井下瓦斯浓度的变化与煤炭安全生产有着密切的关系，然而现有的瓦斯浓度预测方法还存在一些问题。为了解决这些问题，论文提出了一种基于时空序列的瓦斯浓度预测方法，其主要研究工作和创新如下：

（1）针对原始瓦斯浓度数据中由于存在异常值而引起的瓦斯浓度预测精度下降的问题，论文提出了一种基于双生成对抗网络（DUAL-ADGAN）的检测模型用于检测瓦斯数据中的异常值。DUAL-ADGAN模型由训练网络和异常值检测网络构成，首先训练网络负责训练模型参数；其次将训练好的参数共享给异常值检测网络；最后在异常值检测网络中分别计算重构损失、预测损失和判别损失，并由以上三种损失值计算得到异常分值，与动态阈值相比较进行异常值检测。实验结果表明DUAL-ADGAN模型在瓦斯浓度数据集上F1值相比于次优模型提升了7.76%，验证了该模型能够有效的检测时序数据中的异常值，并提高了模型训练的稳定性，降低了异常值误检、漏检率。

（2）针对现有瓦斯浓度预测模型未能充分利用数据中的时空特征以及短期非平稳特征的问题，论文提出了一种基于时空序列的瓦斯浓度预测模型（GSN）用于提高瓦斯浓度的预测精度。GSN模型分别由时间特征提取模块、空间特征提取模块和特征融合与预测模块构成。首先，时间特征提取模块通过引入自注意力机制和差分结构用于提取瓦斯浓度数据之间的自相关性特征和非平稳特征；其次，采用图卷积神经网络提取不同位置瓦斯浓度数据之间的空间特征；最后，将上述多种特征进行融合，并通过模型的全连接层预测瓦斯浓度。实验结果表明GSN模型在瓦斯浓度数据集上均方误差相比于次优模型降低了4.7%，验证了GSN模型能够有效提取数据中的时空特征和非平稳特征。

（3）为了将理论研究内容用于指导煤矿的安全生产工作，论文设计并实现了一套基于深度学习的瓦斯浓度预测预警系统，该系统包含瓦斯浓度数据采集与管理、异常值检测与修复、瓦斯浓度预测与预警和系统管理四个功能模块。经过测试，系统各功能模块运行正常，该系统对于煤矿井下的安全生产工作具有一定的指导价值。

The variation of underground gas concentration is closely related to coal safety production, however, there are still some problems in the existing gas concentration prediction methods. In order to solve these problems, a gas concentration prediction method based on time series is proposed in the paper, and its main research works and innovations are as follows:

(1) To address the problem of degradation of gas concentration prediction accuracy due to the presence of anomalies in the raw gas concentration data, a dual generative adversarial network (DUAL-ADGAN) based detection model is proposed to detect the anomalies in the gas data.The DUAL-ADGAN model consists of a training network and an anomaly detection network. Firstly, the training network is responsible for training the model parameters; Secondly, the trained parameters are shared to the anomaly detection network; Finally, the reconstruction loss, prediction loss and discriminant loss are calculated in the anomaly detection network, and the anomaly score is calculated from the above three loss values and compared with the dynamic threshold to detect the anomaly. The experimental results show that the DUAL-ADGAN model improves the F1 value on the gas concentration data set by 7.76% compared with the suboptimal model, which verifies that the DUAL-ADGAN model can effectively detect the anomalous values in the time-series data. It also improves the stability of model training and reduces the rate of false detection and leakage of anomalous values.

(2) To address the problem that existing gas concentration prediction models can not make full use of the spatio-temporal features and short-term non-stationary features in the data, a gas concentration prediction model (GSN) based on spatio-temporal series is proposed to improve the accuracy of gas concentration prediction. The GSN model consists of a temporal feature extraction module, a spatial feature extraction module and a feature fusion prediction module, respectively. Firstly, the temporal feature extraction module is used to extract autocorrelation features between gas concentration data and non-smooth features in the data by introducing self-attention mechanism and difference structure; Secondly, the spatial features between gas concentration data are extracted by using graphical convolutional neural network; Finally, the spatio-temporal features are fused and the gas concentration is predicted by the fully connected layer of the model. The experimental results show that the mean square error of the GSN model on the gas concentration data set is reduced by 4.7% compared with the suboptimal model, which verifies that the GSN model can effectively extract spatio-temporal features and non-stationary features.

(3) In order to apply the theoretical research content to the safety work of coal mines, the paper designs and implements a deep learning based gas concentration prediction and early warning system. The system consists of four functional modules: gas concentration data collection and management, anomaly detection and repair, gas concentration prediction and system management. After testing, the system can effectively perform gas concentration prediction and early warning, which has certain guidance value for the safety production work in underground coal mines.

[1]王海宁.中国煤炭资源分布特征及其基础性作用新思考[J].中国煤炭地质,2018,30(07):5－9.

[2]韩云春，任波，杨理强，陈本良，段昌瑞，邓东生，李志兵．我国煤与瓦斯突出预测指标及技术研究现状浅析[J].陕西煤炭, 2020, 39(01): 51－53+12.

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

[4]孟远,谢东海,苏波,谷翱翔.2010年-2019年全国煤矿生产安全事故统计与现状分析[J].矿业工程研究, 2020, 35(04): 27－33.

[5]魏引尚. 基于概率统计的通风巷道瓦斯积聚危险性分析研究[D].西安科技大学,2004.

[6]屈世甲.基于单一测点瓦斯异常预报警方法的探讨[J].煤矿安全,2012,43(11): 168－170.

[7]Breunig MM, Kriegel HP, Ng RT, et al. LOF: identifying density-based local outliers[C]//Proceedings of the 2000 ACM SIGMOD international conference on Management of data. 2000: 93－104.

[8]Ripan R C, Sarker I H, Hossain S M, et al. A data-driven heart disease prediction model through K-means clustering-based anomaly detection[J]. SN Computer Science, 2021, 2(2): 1－12.

[9]Liu F T, Ting K M, Zhou Z H. Isolation-based anomaly detection[J]. ACM Transactions on Knowledge Discovery from Data (TKDD), 2012, 6(1): 1－39.

[10]Santiago-Paz J, Torres-Roman D, Figueroa-Ypiña A, et al. Using generalized entropies and OC-SVM with Mahalanobis kernel for detection and classification of anomalies in network traffic[J]. Entropy, 2015, 17(9): 6239－6257.

[11]杨梦莹. 基于多传感器数据融合的矿用瓦斯传感器故障诊断方法研究[D].中国矿业大学,2019.

[12]Feng C, Tian P. Time series anomaly detection for cyber-physical systems via neural system identification and bayesian filtering[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 2021: 2858－2867.

[13]Wu D, Jiang Z, Xie X, et al. LSTM learning with Bayesian and Gaussian processing for anomaly detection in industrial IoT[J]. IEEE Transactions on Industrial Informatics, 2019, 16(8): 5244－5253.

[14]Salehi M, Arya A, Pajoum B, et al. Arae: Adversarially robust training of autoencoders improves novelty detection[J]. Neural Networks, 2021, 144: 726－736.

[15]Li D, Chen D, Jin B, et al. MAD-GAN: Multivariate anomaly detection for time series data with generative adversarial networks[C]//International conference on artificial neural networks. Springer, Cham, 2019: 703－716.

[16]Geiger A, Liu D, Alnegheimish S, et al. Tadgan: Time series anomaly detection using generative adversarial networks[C]//2020 IEEE International Conference on Big Data (Big Data). IEEE, 2020: 33－43.

[17]Gulrajani I, Ahmed F, Arjovsky M, et al. Improved training of wasserstein gans[J]. Advances in neural information processing systems, 2017, 30－40.

[18]Ngo P C, Winarto A A, Kou C K L, et al. Fence GAN: Towards better anomaly detection[C]//2019 IEEE 31St International Conference on tools with artificial intelligence (ICTAI). IEEE, 2019: 141－148.

[19]孙肖琦，郑欣.工作面煤与瓦斯突出预测方法综述[J].煤炭技术,2019,38(10):111－114.

[20]崔俊飞.瓦斯地质动态分析及瓦斯涌出实时预警系统[J].工矿自动化,2015,41(03):5－9.

[21]肖鹏，李树刚，宋莹，张天军.瓦斯涌出量的灰色建模及其预测[J].采矿与安全工程学报,2009,26(03):318－321.

[22]许满贵，高帅帅，曹艳军，卢亚楠，王娇娇.基于灰色理论-多元回归分析的瓦斯含量预测[J].煤矿安全,2018,49(09):211－214.

[23]李鑫灵，袁梅，敖选俊，隆能增，张平，许石青.PCA-SVM模型在煤层瓦斯涌出量预测中的应用[J].工业安全与环保,2019,45(10):35－39.

[24]朱志洁，张宏伟，韩军，宋卫华.基于PCA-BP神经网络的煤与瓦斯突出预测研究[J].中国安全科学学报,2013,23(04):45－50.

[25]智登奎,李国勇.基于遗传算法优化神经网络瓦斯浓度预测[J].矿山机械,2013,41(04):117－120.

[26]刘奕君，赵强，郝文利.基于遗传算法优化BP神经网络的瓦斯浓度预测研究[J].矿业安全与环保,2015,42(02):56－60.

[27]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.

[28]Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural computation, 1997, 9(8): 1735－1780.

[29]Cho K, Van Merriënboer B, Gulcehre C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. arXiv preprint arXiv:1406.1078, 2014.

[30]Jia P, Liu H, Wang S, et al. Research on a mine gas concentration forecasting model based on a GRU network[J]. IEEE Access, 2020, 8: 38023－38031.

[31]Liang R, Chang X, Jia P, et al. Mine Gas Concentration Forecasting Model Based on an Optimized BiGRU Network[J]. ACS omega, 2020, 5(44): 28579－28586.

[32]刘剑桥. 基于改进LSTM循环神经网络瓦斯数据时间序列预测研究[D].中国矿业大学,2018.

[33]刘航舵. 基于循环神经网络的瓦斯灾害预测预警研究[D].西安科技大学,2020.

[34]程子均，马六章，张翼翔.基于LSTM-FC的瓦斯浓度时空分布预测[J].计算机工程与应用,2020,56(16):258－264.

[35]李叙达. 改进灰狼算法优化的机器学习瓦斯浓度预测研究[D].辽宁工程技术大学,2022.

[36]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.

[37]丁汀，颜登程，张以文，周珊.融合煤矿多维时序数据的瓦斯异常检测算法[J].计算机集成制造系统,2020,26(06):1651－1659.

[38]戴俊，王俊，朱忠奎，沈长青，黄伟国.基于生成对抗网络和自动编码器的机械系统异常检测[J].仪器仪表学报,2019,40(09):16－26.

[39]郭森森，王同力，慕德俊.基于生成对抗网络与自编码器的网络流量异常检测模型[J].信息网络安全,2022,22(12):7－15.

[40]周杰. 基于生成对抗网络的异常检测方法的研究[D].合肥工业大学,2020.

[41]Bashar M A, Nayak R. TAnoGAN: Time series anomaly detection with generative adversarial networks[C]//2020 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE, 2020: 1778-1785.

[42]魏宗康. 平顶山某矿丁6-31030工作面瓦斯异常特征及诊断准则研究[D].中国矿业大学,2020.

[43]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].安徽理工大学,2016.

[44]Fang W, Chen Y, Xue Q. Survey on research of RNN-based spatio-temporal sequence prediction algorithms[J]. Journal on Big Data, 2021, 3(3): 97－110.

[45]姚庆安，张鑫，刘力鸣，冯云丛，金镇君.融合注意力机制和多尺度特征的图像语义分割[J].吉林大学学报(理学版),2022,60(06):1383－1390.

[46]宋超. 基于注意力机制的自然场景中英文文本识别算法研究[D].湖南大学,2021.

[47]Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30－44.

[48]徐冰冰,岑科廷,黄俊杰,沈华伟,程学旗.图卷积神经网络综述[J].计算机学报,2020,43(05):755－780.

[49]Bai J, Zhu J, Song Y, et al. A3T-GCN: Attention Temporal Graph Convolutional Network for Traffic Forecasting[J]. International Journal of Geo-Information, 2021, 10(7):485－494.

[50]Zhao L, Song Y, Zhang C, et al. T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, PP(99):1－11.

[51]Wu Z, Pan S, Long G, et al. Connecting the dots: Multivariate time series forecasting with graph neural networks[C]//Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 2020: 753－763.

[52]Bai L, Yao L, Li C, et al. Adaptive graph convolutional recurrent network for traffic forecasting[J]. Advances in neural information processing systems, 2020, 33: 17804－17815.