National and Local Joint Engineering Laboratory of Internet Application Technology on Mine, China University of Mining and Technology
School of Information and Control Engineering, China University of Mining and Technology
School of Safety Engineering, China University of Mining and Technology
Tiandi (Changzhou) Automation Co., Ltd.

煤矿气体传感器用于混合气体检测时测量信号之间有交叉干扰，难以保证检测准确性。对于相同的待识别气体，传统气体识别算法的识别精度低于基于神经网络的气体识别算法，神经网络通过调整其网络层、每层神经元的数量、神经元的激活函数和各层网络之间的权重等来实现更高的气体识别精度。介绍了煤矿混合气体检测系统结构，通过构建气体传感阵列，利用其多维空间气体响应模式，并结合特定的气体识别算法，实现对混合气体的定性定量识别。重点分析了几种面向煤矿混合气体检测的神经网络算法并进行了对比分析，主要包括反向传播（BP）神经网络、卷积神经网络（CNN）、循环神经网络（RNN）、径向基函数（RBF）神经网络：BP神经网络通常可以达到较高的分类精度，然而需要训练大量的参数，训练时间长，通常为了减少时长和提高精度，可以将BP神经网络与其他算法相结合；CNN可以自动提取数据特征，精度和训练速度都优于BP神经网络，但其易于陷入局部最优；RNN可以使用更少的数据并提取更有效的特征，但容易出现梯度消失等问题；RBF神经网络具有较强的鲁棒性和在线学习能力，但其通常需要大量数据完成模型训练。神经网络算法的应用将大幅提升煤矿混合气体的检测精度，保障煤矿智能化的实现。

When coal mine gas sensors are used for mixed gas detection, there is cross interference between measurement signals. It is difficult to ensure detection accuracy. For the same gas to be identified, the recognition precision of traditional gas recognition algorithms is lower than that of gas recognition algorithms based on neural networks. Neural networks achieve higher gas recognition accuracy by adjusting their network layers, the number of neurons in each layer, the activation function of neurons, and the weights between each layer of networks. This paper introduces the structure of a coal mine mixed gas detection system. By constructing a gas sensor array, utilizing its multi-dimensional gas response mode, and combining specific gas recognition algorithms, the qualitative and quantitative recognition of mixed gases is achieved. Several neural network algorithms for mixed gas detection in coal mines are analyzed and compared. The algorithms mainly include backpropagation (BP) neural network, convolutional neural network (CNN), recurrent neural network (RNN), and radial basis function (RBF) neural network. BP neural network can usually achieve high classification precision, but requires training a large number of parameters and a long training time. Usually, in order to reduce time and improve precision, BP neural networks can be combined with other algorithms. CNN can automatically extract data features, with better precision and training speed than BP neural networks. But it is prone to falling into local optima. RNN can use less data and extract more effective features, but it is prone to problems such as gradient vanishing. RBF neural networks have strong robustness and online learning capability, but they usually require a large amount of data to complete model training. The application of neural network algorithms will significantly improve the detection precision of mixed gases in coal mines, ensuring the implementation of intelligent coal mines.

国家自然科学基金青年基金项目（62204260，52174222）；科技部长三角联合攻关项目（2022CSJGG0703）；天地科技股份有限公司科技创新创业资金专项项目（2019-TD-ZD007）。

焦明之，沈中丽，周扬明，等. 面向煤矿混合气体检测的神经网络算法研究进展[J]. 工矿自动化，2023，49（9）：115-121.

JIAO Mingzhi, SHEN Zhongli, ZHOU Yangming, et al. Research progress on neural network algorithms for mixed gas detection in coal mines[J]. Journal of Mine Automation，2023，49（9）：115-121.