瓦斯突出是煤矿五大灾害之一，及时、准确、有效预测瓦斯浓度是预防瓦斯突出灾害的关键。神经网络由于其强大的自学习能力和鲁棒性，在瓦斯浓度预测领域受到了广泛应用。但传统神经网络预测模型大多没有考虑瓦斯浓度数据的时空特性，仅仅采用单点时间序列预测方法来实现对瓦斯浓度的预测，影响了采集工作面瓦斯浓度预测的准确性。本文通过分析瓦斯浓度数据的时空特性，构建了一种基于时空数据的时空图神经网络瓦斯浓度预测模型——AGCN-WaveNet。本文的主要研究内容如下：

首先，针对现有的大多数时空图神经网络模型在捕获时空数据的长时间依赖时效果较差的问题，本文使用图卷积网络结合WaveNet模型构造时空图神经网络模型。在空间维度上使用图卷积方式聚合邻居节点的信息，在时间维度上引入WaveNet模型来提取数据的时间特征。通过使用WaveNet模型中的扩张因果卷积模块，使模型可以使用较少的网络层数情况下获得非常大的感受野，也有效的避免了梯度爆炸等问题。

其次，针对大多数时空图神经网络在进行建模时，都假定各节点之间的空间关系是固定不变的，但固定的空间结构并不一定可以完整的表示节点之间的空间关系，可能会造成空间关系的缺失而导致模型性能降低的问题。本文通过引入注意力机制来实现各节点之间的空间关联强度的动态调整，使模型可以更完整的表示节点之间的空间关系。通过在公共数据集下的实验性能比较表明，注意力机制的引入使模型的预测精度得到了提升。

最后，针对传统神经网络瓦斯浓度预测模型没有考虑瓦斯浓度数据的时空特性导致预测精度偏低的问题，本文采用基于AGCN-WaveNet的瓦斯浓度预测模型在实测瓦斯浓度数据集上进行了实例应用。根据矿井内风向和瓦斯浓度监测点之间的距离，采用改进的高斯核函数对瓦斯浓度数据的空间结构进行构造。实验数据的来源为某矿井工作面多个瓦斯浓度监测点收集到的数据。通过将模型预测的结果与现有的瓦斯浓度预测模型进行对比表明，本文所提模型由于同时考虑了瓦斯浓度数据的时空特性，可以更好的拟合瓦斯浓度的变化趋势，在预测精度上有较大的提升。在AGCN-WaveNet瓦斯浓度预测模型的基础上，本文采用C#设计了一个简易的瓦斯浓度预测系统。该系统界面简洁，操作简单，能够对工作面瓦斯含量做出准确预报，可以有效的预防矿井瓦斯灾害的发生。

Gas outburst is one of the five major disasters in coal mines. Timely, accurate and effective prediction of gas concentration is the key to prevent gas concentration disasters. Neural networks have been widely used in the field of gas concentration prediction due to their strong self-learning ability and robustness. However, most of the traditional neural network prediction models do not consider the spatiotemporal characteristics of gas concentration data, and only use the single-point time series prediction method to predict the gas concentration, which affects the accuracy of the gas concentration prediction at the collecting face. In this paper, by analyzing the spatiotemporal characteristics of gas concentration data, a spatiotemporal graph neural network gas concentration prediction model based on spatiotemporal data is constructed——AGCN-WaveNet. The main research contents of this paper are as follows :

        First, to address the problem that most existing spatiotemporal graph network models are less effective in capturing the long-range temporal sequences of spatiotemporal data, this paper uses graph convolutional network combined with WaveNet model to construct spatiotemporal graph network models. In the spatial dimension, use graph convolutional network to aggregate the information of neighbor nodes. In the temporal dimension, the WaveNet network is introduced to extract the time characteristics of data. By using the dilated causal convolution module of the WaveNet model, the spatiotemporal graph network model can be used to obtain very large receptive field with a small number of network layers. It also effectively avoids the problems such as gradient explosion.

Secondly, when modeling for most spatiotemporal network, it is assumed that the spatial relationship between nodes is fixed. However, the fixed spatial structure does not necessarily represent the spatial relationship between nodes completely, which may lead to the lack of spatial relationship and reduce the performance of the model. In this paper, the attention mechanism is introduced to realize the dynamic adjustment of spatial correlation strength between nodes, so that the model can more completely represent the spatial relationship between nodes. The comparison of experimental performance under public datasets shows that the introduction of attention mechanism improves the prediction accuracy of the model.

Finally, in view of the low prediction accuracy of the traditional single-point gas concentration prediction model, this paper uses the gas concentration prediction model based on the AGCN-WaveNet model to apply it to the measured gas concentration data set. The spatial structure of the gas concentration data is constructed using an improved Gaussian function based on the wind direction and the distance between gas concentration monitoring points in the mine. The experimental data were collected from a number of gas concentration monitoring points at the working face of a mine. A comparison of the model prediction results with existing gas concentration prediction models shows that the model proposed in this paper can better fit the changing trend of gas concentration and improve the prediction accuracy as it also takes into account the spatial and temporal characteristics of gas concentration data. Based on the AGCN-WaveNet gas prediction model, a simple gas concentration prediction system is designed using C#. The system has a simple interface, is easy to operate and can make accurate forecasts of gas content at the working face, which can effectively prevent the occurrence of gas disasters in mines.

[1] 王长建,汪菲,叶玉瑶,等.基于供需视角的中国煤炭消费演变特征及其驱动机制[J].自然资源学报,2020,35(11):2708-2723.

[2] 何则,周彦楠,刘毅.2050年中国能源消费结构的系统动力学模拟——基于重点行业的转型情景[J].自然资源学报,2020,35(11):2696-2707.

[3] 王安建,高芯蕊.中国能源与重要矿产资源需求展望[J].中国科学院院刊,2020,35(03):338-344.

[4] 冯烽.中国能源消费驱动因素分析与促进“双循环”发展的启示[J].中国能源,2021,43(02):21-27+43.

[5] 余明高,阳旭峰,郑凯,等.我国煤矿瓦斯爆炸抑爆减灾技术的研究进展及发展趋势[J].煤炭学报,2020,45(01):168-188.

[6] 王晓蕾, 袁印, 陈鄢. 中国矿井瓦斯浓度检测技术现状及发展趋势[J]. 科学技术与工程, 2020, 20(23):8-13.

[7] 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.

[8] 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].西安科技大学学报,2018,38(06):1027-1035.

[9] 刘鹏,魏卉子,景江波,等.基于增强CART回归算法的煤矿瓦斯涌出量预测技术[J].煤炭科学技术,2019,47(11):116-122.

[10] 高魁,乔国栋,刘泽功,等.煤与瓦斯突出机理分类研究构想及其应用探讨[J].采矿与安 全工程学报,2019,36(05):1043-1051.

[11] 徐刚,王磊,王海涛,等.基于灰色理论与BP神经网络的煤层瓦斯含量预测方法[J].煤炭技术,2019,38(11):82-85.

[12] 李绍良, 王茜, 张毅,等.基于混沌相空间重构的IGA-LSSVM在线煤矿工作面瓦斯浓度预测模型的研究[J].中国煤炭,2019,45(05):41-45+54.

[13] 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(1):33-38.

[14] Shi X L, Lin G, Que H, et al. Grey relational analysis using Gaussian process regression method for dissolved gas concentration prediction[J]. International Journal of Machine Learning & Cybernetics, 2018, 5(13):1-10.

[15] 冯亮, 张立强. 灰色理论预测煤层瓦斯涌出量的研究与应用[J]. 煤炭与化工, 2016,39(01):44-47.

[16] 熊祖强, 王晓蕾. 矿井相对瓦斯涌出量动态无偏灰色马尔科夫预测[J]. 安全与环境学报,2015, 15(03):15-18.

[17] 秦志.基于灰色理论的回采工作面瓦斯涌出量动态预测研究[J].能源与环保,2018,40(02):17-21.

[18] 吕坤坤,吕品,张文庆.基于灰色GM(2,1)模型的工作面瓦斯涌出量预测[J].煤炭技术,2014,33(08):30-32.

[19] 李忠辉,马云波,郑安琪,等.掘进工作面突出危险性预测及危险区可视化研究[J].中国安全科学学报,2020,30(11):53-59.

[20] 王菲茵, 王启飞, 张小千. 瓦斯异常涌出混沌时间序列预测模型的建立[J]. 中国煤炭, 2017, 43(8):7-15.

[21] 程健,白静宜,钱建生,等.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.

[22] 耿越.基于混沌粒子群神经网络的瓦斯浓度预测[J].中国煤炭,2017,43(03):124-129.

[23] 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.

[24] 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.

[25] 李栋,孙振明,李梅,等.基于混沌粒子群的AWLSSVM瓦斯预测研究[J].煤矿安全, 2020,51(08): 193-198+205.

[26] 李冬,彭苏萍,杜文凤,等.煤层瓦斯突出危险区综合预测方法[J].煤炭学报,2018,43 (02):466-472

[27] 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(6): 965-971.

[28] Jia P, H Liu, Wang S, et al. Research on a Mine Gas Concentration Forecasting Model Based on a GRU Network[J]. IEEE Access, 2020, 8:38023-38031.

[29] Lu P, Chen N, Mao S, et al. LSTM based encoder-decoder for short-term predictions of gas concentration using multi-sensor fusion[J]. Process Safety and Environmental Protection, 2020, 137: 93-105.

[30] Li M, Zhu Z. Spatial-temporal fusion graph neural networks for traffic flow forecast ting[C]//Proceedings of the AAAI conference on artificial intelligence, 2021:4189-4196.

[31] Fan J, Bai J, Li Z, et al. A GNN-RNN Approach for Harnessing Geospatial and Temporal Information: Application to Crop Yield Prediction [C]//Proceedings of the AAAI conference on artificial intelligence, 2021:1546-1553.

[32] Gao Z, Xu Y, et al. Conditional Local Convolution for Spatio-temporal Meteorological Forecasting[C]//Proceedings of the AAAI conference on artificial intelligence, 2022:2354 -2367.

[33] Zhang H, Chen L, Cao J, et al. A combined traffic flow forecasting model based on graph convolutional network and attention mechanism[J]. International Journal of Modern Physics C, 2021, 32(12): 215-223.

[34] Wang X,Ma Y,Wang Y,et al. Traffic flow prediction via spatial temporal graph neural network[C]// Proceedings of The Web Conference 2020,2020: 1082-1092.

[35] Yu B, Yin H, Zhu Z, et al. Spatio-temporal graph convolutional networks: a deep learning framework for traffic forecasting[C]//International Joint Conference on Artificial Intelligence,2018:3634-3640.

[36] Dai R, Xu S, Gu Q, et al. Hybrid spatio-temporal graph convolutional network: Improving traffic prediction with navigation data[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2020: 3074-3082.

[37] Pan B,Cai H,Huang D,et al. Spatio-temporal graph for video captioning with knowledgee distillation [C]//Computer Vision and Pattern Recognition,2020:1682-1697.

[38] Yu C, Ma X, Ren J, et al. Spatio-temporal graph transformer networks for pedestrian trajectory prediction[C]//European Conference on Computer Vision. Springer, Cham, 2020: 507-523.

[39] 李成武, 王义林, 王其江,等. 直接法瓦斯含量测定结果准确性实验研究[J]. 煤炭学报, 2020, 45(1):8.

[40] 李欢. 基于时空序列的瓦斯浓度预测方法研究[D].中国矿业大学,2019.

[41] 冯宁,郭晟楠,宋超,等.面向交通流量预测的多组件时空图卷积网络[J].软件学报,2019,30(03):759- 769.

[42] Li Y, Yu R, Sha C, et al. Diffusion convolutional recurrent neural network: Data-driven traffic forecasting[C]//International Conference on Learning Representations,2018:2764-2778.

[43] 汤文兵, 任正云, 韩芳. 基于注意力机制的协同卷积动态推荐网络[J]. 自动化学报, 2021, 47(10):11-23.

[44] Oord A, Dieleman S, Zen H, et al. WaveNet: A generative model for raw audio[J]. 2016. arXiv preprint arXiv:1609.03499.

[45] Tian C,Chan W. Spatial-temporal attention WaveNet: a deep learning framework for traffic prediction considering spatial temporal dependencies[J] .IET Intelligent Transport Systems,2021,15(4):549-561.

[46] Rueda F D,Alejandro D . Short-term load forecasting using encoder-decoder WaveNet: application to the french grid[J]. Energies, 2021, 14(9):2524-2536.

[47] Dauphin Y N,Fan A,Auli M,et al. Language modeling with gated convolutional networks[C]// International conference on machine learning,2017: 933-941.

[48] Sutskever I, Vinyals O, Le Q V. Sequence to sequence learning with neural networks[J]. Advances in neural information processing systems, 2014, 27(02):369-381.

[49] 汤进. 基于瓦斯浓度时间序列的大数据预测[D].中国矿业大学,2018.

[50] 魏永强. 煤矿瓦斯监测数据插值与预测研究[D].中国矿业大学(北京),2015.