世甲, 李鹏. 基于支架工作阻力大数据的工作面顶板矿压预测技术研究[J]. 矿业安全与环保, 2019, 46(2): 92–97.

[42] 贾澎涛, 苗云风. 基于堆叠LSTM的多源矿压预测模型分析[J]. 矿业研究与开发, 2021, 41(08): 79–82.

[43] 杨硕. 基于PSO-BP神经网络的浅埋煤层工作面顶板矿压预测研究[D]. 西安: 西安科技大学, 2010.

[44] 赵铭生, 刘守强, 纪润清, 刘德民, 曹欢, 胡棉舒. 基于遗传算法优化BP神经网络的华北型煤田矿压破坏带深度预测[J]. 矿业研究与开发, 2020, 40(06): 89–93.

[45] 巩师鑫, 任怀伟, 杜毅博, 赵国瑞, 文治国, 周杰. 基于MRDA-FLPEM集成算法的综采工作面矿压迁移预测[J]. 煤炭学报, 2021, 46(S1): 529–538.

[46] 曾庆田, 吕珍珍, 石永奎, 田广宇, 林泽东, 李超. 基于Prophet+LSTM模型的煤矿井下工作面矿压预测研究[J]. 煤炭科学技术, 2021, 49(7): 16–23.

[47] 李泽萌. 基于LSTM的采煤工作面矿压预测方法研究[D]. 西安: 西安科技大学, 2020.

[48] 王家臣, 许家林, 杨胜利, 王兆会. 煤矿采场岩层运动与控制研究进展——纪念钱鸣高院士“砌体梁”理论40年[J]. 煤炭科学技术, 2023, 51(1): 80–94.

[49] Zuo C, Qian J, Feng S, et al. Deep learning in optical metrology: a review[J]. Light: Science & Applications, 2022, 11(1): 39–92.

[50] Zekić-Sušac M, Has A, Knežević M. Predicting energy cost of public buildings by artificial neural networks, CART, and random forest[J]. Neurocomputing, 2021, 439: 223–233.

[51] LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278–2324.

[52] Tan Z, Pan P. Network fault prediction based on CNN-LSTM hybrid neural network[C]// 2019 International Conference on Communications, Information System and Computer Engineering (CISCE). IEEE, 2019: 486–490.

[53] Jain A, Zamir A R, Savarese S, et al. Structural-rnn: Deep learning on spatio-temporal graphs[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 5308–5317.

[54] 杨丽, 吴雨茜, 王俊丽, 等. 循环神经网络研究综述[J]. 计算机应用, 2018, 38(A02): 1–6.

[55] Chaudhari S, Mithal V, Polatkan G, et al. An attentive survey of attention models[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2021, 12(5): 1–32.

[56] Tay Y, Dehghani M, Bahri D, et al. Efficient transformers: A survey[J]. ACM Computing Surveys, 2022, 55(6): 1–28.

[57] Zerveas G, Jayaraman S, Patel D, et al. A transformer-based framework for multivariate time series representation learning[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 2021: 2114–2124.

[58] Taylor S J, Letham B. Forecasting at scale[J]. The American Statistician, 2018, 72(1): 37–45.

[59] Herzen J, Lässig F, Piazzetta S G, et al. Darts: User-friendly modern machine learning for time series[J]. The Journal of Machine Learning Research, 2022, 23(1): 5442–5447.

[60] Torres J F, Hadjout D, Sebaa A, et al. Deep learning for time series forecasting: a survey[J]. Big Data, 2021, 9(1): 3–21.

[61] Hyndman R J, Athanasopoulos G. Forecasting: principles and practice[M]. OTexts, 2018.

[62] XIONG R, YANG Y, HE D, et al. On layer normalization in the transformer architecture[C]//Proceedings of the 37th International Conference on Machine Learning. Vienna, Austria: PMLR, 2020: 10524–10533.

[63] Boullé N, Nakatsukasa Y, Townsend A. Rational neural networks[J]. Advances in Neural Information Processing Systems, 2020, 33: 14243–14253.

[64] 苏振国, 邓志刚, 李国营, 等. 顶板深孔爆破防治小煤柱冲击地压研究[J]. 矿业安全与环保, 2019, 46(4): 21–25.

[65] 文冰梅, 赵联文, 黄磊. AIC准则与留一法交叉验证渐近等价的证明[J]. 统计与决策, 2022, 38(6): 40–43.

[66] 秦玉金, 苏伟伟, 姜文忠, 等. 我国矿井瓦斯涌出量预测技术研究进展及发展方向[J]. 煤矿安全, 2020, 51(10): 52–59.

[67] Chaudhari S, Mithal V, Polatkan G, et al. An attentive survey of attention models[J]. ACM Transactions on Intelligent Systems and Technology , 2021, 12(5): 1–32.

[68] 黄飞虎, 赵红磊, 弋沛玉, 李沛东, 彭舰. 一种改进Transformer的电力负荷预测方法[J]. 现代电力, 2023, 40(1): 50–58.

[69] Torres J F, Hadjout D, Sebaa A, et al. Deep learning for time series forecasting: a survey[J]. Big Data, 2021, 9(1): 3–21.