瓦斯事故长期以来制约着我国煤炭行业的可持续发展，因此瓦斯浓度监测是煤炭安全监控的重要一环。由于煤矿井下复杂的环境和传感器自身设计的缺陷，可能导致煤矿井下甲烷传感器产生伪数据。伪数据是指与现场实际情况不相符合的监测数据，这些数据会影响煤矿安全监控系统的控制与决策。且随着煤矿监控系统升级，每日都会接收到大量的数据，如何确保算法模型能及时有效的处理分析数据，是当前研究热点。因此本文展开提升甲烷传感器伪数据识别模型性能的研究。主要的研究工作如下：

    （1）为建立实验数据样本，本文以宁夏某煤矿综采工作面甲烷传感器监测数据为研究对象。首先通过灰色关联分析法寻找工作面甲烷浓度关联因素；其次基于改进的小波阈值法去除传感器监测过程中存在的噪声干扰；再次运用支持向量机识别伪数据中的异常数据，识别率高达98.5%；然后使用三次平滑指数插补缺失数据；最后通过心跳传输机制压缩甲烷传感器数据，减少数据量。上述数据预处理工作，为提高模型预测精度奠定基础。

    （2）为识别因传感器精度漂移而产生的伪数据以长短期记忆神经网络为甲烷浓度预测主网络，将长短期记忆神经网络预测的残差通过差分整合移动平均自回归模型进一步提高网络预测精度，将预测后得到的均方根误差作为伪数据识别的参考指标，最后分析实验结果得出ARIMA-LSTM混合模型可以识别出精度偏移10%以上的伪数据，填补了对甲烷传感器精度漂移伪数据识别的空缺。

    （3）为进一步提升伪数据识别的速度，依托云边端框架将伪数据识别任务分层完成，云层主要负责模型算法训练，边层主要负责通过ARIMA-LSTM混合模型识别精度漂移数据和监测数据传输，端层主要负责区域边缘范围内环境数据的采集、数据预处理和传输。为加快混合神经网络模型的运行速度，提出一种基于Spark的分布式甲烷传感器伪数据识别模型的训练方法，结果证明通过分布式计算方法ARIMA-LSTM混合模型运行速度提升4.2倍，伪数据识别模型运行的速度得到提升。

      本文是依托综采工作面展开的伪数据识别方法的研究，通过云边端框架实现分层完成伪数据识别任务，使得在不减少采集数据量的情况下减轻边缘服务器负荷，减少事故误报警和漏报警的情况，并降低事故处置的延迟，且在边缘服务器处有效识别出精度漂移伪数据，还加快模型的更新速度，这对后续传感器的标校和协助监控系统控制决策具有较大意义。

      Methane concentration monitoring is a crucial component of coal safety supervision since gas mishaps have long been a barrier to the sustainable development of coalmine in China. The methane sensor in underground coal mine may produce false data because of the complicated environment in underground coal mines and flaws in sensor design. The data from methane sensors may not accurately reflect the true concentration , which will affect the control and decision-making of the coal mine safety monitoring system. The control and decision-making of the coal mine safety supervision system may be impacted by this false data. The current research focus on making sure that the algorithm can timely and efficiently analyze data. As coal mine supervision systems are improved, a considerable volume of data is received every day. Therefore, the goal of this work is to improve the performance of the  false data recognition model of methane  sensor. The main research work is as follows:

   （1）This paper uses methane sensor monitoring data from a fully automated coal mining face in Ningxia as the research object to create experimental data samples. First, the parameters influencing the concentration of methane in the mining face are discovered using the grey relational analysis. Second, the noise interference in the sensor monitoring is removed by an improved wavelet threshold method. Third, the support vector machine is utilized to detect anomalous data in fake data, with an accuracy rate of up to 98.5%. The triple exponential smoothing approach is then applied to interpolate the missing data. Finally, The heartbeat transmission mechanism is used to compress the data from the methane sensor and reduce the data volume. The above data preprocessing work lays a foundation for improving the accuracy of model prediction.

    （2）In order to recognize false data brought, a long short-term memory neural network is utilized as the primary network for predicting methane concentration by sensor accuracy drift. The autoregressive integrated moving average model is used to further enhance the residual predicted by the long short-term memory neural network, which can increase the accuracy of the network's prediction. As a reference index for false data recognition, We adopt the root mean square error as the evaluation criterion of the prediction. Finally, the experimental findings demonstrate that the ARIMA-LSTM hybrid model has an accuracy offset of more than 10% for detecting bogus data. It fills the gap of false data identification of methane sensor accuracy drift.

   （3）A cloud-edge-end framework is utilized to carry out the task in a hierarchical fashion in order to speed up the recognition of false data. The end layer is primarily in charge of gathering, preprocessing, and transmitting environmental data within the regional edge range, while the cloud layer trains the model. The edge layer is primarily in charge of identifying accuracy drift data and monitoring data transmission through the ARIMA-LSTM hybrid model. A distributed training strategy based on Spark is suggested in order to accelerate the performance of the hybrid neural network model. The results show that the ARIMA-LSTM speed of hybrid model operation is increased by 4.2 times through distributed computing methods, which improves the running speed of the false data identification model.

       The method of recognizing false data presented in this paper is based on a comprehensive excavation face, and the task of identifying false data is carried out in layers using a cloud-edge-end frame. By minimizing the stress on the edge server and the number of false alarms and missed alarms while maintaining the same volume of collected data, this approach shortens the training time taken to handle an accident. The proposed technique successfully locates precision drift false data on the edge server and quickens the model updating rate. Future developments on sensor calibrating and assisting the control decision of supervision system will be significantly impacted by this research.

[1]武治戎. 煤矿综采工作面自动化系统的设计及应用分析[J]. 内蒙古煤炭经济, 2022, 40(19): 16-18.

[2]许鹏飞. 2000-2021 年我国煤矿事故特征及发生规律研究[J]. 煤炭工程, 2022, 54(07): 129-133.

[3]王云刚,崔春阳,张飞燕,姚康,康军保. 2011—2020 年我国较大及以上煤矿事故统计分析及研究[J/OL]. 安全与环境学报, 2023-04-11:

[4]关于加快煤矿智能化发展的指导意见[N]. 中国煤炭报, 2020-03-05(002).

[5]中华人民共和国应急管理部. 应急管理部关于修改《煤矿安全规程》的决定:中华人民共和国应急管理部令第 8 号［EB/OL］, 煤炭工业出版社, 2022-1-25:

[6]王金川. 甲烷传感器健康状态的综合评价方法研究[D]. 徐州: 中国矿业大学, 2019.

[7]付军军. 塔山矿采煤工作面甲烷传感器安设位置探讨[J]. 科技信息, 2012, 29(32): 386-386.

[8]Li D, Wang Y, Wang J, et al. Recent advances in sensor fault diagnosis: A review[J]. Sensors and Actuators A Physical, 2020, 309: 11990.

[9]陈杰. 矿用 CH_4、 CO 传感器失效机理与解决方案[D]. 北京: 煤炭科学研究总院, 2014.

[10]郭进阳. 甲烷传感器误报警、不报警故障原因分析及对策[J]. 技术与市场, 2015, 22(08): 189-191.

[11]中国煤炭科工集团有限公司.煤矿安全监控系统升级改造验收规范［EB/OL］. 国家煤矿安全监察局, 2019-12-56:

[12]丁恩杰, 俞啸, 夏冰, 等 .矿山信息化发展及以数字孪生为核心的智慧矿山关键技术[J]. 煤炭学报, 2022, 47(01): 564-578.

[13]王国法. 煤矿智能化最新技术进展与问题探讨[J]. 煤炭科学技术, 2022, 50(01): 1-27.

[14]邹哲强, 张立斌, 蒋泽, 等. 煤矿监控系统产生伪数据的原因和应对措施[J]. 工矿自动化, 2013, 39(09): 1-4.

[15]冯亚飞, 肖运江. 煤矿安全监控系统常见误报警原因分析及软件处理方法[J]. 工矿自动化, 2013, 39(04): 90-92.

[16]王启峰. 煤矿安全监控系统异常监测数据处理方法探索[J]. 电子世界, 2013, 39(24): 20-21.

[17]郭江涛. 煤矿安全监控系统现状及发展趋势[J]. 煤矿机械, 2017, 38(03): 1-3.

[18]何敏. 新旧安全监控系统瓦斯伪数据的异同点研究[J]. 煤矿机械, 2020, 41(02): 36-39.

[19]Liu H, Shah S, Wei J. On-line outlier detection and data cleaning[J]. Computers & Chemical Engineering, 2004, 28(9): 1635-1647.

[20]熊君丽.高维空间下基于密度的离群点探测算法实现[J]. 现代电子技术, 2006, 30(15): 67-69.

[21]Almeida J, Barbosa L, Pais A, et al. Improving hierarchical cluster analysis: A new method with outlier detection and automatic clustering[J]. Chemometrics & Intelligent Laboratory Systems, 2007, 87(2): 208-217.

[22]Yang Z, Meratnia N, Havinga P. Ensuring high sensor data quality through use of online outlier detection techniques[J]. International Journal of Sensor Networks, 2010, 7(3): 141-151.

[23]胡云, 施珺, 王崇骏, 等. 基于全局最近邻的离群点检测算法[J]. 计算机应用, 2011, 31(10): 2778-2718.

[24]姜立明, 柴瑞敏. 基于单元格和属性权重的离群点检测[J]. 计算机应用与软件, 2012, 29(10): 216-218.

[25]王勇, 殷大发, 邸彭浩, 等. 煤矿瓦斯监测系统异常数据识别技术研究与应用[J]. 工矿自动化, 2013, 41(4): 83-86.

[26]陈超凡, 李连林. 煤矿安全监控网络异常数据自动甄别系统[J]. 煤矿安全, 2013, 44(7): 105-107.

[27]Lee J, Cho N W. Fast outlier detection using a grid-based algorithm [J]. PLoS ONE, 2016, 11(11): e0165972.

[28]冯震, 付敬奇, 熊南.一种快速的离群点检测方法[J]. 电子测量与仪器学报, 2016, 30(11): 1726-1734.

[29]黄凯峰. 煤矿安全监测监控系统瓦斯浓度异常信号辨识方法研究[D]. 安徽: 安徽理工大学, 2016.

[30]谢俊生. 煤矿安全监控系统联网报警数据识别技术[J]. 工矿自动化, 2017, 43(2): 90-94.

[31]严宏, 杨波, 杨红雨.基于异方差高斯过程的时间序列数据离群点检测[J]. 计算机应用, 2018, 38(05): 1346-1352.

[32]吴海波, 施式亮, 念其锋. 瓦斯浓度流数据实时异常检测方法[J]. 计算机与数字工程, 2019, 47(5): 1086-1090.

[33]Lyu P Y, Chen N, Mao S J, et al. LSTM based encod-er-decoder for short-term predictions of gas concentration using multi-sensor fusion[J]. Process Safety and Environmental Pro-tection, 2020, 137: 93-105.

[34]王晨辉, 王恩东, 高晓锋. 基于CNN-BiLSTM特征融合的异常检测算法[J]. 计算机应用与软件, 2022, 39(12): 272-277.

[35]申琢. 基于云计算和大数据挖掘的矿山事故预警系统研究与设计[J]. 中国煤炭, 2017, 43(12): 109-114.

[36] 裴秋艳. 煤矿海量通风安全数据挖掘算法及智能分析系统研究[D]. 太原: 太原理工大学, 2017.

[37]毛善君, 刘孝孔, 雷小锋, 等. 智能矿井安全生产大数据集成分析平台及其应用[J]. 煤炭科学技术, 2018, 46(12): 169-176.

[38]郑凯. 大数据框架下的多源矿井瓦斯地质信息融合及应用[D]. 中国矿业大学, 2019.

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

[40]郭海帅. 面向煤矿大数据的分布式瓦斯浓度预测方法研究[D]. 安徽: 安徽大学, 2019.

[41]王恩元, 冯小军, 刘晓斐,等. 煤矿瓦斯灾害风险隐患大数据监测预警云平台与应用[J]. 煤炭科学技术, 2022, 50(1): 142-150.

[42]姜德义, 魏立科, 王翀, 等. 智慧矿山边缘云协同计算技术架构与基础保障关键技术探讨[J]. 煤炭学报, 2020, 45(01): 484-492.

[43]屈世甲, 武福生. 基于边缘计算的采煤工作面甲烷监测模式研究[J]. 煤炭科学技术, 2020, 48(12): 161-167.

[44]卜滕滕, 屈世甲, 武福生, 等. 边缘计算驱动的瓦斯灾害智能感知方法研究[J]. 煤矿安全, 2021, 52(11): 110-116.

[45]陈晓晶. 基于“云-边-端”协同的煤矿火灾智能化防控体系建设[J]. 煤炭科学技术, 2022, 50(12): 136-143.

[46]聂晓艳, 郭丽芳. 端边云一体的煤矿监测预警与应急联动模型研究[J]. 煤炭技术, 2023, 42(01): 261-264.

[47]张轶群. 高性能燃烧催化剂的制备及其甲烷气敏特性的研究[D]. 长春: 吉林大学, 2012.

[48]鲁文帅. 纳米催化剂原位打印的石英微热板甲烷传感阵列[D]. 北京: 清华大学, 2016.

[49]赵文杰. 陶瓷微热板气体传感器阵列及检测系统研究[D]. 哈尔滨: 哈尔滨理工大学, 2015.

[50]翟波, 戴峻, 王晓虹. 光干涉甲烷传感器标校与非线性补偿研究[J]. 煤矿安全, 2017, 48(4): 115-117.

[51]陈长伦, 何建波, 刘伟, 等. 电化学式气体传感器的研究进展[J]. 传感器世界, 2004, 10(04): 11-15.

[52]程勇. 抗干扰技术在瓦斯传感器中的应用[J]. 仪表技术与传感器, 2007, 44(9): 57-58.

[53]刘竹琴, 白泽生. 传感器电路的噪声及其抗干扰技术研究[J]. 现代电子技术, 2011, 34(14): 161-165.

[54]朱正和. 提高甲烷载体催化元件灵敏度的研究[J]. 矿业安全与环保, 2003, 32(06): 21-22.

[55]Srivastava M, Anderson C L, Freed J H. A new wavelet denoising method for selecting decomposition levels and noise thresholds [J]. IEEE Access Practical Innovations Open Solutions, 2016, 4: 3862-3877.

[56]Sherstinsky A. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network [J]. Physica D Nonlinear Phenomena, 2020, 404: 132306.

[57]Xu D, Zhang Q, Ding Y, et al. Application of a hybrid ARIMA-LSTM model based on the SPEI for drought forecasting[J]. Environmental Science and Pollution Research, 2022, 29(3): 4128-4144.

[58]Malhotra Jyotsna, Sethi Kaur Jasleen, Mittal Mamta. Analysis of big data using two mapper files in Hadoop Hadoop [J]. International Journal of Security and Privacy in Pervasive Computing (IJSPPC), 2021, 13(1): 69-77.

[59]Mostafaeipour A, Rafsanjani A J, Ahmadi M, et al. Investigating the performance of Hadoop and Spark platforms on machine learning algorithms[J]. The Journal of Supercomputing, 2021, 77(2): 1273-1300.

[60]Zhao C, He J, Wang Q G. Resilient distributed optimization algorithm against adversarial attacks [J]. IEEE Transactions on Automatic Control, 2019, 65(10): 4308-4315.