瓦斯灾害长期以来一直是阻碍我国煤矿安全生产的主要灾害。所以出于控制瓦斯事故的目的，需要针对煤矿瓦斯浓度进行预测预警。然而，结合目前的瓦斯浓度预测预警技术的应用情况来看，存在两方面的不足之处，首先是预测的精准度差，其次是预警分析不足，正是由于这些问题的存在，使得现有的预测预警技术不能满足煤矿安全生产实际需求。针对上述问题，本文首先基于随机森林算法对预处理后的瓦斯浓度时间序列进行特征选择，从而确定具有强解释性的工作面瓦斯浓度时间序列特征集合。其次以特征选择后集合为研究对象，构建基于CNN-GRU的工作面瓦斯浓度预测模型。最后提出工作面瓦斯浓度动态预警体系，以预测值作为预警数据来源，采用粒子群优化支持向量机模型实现工作面瓦斯浓度智能动态预警，并对所提出模型在云平台上进行封装和原型系统开发。主要的研究内容如下：

（1）由于瓦斯浓度受多种因素影响，并且监测数据含噪较高，各影响因素之间具有复杂的非线性关系，采用单因素进行预测不能有效的反应矿井真实环境情况。因此本文首先运用临近均值法和三次平滑指数法实现各类数据的科学预处理工作，然后提出基于随机森林的瓦斯浓度时间序列特征选择，从众多影响变量中筛选出关键影响因素，从而得到高质量数据为工作面瓦斯浓度模型训练和预测奠定基础。

（2）针对瓦斯浓度预测精度低的问题，基于集成学习的思想，本文创新性地提出运用卷积神经网络（CNN）将瓦斯浓度及其影响因素有效的耦合起来，然后应用门控循环单元神经网络（GRU）在关联特征的基础上进行时序性预测的组合模型。并以玉华煤矿工作面监测数据作为实验样本，通过调整批尺寸、神经元和优化函数等参数提高模型的综合性能。同时与其它传统的深度学习模型进行对比，CNN-GRU模型具有更快的收敛速度和更高的准确率。

（3）针对瓦斯浓度预警分析不足的问题。在瓦斯浓度预测的基础上，研究确定预警指标，通过瓦斯浓度监测数据的统计分析确定预警阈值，将预测值与预警指标阈值进行比较实现瓦斯浓度异常分析，划分预警等级。最后以瓦斯浓度预测值为数据来源，结合预警等级采用粒子群算法优化支持向量机的深度学习模型实现工作面瓦斯浓度智能动态分级预警。

（4）实现采煤工作面瓦斯浓度智能动态预警与可视化，实现了采煤工作面瓦斯浓度动态预警前端界面和后台数据结构的联合使用，采用云平台模式完成算法集成、模型封装及开发部署的工作，为采煤工作面瓦斯浓度动态预警提供软件模型及技术支持。

基于以上研究成果进行了现场试验验证，形成了一种准确高效的采煤工作面瓦斯浓度动态预警方法，为采煤工作面瓦斯浓度态势评估提供了有力依据。

At this stage, there is a hidden danger of disaster in the process of coal mine production in China, that is, gas disaster. Therefore, for the purpose of controlling gas accidents, it is necessary to predict and warn the gas concentration in coal mines. However, combined with the current application of gas disaster early warning technology, there are two deficiencies: the first is the poor accuracy of early warning, and the second is the inadequate early warning analysis. It is precisely because of these problems that coal mine safety production does not meet the actual needs. To solve the above questions, this paper firstly selects the features of the pretreated gas concentration time series based on the random forest algorithm, so as to determine the highly explanatory feature set of the working face gas concentration time series. Secondly, the prediction model of gas concentration in working face based on CNN-GRU was constructed by taking the post-feature selection set as the research object. Finally, a dynamic warning system of working face gas concentration was proposed, and the predicted value was used as the warning data source. Particle swarm optimization support vector machine model was used to realize intelligent dynamic warning of working face gas concentration, and the proposed model was encapsulated and prototype system was developed on cloud platform.

(1) Because the gas concentration is affected by many factors, and the noise content of the monitoring data is high, andthis paper first uses the near mean method and Lagrange interpolation method to realize the scientific preprocessing of all kinds of data, and then proposes the feature selection of gas concentration time series based on random forest to screen the key influencing factors from many influencing variables, so as to obtain high-quality data, which lays a foundation for the training and prediction of gas concentration model in the working face.

(2) According to gas concentration prediction efficiency and low accuracy problem, this paper puts forward innovative use of convolution neural network (CNN) will be effective coupled gas concentration and its influencing factors, then apply gating cycle unit (GRU helped) neural networks on the basis of the correlation characteristics of the combination of chronological prediction model based on the thought of integrated study. With the monitoring data of Yuhua Coal mine working face as experimental samples, the comprehensive performance of the model is improved by adjusting the batch size, neurons and optimization function. Compared with other traditional deep learning models, the CNN-GRU model has faster convergence speed and smaller error.

(3) Aiming at the problem of insufficient gas concentration early warning analysis. Based on what I have learned, the author scientifically analyzes the early warning measures centered on the prediction of gas concentration, classifies the early warning indicators in combination with the actual situation, specifically including static indicators and dynamic indicators, makes scientific statistics and analysis in combination with the real-time data of gas concentration monitoring, defines the basic index threshold, and compares the value with the predicted value, so as to analyze the abnormal situation of gas concentration, According to the different results, several early warning levels are defined. Finally, the intelligent dynamic and refined early warning analysis of gas concentration in working face is realized by using the method of deep learning particle swarm optimization to optimize the classification of support vector machine.

(4) Realize the intelligent dynamic early warning and visual display of gas concentration in coal mining face, realize the joint use of front-end interface and background data structure of gas concentration dynamic early warning in coal mining face, complete algorithm integration, model packaging, development and deployment with the help of cloud platform, and provide software model and technical support for gas concentration dynamic early warning in coal mining face.

According to the above research results, a kind of accurate and efficient dynamic gas concentration warning technology in coal mining face is formed, which provides a strong basis for gas concentration situation assessment in coal mining face.

[1] 武强,涂坤,曾一凡,等.打造我国主体能源(煤炭)升级版面临的主要问题与对策探讨[J]. 煤炭学报, 2019, 44(6):12.

[2] 2021煤炭行业发展年度报告[R]. 煤炭工业协会, 2022.

[3] 张宏. 煤炭行业"十四五"发展面临的形势与任务[J]. 企业观察家, 2020(11):56-60.

[4] 李树刚, 常心坦, 林海飞. 西北矿区安全生产存在问题及对策研究[J]. 采矿与安全工程学报, 2006, 23(1):52-55.

[5] 袁亮，杨科. 再论废弃矿井利用面临的科学问题与对策[J]. 煤炭学报，2021，46(01): 16-24.

[6] 袁亮.煤及共伴生资源精准开采科学问题与对策[J]. 煤炭学报，2019，44(01): 1-9.

[7] 王双明.对我国煤炭主体能源地位与绿色开采的思考[J].中国煤炭,2020,46(02):11-16.

[8] 袁建平. 煤矿智能化开采技术的创新与管理[J]. 煤矿机电, 2016(03):77-79.

[9] 陈浩. 当代煤矿智能化综采技术发展与策略浅析[J]. 矿业装备, 2019(03):118-119.

[10] 牛莉霞, 李肖萌. 5G时代智慧矿山安全管理新模式[J]. 中国安全科学学报, 2021, 31(6):29-36.

[11] 王虹桥. "互联网+"背景下煤炭工业两化深度融合的思考[J]. 煤炭经济研究, 2015, 35(10):6-11.

[12] 黄晓军. 数学算法在人工智能中的应用[J]. 电子技术(上海), 2021(3):64-65.

[13] 葛路军,庞叶青.二次优化的灰色马尔科夫模型预测煤矿工作面瓦斯浓度[J].煤,2017,26(11):19-22+25.

[14] HAN Tingting, WU Shiyue, WANG Pengjun. Prediction of Gas Concentration Based on Markov Residual Correction[J]. Industrial and Mining Automation, 2014, 40(3):28-31.

[15] 武艳蒙,邱春荣,吕晓波. 基于模糊信息粒化与马尔科夫修正的瓦斯浓度预测[J].煤炭技术,2018,37(05):173-175.

[16] 吕品,马云歌,周心权. 上隅角瓦斯浓度动态预测模型的研究及应用[J].煤炭学报,2006, 31(4): 462-465.

[17] Qu Zhiming, Liang Xiaoying. Application of grey relation clustering and CGNN in gas concentration prediction in top corner of coal mine[A]. 2009 International Conference on Computational Intelligence and Natural Computing[C]. 2009.

[18] 段霄鹏. 基于商空间的煤矿瓦斯数据挖掘研究[D].太原理工大学,2010.

[19] 付华,代巍.基于ACPSO的PSR-MK-LSSVM瓦斯浓度动态预测方法[J].传感技术报,2016,29(06):903-908.

[20] 付华,丰盛成,刘晶,唐博.基于DE-EDA-SVM的瓦斯浓度预测建模仿真研究[J].传感技术学报,2016,29(02):285-289.

[21] 魏林,白天亮,付华,尹玉萍.基于EMD-LSSVM的瓦斯浓度动态预测模型[J].安全与环境学报,2016,16(02):119-123.

[22] 付华,訾海,孟祥云,孙璐.一种EKF-WLS-SVR与混沌时间序列分析的瓦斯动态预测新方法[J].传感技术学报,2015,28(01):126-131.

[23] 李欢,贾佳,杨秀宇,宋春儒. 煤矿综采工作面瓦斯浓度预测模型[J].工矿自动化,2018, 44(12): 48-53.

[24] 白峻尧. 基于SVM的瓦斯涌出的自学习模型研究应用[D].西安科技大学，2006.

[25] 高莉,胡延军,于洪珍. 基于W­RBF的瓦斯时间序列预测方法[D]. 煤炭学报，2008, 33(1): 68-70.

[26] Gao Li, Yu Hongzhen. Prediction of gas emission based on information fusion and chaotic time series [J]. Journal of China University of Mining & Technology, 2006, 16(1): 94-96.

[27] 姚胥华,邱本花. 基于改进BP神经网络的矿井瓦斯浓度预测算法[J].煤炭技术,2017, 36(05): 182-184.

[28]李树刚,马莉,潘少波,石新莉.基于循环神经网络的煤矿工作面瓦斯浓度预测模型研究[J].煤炭科学技术,2020,48(01):33-38.

[29] 崔小彦,赵国强.混沌预测在井下瓦斯浓度监测的应用[J].辽宁工程技术大学学报(自然科学版),2008(S1):184-186.

[30] 程健,白静宜,钱建生,李世银.基于混沌时间序列的煤矿瓦斯浓度短期预测[J].中国矿业大学学报,2008(02):231-235.

[31] 赵金宪,于光华.瓦斯浓度预测的混沌时序RBF神经网络模型[J].黑龙江科技学院学报,2010,20(02):131-134.

[32] Zhao Jinxian, Jin Hongzhang, Yu Guanghua. Coalmine gas concentration forecasting based on chaotic theory and neural network model[A]. Proceedings of the Third International Conference on Information and Computing Science[C]. 2010.

[33] Liu Xiaosheng, Huang Xiaobin. Research on pre­warning expert system of coal mine gas safety based on object­oriented[A]. 2008 IEEE International Symposium on Knowledge Acquisition and Modeling Workshop[C]. 2008.

[34] 熊廷伟. 煤矿瓦斯爆炸预警技术研究[D].重庆大学,2005.

[35] 李华. 矿井掘进瓦斯爆炸实时智能预警监控系统[D].西安科技大学,2005.

[36] 牛聚粉. 基于MapX的煤与瓦斯突出预警技术研究[D].中国地质大学（北京）,2009.

[37] 付贵祥,周红军,郭继茹.基于物联网的煤矿安全综合智能预警系统[J].工矿自动化,2014,40(04):99-101.

[38] Xie Guomin, Fu Hua, Wang Shiyun. Study on early warning of gas in coal mine based on extension data mining[A]. 2010 2nd International Conference on Education Technology and Computer[C]. 2010.

[39]Song Jinxing, Liu Yufang. Study on the mine gas disaster prevention system by GIS[A].2010 2nd Conference on Environmental Science and Information Application Technology[C]. 2010.

[40] 黄为勇. 基于支持向量机数据融合的矿井瓦斯预警技术研究[D].中国矿业大学,2009.

[41] 李骥平. 商空间理论在煤矿预警中的应用[D].太原理工大学,2010.

[42] 邵良杉,付贵祥.基于数据融合理论的煤矿瓦斯动态预测技术[J].煤炭学报,2008(05):551-555.

[43] HINTON G E,SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science,2006,313(5786):504-507.

[44] Yu Z , Hong Z , Lingdong K . Research of Coal-Gas Outburst Forecasting Based on Artificial Immune Network Clustering Model.[J]. IEEE, 2009:23-27.

[45] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[C].Advances in Neural Information Processing Systems,2012:1097-1105.

[46] Ruilin Z , Lowndes I S . The application of a coupled artificial neural network and fault tree analysis model to predict coal and gas outbursts[J]. International Journal of Coal Geology, 2010, 84(2):141-152.

[47]Beringer N, Graves A , Schiel F ,et al. Classifying Unprompted Speech by Retraining LSTM Nets.[J]. Lecture Notes in Computer Science, 2005, 58(1956):575-581.

[48] Erfani S M , Rajasegarar S , Karunasekera S , et al. High-dimensional and large-scale anomaly detection using a linear one-class SVM with deep learning.[J]. Pattern Recognition, 2016, 58(C):121-134.

[49]Wang J , Zhang J , Wang X . Bilateral LSTM: A Two-Dimensional Long Short-Term Memory Model With Multiply Memory Units for Short-Term Cycle Time Forecasting in Re-entrant Manufacturing Systems[J]. IEEE Transactions on Industrial Informatics, 2018, 14(2):748-758.

[50] 郭丽丽,丁世飞.深度学习研究进展[J].计算机科学,2015,42(05):28-33.

[51] 缪燕子.多传感器信息融合理论及在矿井瓦斯突出预警系统中的应用研究[D].中国矿业大学,2009.

[52] 马小平,胡延军,缪燕子.物联网、大数据及云计算技术在煤矿安全生产中的应用研究[J].工矿自动化,2014,40(04):5-9.

[53] 缪燕子,方健,马小平.多传感器信息融合在水环境监测中的应用[J].西安科技大学学报,2006(02):212-214+244.

[54] Chuanqu Zhu.Mechanical analysis on deformation of surrounding rock with road-in packing of gob-side entry retaining in fully-mechanized sub-level caving face[J].Journal of Coal Science & Engineering(China),2008,14(01):24-28.

[55] 朱川曲.基于神经网络的综采工作面人-机-环境系统的可靠性研究[J].煤炭学报,2000(03):268-272.

[56]朱川曲.采煤工作面瓦斯涌出量预测的神经网络模型[J].中国安全科学学报,1999(02):45-48.

[57] 刘晓悦,李朋园.基于PCA和IFOA-BP煤与瓦斯突出软测量建模[J].矿业研究与开发,2018,38(04):109-115.

[58] 闫乐林,徐精彩,许满贵,等.未确知数学在煤矿安全预评价中的应用[J].矿业安全与环保,2004(02):24-26+1

[59] 许满贵.煤矿动态综合安全评价模式及应用研究[D].西安科技大学,2006.

[60] 施式亮.矿井安全非线性动力学评价模型及应用研究[D].中南大学,2000.

[61] 施式亮,伍爱友.GM(1,1)模型与线性回归组合方法在矿井瓦斯涌出量预测中的应用[J].煤炭学报,2008(04):415-418.

[62] 念其锋,施式亮,李润求,等.基于PNN的煤矿安全生产风险综合预警研究[J].中国安全生产科学技术,2013,9(10):71-77.

[63] 何利文,施式亮,宋译,等.基于支持向量机(SVM)的回采工作面瓦斯涌出混沌预测方法研究[J].中国安全科学学报,2009,19(09):42-46+180.

[64] 伍爱友,施式亮,王从陆.基于神经网络和遗传算法的城市火灾风险评价模型[J].中国安全科学学报,2006(11):108-113.

[65] 伍爱友,田云丽,宋译,等.灰色系统理论在矿井瓦斯涌出量预测中的应用[J].煤炭学报,2005(05):47-50.

[66] Soutner D , Müller, Luděk. Application of LSTM neural networks in language modelling[J]. Lecture Notes in Computer Science, 2013, 8082:105-112.

[67] WORDEN K,STASZEWSKI WJ,HENSMAN J J. Natural computing for mechanical systems research:A tutorial overview[J]. Mechanical Systems and Signal Processing,2011,25(1):4-111.

[68] Read J ,Perez-Cruz F. Deep Learning for Multi-label Classification[J]. Machine Learning,2014,85(3):333-359.

[69] Li H,Ota K,Dong M .Learning IoT in Edge: Deep Learning for the Internet of Things with Edge Computing[J]. IEEE Network, 2018,32(1):96-101.

[70] 毛勇华,桂小林,李前,等.深度学习应用技术研究[J].计算机应用研究,2016,33(11):3201-3205.

[71] S.Dharumarajan,Thomas F.A.Bishop.Desertification status mapping in Muttuma Watershed by using Random Forest Model[J].Sciences in Cold and Arid Regions,2022,14(01):32-42.

[72]Yannan Zhao,Yuan Li,Lifen Zhang,Qiuliang Wang.Groundwater level prediction of landslide based on classification and regression tree[J].Geodesy and Geodynamics,2016,7(05):348-355.

[73] 张贺,蔡江辉,张继福,乔衎.信息熵度量的离群数据挖掘算法[J].智能系统学报,2010,5(02):150-155.

[74] Strobl C, Boulesteix A L ,Zeileis A, et al. Bias in random forest variable importance measures: Illustrations, sources and a solution[J]. Bmc Bioinformatics, 2007, 8.

[75] Menze B H , Kelm M B , Masuch R , et al. A comparison of random forest and its Gini importance with standard chemometric methods for the feature selection and classification of spectral data[J]. Bmc Bioinformatics, 2009, 10(1):1-16.

[76] Al-Abed M A , Al-Bashir A K , Al-Rawashdeh A , et al. Estimation of cerebral blood flow velocity during breath-hold challenge using artificial neural networks[J]. Computers in Biology and Medicine, 2019, 115(5):115-128.

[77] HINTON G,OSINDERO S,TEH Y．A fast learning algorithm for deepbelief nets[J]．Neural Computation,2006,18(7):1527-1554

[78] Schmidhuber J ,Gers F , Eck D . Learning Nonregular Languages: A Comparison of Simple Recurrent Networks and LSTM[J]. Neural Computation, 2014, 14(9):2039-2041.

[79] Monner D,Reggia J A . A generalized LSTM-like training algorithm for second-order recurrent neural networks[J]. Neural Networks, 2012, 25(none):70-83.

[80] ZHENG H，YUAN J，CHEN L. Short-Term Load Forecasting Using EMD-LSTM Neural Networks with a Xgboost Algorithm for Feature Importance Evaluation[J]. Energies, 2017, 10(8).

[81] Suk H I , Lee S W , Shen D . Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis[J]. NeuroImage, 2014, 101:569-582.

[82] CHEONEUM P. Korean Coreference Resolution with Guided Mention Pair Model Using Deep Learning[J]. ETRI Journal, 2016, 38(6).

[83] 宋厚岩,王汉军.基于GRU和PCNN的电力知识抽取[J].计算机系统应用,2021,30(09):200-205.

[84] Lecun Y, Bottou L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11):2278-2324.

[85] Yandong L I, Hao Z, Lei H. Survey of convolutional neural network[J]. Journal of Computer Applications, 2016.

[86] Keskes H , Braham A . Recursive Undecimated Wavelet Packet Transform and DAG SVM for Induction Motor Diagnosis[J]. IEEE Transactions on Industrial Informatics, 2015, 11(5):1059-1066.

[87] M Sundermeyer,R Schlüter,H Ney. LSTM Neural Networks for Language Modeling.Interspeech[J],2012,31(43):601-608.

[88] 汪海燕,黎建辉,杨风雷.支持向量机理论及算法研究综述[J].计算机应用研究,2014,31(05):1281-1286.

[89] 王霞,董永权,于巧,耿娜.结构化支持向量机研究综述[J].计算机工程与应用,2020,56(17):24-32.

[90] 赵毅. 粒子群优化算法的改进和应用研究[D].沈阳工业大学,2019.

[91] 汤可宗. 遗传算法与粒子群优化算法的改进及应用研究[D].南京理工大学,2011.

[92] 秦媛. 粒子群算法改进及其应用研究[D].南京邮电大学,2018.

[93] Erdogmus P . Particle Swarm Optimization with Applications[M]. 2018.

[94] 董方. 粒子群算法研究及其在动态优化中的应用[D].浙江大学,2014.

[95]陈功,苗瑞,张洁.基于PSO-SVM的带钢表面缺陷分类研究[J].内蒙古大学学报(自然科学版),2015,46(04):435-441.

[96] 林升梁,刘志.基于RBF核函数的支持向量机参数选择[J].浙江工业大学学报,2007(02):163-167.

[97] 张洵. 粒子群优化算法的改进研究[D].渤海大学,2017.