Yang Huan, Wang Jian, Zhang Haoliang
Changzhou Research Institute Co., Ltd., China Coal Technology and Engineering Group, Changzhou 213000, China.
Tiandi (Changzhou) Automation Co., Ltd., China Coal Technology and Engineering Group, Changzhou 213000, China.
ACS Omega. 2024 May 7;9(20):22136-22144. doi: 10.1021/acsomega.4c00519. eCollection 2024 May 21.
The gas emission from the coal mine working face is influenced by multiple factors, resulting in the real-time value, fluctuation, and trend changes of gas concentration being relatively independent and interrelated. This paper establishes a gas multi-indicator warning method that can comprehensively warn the status of real-time value, fluctuation, and trend changes of gas concentration from the working face. This paper proposes six basic indicators and includes two main research contents: intelligent threshold partition and gas multi-indicator warning. First, this paper proposes an intelligent threshold partition algorithm based on GF-KMeans (genetic fixed-centered K-means), which combines a genetic algorithm (GA) and an FC-KMeans (fixed-centered K-means) algorithm to dynamically partition the threshold range corresponding to the gas warning level. The GA solves the local optimal problem in the traditional K-Means algorithm, enhancing its stability and predictability. The FC-KMeans algorithm achieves a more precise control in the initial clustering center selection. Second, this paper researches a gas multi-indicator warning method based on a multihead optimal attention (MOA)-Transformer. By using the multihead optimization attention mechanism to represent classification features and utilizing Transformer's encoder structure to classify gas warning. The experimental result shows that the accuracy of the MOA-Transformer method is 86.17%, which is 3.45% higher than that of the Transformer method. The precision of the MOA-Transformer method is 88.78%, which is 3.75% higher than that of the Transformer method. The recall of the MOA-Transformer method is 85.23%, which is 4.70% higher than that of the Transformer method. The macro-F1 of the MOA-Transformer method is 86.96%, which is 4.39% higher than that of the Transformer method. The results fully demonstrate the superiority of the MOA-Transformer method in gas warning tasks.
煤矿工作面的瓦斯涌出受多种因素影响，导致瓦斯浓度的实时值、波动情况及趋势变化相对独立又相互关联。本文建立了一种瓦斯多指标预警方法，可综合预警工作面瓦斯浓度的实时值、波动情况及趋势变化状态。本文提出了六个基本指标，并包括两个主要研究内容：智能阈值划分和瓦斯多指标预警。首先，本文提出了一种基于GF-KMeans（遗传固定中心K均值）的智能阈值划分算法，该算法将遗传算法（GA）和FC-KMeans（固定中心K均值）算法相结合，动态划分瓦斯预警等级对应的阈值范围。遗传算法解决了传统K-Means算法中的局部最优问题，提高了其稳定性和可预测性。FC-KMeans算法在初始聚类中心选择上实现了更精确的控制。其次，本文研究了一种基于多头最优注意力（MOA）-Transformer的瓦斯多指标预警方法。通过使用多头优化注意力机制来表示分类特征，并利用Transformer的编码器结构对瓦斯预警进行分类。实验结果表明，MOA-Transformer方法的准确率为86.17%，比Transformer方法高3.45%。MOA-Transformer方法的精确率为88.78%，比Transformer方法高3.75%。MOA-Transformer方法的召回率为85.23%，比Transformer方法高4.70%。MOA-Transformer方法的宏F1值为86.96%，比Transformer方法高4.39%。结果充分证明了MOA-Transformer方法在瓦斯预警任务中的优越性。
