瓦斯灾害是煤矿领域的重大安全问题。随着煤矿智能化水平的不断提高，瓦斯浓度预测预警技术对煤矿灾害的防治起着至关重要的作用。充分利用煤矿井下大量的瓦斯数据进行瓦斯浓度预测预警，能有效提高瓦斯灾害的预警能力。因此，准确、可靠的瓦斯浓度预测预警对煤矿安全生产有着重大意义。本文以陕西某矿综采面瓦斯监测数据为研究对象，进行了瓦斯浓度预测及预警方法的研究。

本文为提高综采面瓦斯浓度预测精度和预警效率，将深度学习中的预测模型与分布式处理框架相结合，搭建了基于Spark Streaming的瓦斯浓度预警框架。首先对影响工作面瓦斯浓度的因素进行分析，并利用灰色关联度法进行瓦斯浓度影响因素的验证，最终选取7种瓦斯浓度影响因素构建瓦斯浓度影响因素体系，同时基于矿井瓦斯实测数据，建立单一的门控循环单元（GRU）预测模型，利用主成分分析方法对GRU模型进行数据降维以提高模型的预测精度，并从预测模型的隐含层神经元个数、隐含层数、批大小和时间步长四个方面进行参数优化。其次，为提高单一GRU模型的预测精度，采用遗传算法（GA）和粒子群算法（PSO）进行模型优化，分别建立PCA-GA-GRU模型和PCA-PSO-GRU模型，经过对比分析后得出 PCA-PSO-GRU模型的平均绝对误差为0.0121，均方根误差为0.0159，模型拟合度为0.974，模型训练时间为65s，该模型预测精度高、预测效果好。最后，在瓦斯浓度监测数据的基础上，利用统计分析确定瓦斯浓度的预警阈值，通过预测值与实测值的对比进行瓦斯浓度异常值判定和预警等级的划分，并将PCA-PSO-GRU预测模型与kafka系统和RDD数据集相结合，建立基于Spark Streaming瓦斯浓度预警框架，利用瓦斯浓度实测数据进行框架预警效率的验证。

实验结果表明：基于Spark Streaming瓦斯浓度预警框架的预警准确率在90%以上，整个预警处理时间为7s左右。该框架极大提高了瓦斯浓度预警速度，为矿井安全生产和瓦斯灾害防治提供了支持和决策。

Gas disaster was a major safety issue in the coal mine field. With the continuous improvement of the level of coal mine intelligence, gas concentration prediction and pre-warning technology plays a vital role in the prevention and control of coal mine disasters. Making full use of a large amount of gas data underground in coal mines for gas concentration prediction and pre-warning can effectively improve the pre-warning capabilities of gas disasters. Therefore, accurate and reliable gas concentration prediction and pre-warning are of great significance to coal mine safety production. This paper takes the gas monitoring data of a fully mechanized mining face in a mine in Shaanxi as the research object, and conducts the research on gas concentration prediction and pre-warning methods.

In order to improve the prediction accuracy and pre-warning efficiency of gas concentration in fully mechanized mining face, this paper combines the prediction model in deep learning with the distributed processing framework, and builds a pre-warning framework of gas concentration based on Spark Streaming. Firstly, the factors affecting the gas concentration in the working face are analyzed, and the influencing factors of the gas concentration are verified by the grey relational degree method, and seven influencing factors of the gas concentration are selected to construct the influencing factor system of the gas concentration. Based on mine gas measured data at the same time, to create a single gated recurrent unit model (GRU), using the method of principal component analysis and the model for data dimension reduction GRU helped to improve the predictive accuracy of the model and the model for prediction of hidden layer neurons number, hidden layer, batch size and time step four aspects for parameter optimization. Secondly, in order to improve the prediction accuracy of the single GRU model, genetic algorithm (GA) and particle swarm optimization algorithm (PSO) were used to optimize the model. The PCA-GA-GRU model and PCA-PSO-GRU model were established respectively. After comparative analysis, the average absolute error of the PCA-PSO-GRU model was 0.0121. The root mean square error is 0.0159, the model fitting degree is 0.974, and the model training time is 65s. The model has high prediction accuracy and good prediction effect. Finally, based on the monitoring data of gas concentration, the pre-warning threshold of gas concentration was determined by statistical analysis, and the abnormal value of gas concentration was judged and the early warning level was divided by comparing the predicted value with the measured value, and the PCA-PSO GRU prediction model was combined with Kafka system and RDD data set. The pre-warning framework of gas concentration based on Spark Streaming was established, and the pre-warning efficiency of the framework was verified by the measured data of gas concentration.

The experimental results show that the warning accuracy rate based on the Spark Streaming gas concentration warning framework is above 90%, and the whole pre-warning processing time is about 7s. This framework has greatly improved the speed of early warning of gas concentration, and provided support and decision-making for mine gas safety production and disaster prevention.

[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[J]．煤炭学报，2019，44(6)：1625－1636．

[2]煤炭是保障国民经济持续较快发展的重要能源支撑——煤炭工业壮丽70年综合评述[J]．中国煤炭，2019，45(10)：5-9．

[3]齐翊希．煤炭经济发展态势与对策分析[J]．现代工业经济和信息化，2019，9(07)：7-8．

[4]Anna Manowska, Michał Mazurek. Prospects for development and hard coal economy limitations in the context of ensuring national energy security[J]. IOP Conference Series: Earth and Environmental Science, 2018, 198(1).

[5]张镭．煤矿瓦斯防治技术现状与问题[J]．中国科技信息，2020(17)：105-106．

[6]Yiwen Zhang, Haishuai Guo, Zhihui Lu, et al. Distributed gas concentration prediction with intelligent edge devices in coal mine[J]. Engineering Applications of Artificial Intelligence, 2020, 92.

[7]关旭．煤矿智能化开采技术的创新与管理[J]．内蒙古煤炭经济，2020(14)：86-87．

[8]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，2019，9(03)：118-119．

[9]符大利．基于数据分析的无人化工作面生产模式研究[J]．煤矿安全，2017，48(05)：214-216．

[10]杨良斌．数据挖掘领域研究现状与趋势的可视化分析[J]．图书情报工作，2015，60(S2)：142-147．

[11]K. S. Ramshree. A Survey of Data Mining Techniques for Social Network Analysis[J]. Software Engineering, 2017, 9(1).

[12]邱宜萍，孙永红．“数字煤矿安全”广域网络在煤矿监察中的应用[J]．煤矿安全，2004，30(12)：66．

[13]谭章禄，陈晓．煤炭企业信息化建设现状及发展对策[J]．工矿自动化，2016，42(07)：63-65．

[14]方巍，郑玉，徐江．大数据：概念、技术及应用研究综述[J]．南京信息工程大学学报(自然科学版)，2014，6(05)：405-419．

[15]郑学召，童鑫，郭军，等．煤矿智能监测与预警技术研究现状与发展趋势[J]．工矿自动化，2020，46(06)：35-40．

[16]Xiaojie Zhu, Ze Jiang, Xiaobing Zhao, et al. A Novel Fuzzy Fusion Algorithm of Multi-sensor Data and Its Application in Coalmine Gas Monitoring[J]. IIETA, 2019, 18(6).

[17]樊雅琴，王炳皓，王伟，等．深度学习国内研究综述[J]．中国远程教育，2015(06)：27-33+79．

[18]荀晓玉，苏璨，李威，等．基于CNN-LSTM的煤矿瓦斯浓度预测[J]．现代信息科技，2020，4(20)：149-152．

[19]Mário P. éstias. Survey of Convolutional Neural Networks on Edge with Reconfigurable Computing. 2019, 12(8).

[20]Debnath Bhattacharyya, N. Thirupathi Rao. A Detailed Review on The Methods to Process an Application in Natural Language Processing[J]. International Journal of Computer Science and Information Technology for Education, 2019, 4(2).

[21]Bo Peng, Yang-sheng Jiang, Yun Pu. Review on Automatic Pavement Crack Image Recognition Algorithms. 015, 9(2): 3-20.

[22]包武杰，黄浩．基于语音增强方法的语音端点检测[J]．现代电子技术，2017，40(22)：1-4+9．

[23]徐磊，李希建．基于大数据的矿井灾害预警模型[J]．煤矿安全，2018，49(03)：98-101．

[24]李树刚，马莉，潘少波，等．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.

[25]Zakaria Alameer, Ahmed Fathalla, Kenli Li, et al. Multistep-ahead forecasting of coal prices using a hybrid deep learning model[J]. Resources Policy, 2020, 65.

[26]Dongjun Li, Zhenxin Zhang, Zhihua Xu, et al. Image-based Hierarchical Deep Learning Framework for Coal and Gangue Detection[J]. Journal of Engineering, 2020.

[27]张雨．基于深度学习的井下瓦斯浓度预测系统设计与实现[D]．徐州：中国矿业大学，2019．

[28]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：西安科技大学学报，2018，38(06)：1027-1035．

[29]宋爽．基于深度学习的采空区卸压瓦斯抽采智能评价方法研究[D]．西安科技大学，2019．

[30]智宁，毛善君，李梅，等．基于深度融合网络的煤矿图像尘雾清晰化算法[J]．煤炭学报，2019，44(02)：655-666．

[31]王书芹．基于深度学习的瓦斯时间序列预测与异常检测[D]．北京：中国矿业大学，2018．

[32]Guoxin Zhang, Zengcai Wang, Lei Zhao, et al. Coal-Rock Recognition in Top Coal Caving Using Bimodal Deep Learning and Hilbert-Huang Transform[J]. Shock and Vibration, 2017, 2017.

[33]Dominik Ślęzak, Marek Grzegorowski, Andrzej Janusz, et al. A framework for learning and embedding multi-sensor forecasting models into a decision support system: A case study of methane concentration in coal mines[J]. Information Sciences, 2018, 451-452.

[34]姚青华，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．

[35]Pingyang Lyu, Ning Chen, Shanjun Mao. LSTM based encoder-decoder for short-term predictions of gas concentration using multi-sensor fusion[J]. Process Safety and Environmental Protection, 2020, 137(04): 153-156.

[36]钱建生，邱春荣，李紫阳，等．深度学习耦合粒子群优化SVM的瓦斯浓度预测[J]．煤矿安全，2016，47(11)：173-176．

[37]吴兵，郭志国，王紫薇．基于ARIMA-GM模型的采掘工作面瓦斯涌出预测[J]．煤矿安全，2015，46(11)：152-155．

[38]马迅．分源预测法在回采工作面瓦斯涌出量预测中的应用[J]．山东煤炭科技，2020(06)：101-103．

[39]林海峰．开采保护层条件下的工作面瓦斯涌出量预测[J]．能源技术与管理，2018，43(05)：33-35．

[40]Pingyang Lyu, Ning Chen, Shanjun Mao. LSTM based encoder-decoder for short-term predictions of gas concentration using multi-sensor fusion[J]. Process Safety and Environmental Protection, 2020, 137.

[41]Lizhi Zhang, Zaihua Yang. Prediction of Gas Concentration Based on ARIMA and GARCH Model[A]. Science and Engineering Research Center.Proceedings of 2019 2nd International Conference on Sustainable Energy, Environment and Information Engineering(SEEIE 2019)[C]. Science and Engineering Research Center:Science and Engineering Research Center, 2019: 5.

[42]董丁稳，李树刚，常心坦，等．工作面多变量瓦斯体积分数时间序列预测模型[J]．采矿与安全工程学报，2012，29(01)：135-139．

[43]吴雅琴，李惠君，徐丹妮．基于IPSO-Powell优化SVM的煤与瓦斯突出预测算法[J]．工矿自动化，2020，46(04)：46-53．

[44]赵美成，贺安民，屈世甲．综采工作面瓦斯数据时间序列预测方法研究[J]．工矿自动化，2019，45(06)：80-85．

[45]徐刚，王磊，金洪伟，等．因子分析法与BP神经网络耦合模型对回采工作面瓦斯涌出量预测[J]．西安科技大学学报，2019，39(06)：965-971．

[46]李欢，贾佳，杨秀宇，等．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．

[47]郭思雯，陶玉帆，李超．基于时间序列的瓦斯浓度动态预测[J]．工矿自动化，2018，44(09)：20-25．

[48]Yiwen Zhang, Haishuai Guo, Zhihui Lu, et al. Distributed gas concentration prediction with intelligent edge devices in coal mine[J]. Engineering Applications of Artificial Intelligence, 2020, 92.

[49]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．

[50]马莉，潘少波，代新冠，等．基于PSO-Adam-GRU的煤矿瓦斯浓度预测模型[J]．西安科技大学学报，2020，40(02)：363-368．

[51]张昭昭，乔俊飞，余文．基于动态神经网络的瓦斯浓度实时预测方法[J]．控制工程，2016，23(04)：478-483．

[52]郝天轩，张海波．基于灰色系统理论的瓦斯含量多变量可视化预测方法实现[J]．工矿自动化，2013，39(06)：6-9．

[53]龚星宇，常心坦，贾澎涛．独立成分分析在瓦斯浓度预测中的应用研究[J]．工矿自动化，2015，41(04)：82-86．

[54]Anne Nassauer, Nicolas M. Legewie. Methodologische Entwicklungen in der Gewaltforschung. Videodatenanalyse, Mixed Methods und Big Data Methodological developments in violence research Video data analysis, mixed methods, and big data [J]. Österreichische Zeitschrift für Soziologie: Vierteljahresschrift der Österreichischen Gesellschaft für Soziologie, 2020, 45(Suppl 1).

[55]毕倪飞，丁光耀，陈启航，等．数据流计算模型及其在大数据处理中的应用[J]．大数据，2020，6(03)：73-86．

[56]张贤德．基于Spark平台的实时流计算推荐系统的研究与实现[D]．江苏大学，2016．

[57]戚红雨．流式处理框架发展综述[J]．信息化研究，2019，45(06)：1-8.

[58]于苹苹，倪建成，姚彬修，等．基于Spark框架的高效KNN中文文本分类算法[J]．计算机应用，2016，36(12)：3292-3297．

[59]金鑫，李龙威，苏国华，等．基于Spark框架和PSO优化算法的电力通信网络安全态势预测[J]．计算机科学，2017，44(S1)：366-371．

[60]Hesamaldin Hajialian, Cristian Toma. Network Anomaly Detection by Means of Machine Learning: Random Forest Approach with Apache Spark[J]. Informatica Economica, 2018, 22(4).

[61]Tanvir Habib Sardar, Zahid Ansari. An Analysis of Distributed Document Clustering Using MapReduce Based K -Means Algorithm[J]. Journal of The Institution of Engineers (India): Series B, 2020(prepublish).

[62]Malhotra Jyotsna, Sethi Kaur Jasleen, Mittal Mamta. Analysis of Big Data Using Two Mapper Files in Hadoop[J]. International Journal of Security and Privacy in Pervasive Computing (IJSPPC), 2021, 13(1).

[63]施炤．基于Spark Streaming的实时数据处理系统设计与实现[J]．现代信息科技，2020，4(20)：10-12．

[64]阮晓龙，贺路路．基于ELK+Kafka的智慧运维大数据分析平台研究与实现[J]．软件导刊，2020，19(06)：150-154．

[65]于金良，朱志祥，梁小江．基于Flume的MySQL数据自动收集系统[J]．计算机技术与发展，2016，26(12)：137-141．

[66]唐立，吴新胜，李亚平，等．Spark框架下DA优化的教学大数据分析算法[J]．重庆科技学院学报(自然科学版)，2020，22(05)：86-90．

[67]谭亮，周静．基于Spark Streaming的实时交通数据处理平台[J]．计算机系统应用，2018，27(10)：133-139．

[68]段雄英，赵洋洋，陈艳霞，等．基于Spark的高压断路器机械寿命预测评估[J]．高压电器，2020，56(09)：80-86．

[69]杨帆，何正伟，刘力荣．基于Spark和小波分析的水上交通异常数据实时检测方法研究[J]．陕西理工大学学报(自然科学版)，2019，35(01)：35-41．

[70]章茂庭，杨楠，蒋顺英，等．基于Spark框架的实时交通流量预测方法[J]．现代信息科技，2020，4(04)：1-8．

[71]邵帆．基于Spark的煤与瓦斯突出预警研究[D]．西安科技大学，2019．

[72]彭相洲，陈雨．一种多趋势时间序列预测的神经网络模型[J]．计算机应用研究，2020，37(S2)：47-49．

[73]张俭让，马彦龙．基于时间序列的综采工作面瓦斯浓度预测研究[J]．煤炭技术，2019，38(03)：111-113．

[74]Shuai-Jing Ren, Cai-Ping Wang, Yang Xiao, et al. Thermal properties of coal during low temperature oxidation using a grey correlation method[J]. Fuel, 2020, 260.

[75]赵蔷．主成分分析方法综述[J]．软件工程，2016，19(06)：1-3．

[76]Xiaoying Guan, Shengjia Xie, Guo Chen, et al. Modal parameter identification by adaptive parameter domain with multiple genetic algorithms[J]. Journal of Mechanical Science and Technology, 2020, 34(12).

[77]喻祥，孙辉，赵嘉，等．自适应多策略粒子群优化算法的研究综述[J]．南昌工程学院学报，2016，35(03)：71-75．