基于时空图卷积网络的瓦斯体积分数预警效果研究  

Research on performance of gas volume fraction early-warning based on spatio-temporalgraph convolutional network

作　　者：高翼飞 张晓航[1] 畅明 葛帅帅 陈伟 GAO Yifei;ZHANG Xiaohang;CHANG Ming;GE Shuaishuai;CHEN Wei(School of Economics and Management,Beijing University of Posts and Telecommunications,Beijing 100876,China;School of Application and Innovation in Information Technology,Yuncheng Vocational and Technical University,Yuncheng Shanxi 044031,China;School of Continuing Education,Yuncheng Vocational and Technical University,Yuncheng Shanxi 044031,China;College of CML Engineering and Architecture,Yuncheng Vocational and Technical University,Yuncheng Shanxi 044031,China)

机构地区：[1]北京邮电大学经济管理学院,北京100876 [2]运城职业技术大学信息技术应用创新学院,山西运城044031 [3]运城职业技术大学继续教育学院,山西运城044031 [4]运城职业技术大学建筑工程学院,山西运城044031

出　　处：《中国安全生产科学技术》2024年第1期58-64,共7页Journal of Safety Science and Technology

基　　金：国家自然科学基金项目(72271034);北京邮电大学博士研究生创新基金项目(CX2021132)。

摘　　要：为了提升瓦斯体积分数预警效果,提出1种融合时空特征的瓦斯体积分数预警模型(STGCN),以图神经网络作为基本框架对同一工作面多传感器进行统一的训练和推断,并通过图卷积的方式捕捉瓦斯体积分数数据的时空特征。在此基础上,提出瓦斯体积分数分级预警方法,将预测扩展为分级预警。研究结果表明:STGCN在瓦斯体积分数预测和预警任务上取得更好的准确率和效率。研究结果可为矿井瓦斯灾害防治提供参考。To enhance the performance of gas volume fraction early-warning,a gas volume fraction early-warning model integrating the spatio-temporal features(STGCN)was proposed.The graph neural network was taken as the foundational framework for unified training and inference of multiple sensors on the same working face,and the spatio-temporal characteristics of gas volume fraction data were captured through graph convolution.On this basis,a grading early-warning method of gas volume fraction was proposed,which extended the prediction to the grading early-warning.The results show that STGCN achieves better accuracy and efficiency in gas volume concentration prediction and early-warning tasks.The research results can provide a reference for the prevention and control of mine gas disasters.

关 键 词：瓦斯体积分数预警 图卷积网络 时空数据 可学习矩阵 分级预警方法 煤矿安全 

分 类 号：X936[环境科学与工程—安全科学]