摘 要

随着优质浅埋藏煤层的逐步开采以及中小煤矿的资源整合，我国煤矿开采开始呈现深度化和规模化趋势，进而因工作面矿压失衡而导致的矿压灾害频发，严重威胁着井下开采人员的生命安全。因此，在煤矿开采中对工作面矿压的预测对保证煤矿井下安全生产具有重要意义。论文以提高矿压预测的精度和效率并支撑工作面矿压监测预警工作为核心，基于深度学习开展对采煤工作面矿压预测方法的研究，具体内容如下：

为了解决矿压预测精度不高的问题，提出了基于IWOA-Highway-BiLSTM的矿压预测方法。该方法首先应用3准则和K近邻算法实现工作面矿压监测数据的预处理；然后构建基于双向长短时记忆网络结构提取矿压数据特征，使用Highway Networks作为隐藏层的层间连接减少模型训练时间；最后利用改进鲸鱼优化算法对模型超参进行寻优，实现工作面矿压的预测。实验结果表明：相较于SVR、RNN、LSTM和GRU模型，论文所构建的基于IWOA-Highway-BiLSTM模型在测试集上的RMSE降低了20.8%、16.7%、14.1%和5.2%；运行时间减少了39.8%、34.9%、38.1%和8.1%，取得了良好的预测效果。

为了解决循环神经网络及其变种结构在处理大规模数据集时性能降低的问题，提出了适用于数据量较大情况下的基于Attention-TCN的矿压预测方法。该方法通过构建基于时序卷积网络结构的模型提取数据特征，为每一个卷积顶端增加注意力层来强化关键特征，提高预测精度。实验结果表明：在数据量较大的数据集上，相较于RNN、LSTM、GRU以及IWOA-Highway-BiLSTM模型，Attention-TCN模型在测试集上的RMSE降低了34.1%、24.1%、17.9%和15.7%，运行时间减少了44.5%、50.4%、24.7%和27.0%，取得了比循环神经网络及其变体结构更好的预测效果。

为了支撑工作面矿压监测及灾害预警工作，通过系统分析、系统设计、功能实现及系统测试四个环节设计并实现了工作面矿压智能监测与预警系统。工作面现场应用结果表明：系统工作平稳，监测及预警效果良好，对于煤矿井下的安全生产具有较好的实际应用价值。

With the gradual exploitation of high-quality shallow buried coal seams and the resource integration of small and medium-sized coal mines, China's coal mining began to show a deep and large-scale trend. The mine pressure disasters caused by the imbalance of mine pressure in the working face occur frequently, which seriously threatens the life safety of underground mining personnel. Therefore, in coal mining, it is of great significance to predict the mine pressure of the working face to ensure the safe production of coal mines. The paper focuses on improving the accuracy and efficiency of mine pressure prediction and supporting the monitoring and early warning of mine pressure in the working face. Based on deep learning, the research on the prediction method of mine pressure in the working face is carried out. The specific work is as follows:

To solve the problem of low accuracy of mine pressure prediction, a mine pressure prediction method based on IWOA-Highway-BiLSTM is proposed.First, the Pauta criteria and k-nearest neighbor algorithm are used to preprocess the rock pressure monitoring data. Second, A bidirectional long-term and short-term memory network structure is constructed to extract the characteristics of mine pressure data, and highway networks is used as the hidden interlayer connection to reduce the training time of the mode. Finally, the improved whale algorithm is used to optimize the model parameters to realize the prediction of mine pressure data. The experimental results show that compared with SVR,RNN,LSTM and GRU,the prediction accuracy of IWOA-Highway-BiLSTM model on the test set is improved by 20.8%, 16.7%, 8.2% and 5.2% respectively; the running time is reduced by 39.8%, 34.9%, 38.1% and 8.1%,and achieved good prediction results.

To solve the problem of reduced performance of cyclic neural network and its variant structure in processing large-scale data sets, a ground pressure prediction method based on Attention-TCN is proposed, which is suitable for large amount of data.By constructing a model based on the structure of time-series convolution network, the data features are extracted, and an attention layer is added at the top of each convolution to strengthen the key features and improve the prediction accuracy. The experimental results show that on the large data set, compared with RNN, LSTM, GRU and IWOA-Highway-BiLSTM models, the RMSE of the Attention-TCN model on the test set is reduced by 34.1%, 24.1%, 17.9% and 15.7%; the running time is reduced by 44.5%, 50.4%, 24.7% and 27.0%, achieved better prediction performance than the recurrent neural network and its variant structure..

To support the mine pressure monitoring and disaster early warning of the working face,the intelligent monitoring and early warning system of mine pressure in working face is designed and realized through four links of system analysis, system design, function realization and system test. The field test results of the working face show that the system runs stably, with good monitoring and early warning effects, and has good practical application value for safe production in coal mines.

[1] 张莉,张建强,宁树正,赵欣,刘亢,王丹凤,黄少青,周洁,邹卓,孙汉英.中国与全球煤炭行业形势对比分析[J].中国煤炭地质,2021,33(S1):17-21+43.

[2] 李辉,庞博,朱法华,孙雪丽,徐静馨,王圣.碳减排背景下我国与世界主要能源消费国能源消费结构与模式对比[J/OL].环境科学:1-13[2022-04-09].DOI:10.13227/ j.hjkx.202112065.

[3] 李波,巨广刚,王珂,张路路,孙东辉.2005—2014年我国煤矿灾害事故特征及规律研究[J].矿业安全与环保,2016,43(03):111-114.

[4] 王超,谢佳玉,徐放艳,孔超.大采高工作面矿压规律自动分析与仿真研究[J].煤炭科学技术,2021,49(04):186-192.

[5] 陶志刚,任树林,庞仕辉,张海江,张斌,何满潮.无煤柱自成巷大数据物联网监测系统及工程应用[J].中国矿业大学学报,2021,50(05):983-991.

[6] 窦林名,王盛川,巩思园,蔡武,李小林.冲击矿压风险智能判识与监测预警云平台[J].煤炭学报,2020,45(06):2248-2255.

[7] 周波,张旺,周泽,李克城.近距离采空区下煤层开采矿压显现特征研究[J].煤矿安全,2021,52(10):217-223.DOI:10.13347/j.cnki.mkaq.2021.10.037.

[8] 庞军林.基于统计分析法的工作面来压预测研究[J].煤矿开采,2012,17(04):93-95.

[9] Jia Wang,Chen Liu,Tian Fu,Lili Zheng. Research on automatic target detection and recognition based on deep learning[J]. Journal of Visual Communication and Image Representation,2019,60:

[10] Hossain Md. Rajib,Hoque Mohammed Moshiul,Siddique Nazmul,Sarker Iqbal H.. Bengali text document categorization based on very deep convolution neural network[J]. Expert Systems With Applications,2021,184:

[11] Li Yachao,Li Junhui,Zhang Min. Deep Transformer modeling via grouping skip connection for neural machine translation[J]. Knowledge-Based Systems,2021,234:

[12] Gonzalez-Garcia A, Barragan-Alcantar D, Collado-Gonzalez I, et al. Adaptive dynamic programming and deep reinforcement learning for the control of an unmanned surface vehicle: Experimental results[J]. Control Engineering Practice, 2021, 111: 104807.

[13] 窦林名,周坤友,宋士康,曹安业,崔恒,巩思园,马小涛.煤矿冲击矿压机理、监测预警及防控技术研究[J].工程地质学报,2021,29(04):917-932.DOI:10.13544/ j.cnki.jeg.2021-0337.

[14] Dou L, Cai W, Cao A, et al. Comprehensive early warning of rock burst utilizing microseismic multi-parameter indices[J]. International Journal of Mining Science and Technology, 2018, 28(5): 767-774.

[15] 刘会田,姚精明.模糊综合评判法在矿井多种灾害预测中的应用[J].能源技术与管理,2008(06):22-24.

[16] 左凌云.基于支架工作阻力大数据的工作面区域矿压分析研究[J].煤炭工程,2019,51(11):60-64.

[17] 张通,赵毅鑫,朱广沛,王少磊,焦振华.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,2016,41(S2):287-296.

[18] 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.

[19] 孙书亮,王元杰,乔中栋,于贵良,马连兄.地音监测技术在冲击矿压预测预报中的应用[J].煤矿开采,2009,14(05):80-82.

[20] 王应启,马良,陈本华.钻屑法在监测冲击矿压中的研究与应用[J].煤矿现代化,2006(S1):153-154.

[21] Cai W, Dou L, Zhang M, et al. A fuzzy comprehensive evaluation methodology for rock burst forecasting using microseismic monitoring[J]. Tunnelling and Underground Space Technology,2018,80:232-245.

[22] Qiu L, Liu Z, Wang E, et al. Early-warning of rock burst in coal mine by low-frequency electromagnetic radiation[J]. Engineering Geology,2020,279:105755.

[23] 窦林名,蔡武,巩思园,韩荣军,刘军.冲击危险性动态预测的震动波CT技术研究[J].煤炭学报,2014,39(02):238-244.

[24] 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(02):364-370.

[25] 窦林名,冯龙飞,蔡武,王华,贺海鸿,焦彪,张敏.煤岩灾变破坏过程的声震前兆识别与综合预警模型研究[J].采矿与安全工程学报,2020,37(05):960-968+976.DOI:10.13545/ j.cnki.jmse.2020.05.012.

[26] Li N, Feng X, Jimenez R. Predicting rock burst hazard with incomplete data using Bayesian networks[J]. Tunnelling and Underground Space Technology,2017,61:61-70.

[27] 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(03):71-81.

[28] 孙刚,周华平,孙克雷.基于支持向量机的矿压监测数据预测研究[J].阜阳师范学院学报(自然科学版),2013,30(03):46-49.

[29] Li J, Yue J, Yang Y, et al. Multi-Resolution Feature Fusion model for coal rock burst hazard recognition based on Acoustic Emission data[J]. Measurement,2017,100:329-336.

[30] 吕文玉,丁科,伍永平,杜旭峰,胡馨,罗香玉,张勇.综采面支架工作阻力的PCA-SVR预测模型[J].西安科技大学学报,2021,41(06):973-978.

[31] Jia R S, Liu C, Sun H M, et al. A situation assessment method for rock burst based on multi-agent information fusion[J]. Computers & Electrical Engineering,2015,45:22-32.

[32] 常峰. 基于GA-BP神经网络的工作面顶板矿压预测模型应用研究[D].中国矿业大学,2019.

[33] 柴敬,王润沛,杜文刚,雷武林,朱旭宝.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(04):64-71.

[34] 刘晓悦,李朋园,周勇君.基于改进果蝇优化GRNN的冲击地压预测研究[J].矿业研究与开发,2017,37(10):106-110.

[35] 靳甜甜. 基于小波分析与ARMA组合模型的矿压显现预报研究[D].西安科技大学,2015.

[36] 程海星,朱磊,宋立平,刘文涛,徐凯.基于逆向传播神经网络的工作面顶板矿压数据预测[J].煤矿安全,2021,52(05):216-220.DOI:10.13347/j.cnki.mkaq.2021.05.038.

[37] 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(02):92-97.

[38] 赵毅鑫,杨志良,马斌杰,宋红华,杨东辉.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(01):54-65.

[39] 闫万梓. 基于多源信息的综采工作面智能来压预警模型研究[D].中国矿业大学,2021.DOI:10.27623/d.cnki.gzkyu.2021.000469.

[40] 金声尧,贾澎涛,王鹏,杨继红,侯长民.高应力围岩灾变信息采集及智能预警[J].矿业研究与开发,2021,41(06):170-175.DOI:10.13827/j.cnki.kyyk.2021.06.029.

[41] 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.DOI:10.13827/j.cnki.kyyk.2021.08.014.

[42] 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(07):16-23.

[43] 柴敬,刘义龙,王安义,屈世甲,欧阳一博.基于GRU和XGBoost的矿压显现规律预测[J].工矿自动化,2022,48(01):91-97.DOI:10.13272/j.issn.1671-251x.2021070062.

[44] 杨成虎.监测监控技术在煤矿安全生产中的应用[J].矿业装备,2021(05):42-43.

[45] Hu M, Tsang E C C, Guo Y, et al. Attribute reduction based on overlap degree and k-nearest-neighbor rough sets in decision information systems[J]. Information Sciences,2022,584:301-324.

[46] Xia J, Zhang J, Wang Y, et al. WC-KNNG-PC: Watershed clustering based on k-nearest-neighbor graph and Pauta Criterion[J]. Pattern Recognition,2022,121:108177.

[47] Santos, M. S., Abreu, P. H., Wilk, S. How distance metrics influence missing data imputation with k-nearest neighbours - ScienceDirect[J]. Pattern Recognition Letters,2020,136:111-119.

[48] 冀汶莉,郗刘涛,王斌.面向不平衡数据集的煤矿监测系统异常数据识别方法[J].工矿自动化,2020,46(01):18-25.

[49] 谭笑枫,李广帅. 深度学习发展综述[C]//国家安全地球物理丛书（十五）——丝路环境与地球物理.,2019:252-260.DOI:10.26914/c.cnkihy.2019.026363.

[50] Dokuz Y, Tufekci Z. Mini-batch sample selection strategies for deep learning based speech recognition[J]. Applied Acoustics,2021,171:107573.

[51] Chotirat S, Meesad P. Part-of-Speech tagging enhancement to natural language processing for Thai wh-question classification with deep learning[J]. Heliyon,2021, 7(10):e08216.

[52] Xu Z, Ma W, Lin P, et al. Deep learning of rock images for intelligent lithology identification[J]. Computers & Geosciences,2021,154:104799.

[53] 杨丽,吴雨茜,王俊丽,刘义理.循环神经网络研究综述[J].计算机应用,2018,38(S2):1-6+26.

[54] Yagmur Gizem Cinar,Hamid Mirisaee,Parantapa Goswami,Eric Gaussier,Ali Aït-Bachir. Period-aware content attention RNNs for time series forecasting with missing values[J]. Neurocomputing,2018,312:

[55] Hochreiter, Sepp, and J. Schmidhuber. Long Short-Term Memory[J].Neural Computation,1997,9(8):1735-1780.

[56] Wei Yunyu,Chen Zezong,Zhao Chen,Tu Yuanhui,Chen Xi,Yang Rui. A BiLSTM hybrid model for ship roll multi-step forecasting based on decomposition and hyperparameter optimization[J]. Ocean Engineering,2021,242:

[57] Cho, Kyunghyun, van Merrienboer, Bart, Gulcehre, Caglar. Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation[J]. Computer science, 2014.

[58] Wang Xiaodi,Tang Mingwei,Yang Tian,Wang Zhen. A novel network with multiple attention mechanisms for aspect-level sentiment analysis[J]. Knowledge-Based Systems,2021,227:

[59] Wang Peng,Si Fengqi,Cao Yue,Shao Zhuang,Ren Shaojun. Prediction of superheated steam temperature for thermal power plants using a novel integrated method based on the hybrid model and attention mechanism[J]. Applied Thermal Engineering,2022,203:

[60] Lin Yi,Zhou Jingchun,Ren Wenqi,Zhang Weishi. Autonomous underwater robot for underwater image enhancement via multi-scale deformable convolution network with attention mechanism[J]. Computers and Electronics in Agriculture,2021,191:

[61] Guo Yanbu,Zhou Dongming,Li Weihua,Cao Jinde,Nie Rencan,Xiong Lei,Ruan Xiaoli. Identifying polyadenylation signals with biological embedding via self-attentive gated convolutional highway networks[J]. Applied Soft Computing Journal,2021,103:

[62] Cheridito Patrick,Jentzen Arnulf,Rossmannek Florian. Non-convergence of stochastic gradient descent in the training of deep neural networks[J]. Journal of Complexity, 2020(prepublish):

[63] 郭启程,杜晓玉,张延宇,周毅.基于改进鲸鱼算法的无人机三维路径规划[J].计算机科学,2021,48(12):304-311.

[64] 吴张倩,苏兆品,武钦芳,张国富.实际噪声下基于时序卷积网络的手机来源识别[J].计算机工程与科学,2021,43(08):1461-1469.