基于GCN-LSTM架构的煤矿瓦斯预测方法  

A Coal Mine Gas Prediction Method Based on GCN-LSTM Architecture

作　　者：高成 盛武[1] Gao Cheng;Sheng Wu(School of Economics and Management,Anhui University of Science and Technology,Huainan 232001,Anhui)

机构地区：[1]安徽理工大学经济与管理学院,安徽淮南232001

出　　处：《武汉工程职业技术学院学报》2025年第1期29-34,59,共7页Journal of Wuhan Engineering Institute

基　　金：安徽省自然科学基金项目(1808085MG212);安徽理工大学2023年研究生创新基金项目(2023cx2162)。

摘　　要：针对煤矿瓦斯数据预测精度低的问题,提出了一种基于图卷积网络(GCN)和长短期记忆网络(LSTM)集成的时空神经网络模型,以提高瓦斯体积分数预测的准确性。通过GCN和LSTM构建时空块挖掘煤矿多源监测数据时空信息,提升模型识别数据趋势的能力,生成更精确的预测结果。实验结果表明,所提方法在拟合效果上优于GRU和LSTM,在均方根误差(RMSE)、平均绝对误差(MAE)和确定系数(R^(2))上的表现分别为0.2266、0.1875和0.8206,均优于GRU和LSTM,进一步验证了该方法在煤矿瓦斯预测中的有效性。To address the issue of low prediction accuracy in coal mine gas concentration data,a spatiotemporal neural network model combining Graph Convolutional Network(GCN)and Long Short-Term Memory(LSTM)networks is proposed.This model enhances prediction accuracy by employing GCN and LSTM layers to construct spatiotemporal blocks that effectively capture the spatial and temporal features of coal mine gas data,thereby improving the model’s capacity to identify complex data patterns and generate more accurate forecasts.Experimental results demonstrate that the proposed method outperforms GRU and LSTM in terms of fitting accuracy,with Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Coefficient of Determination(R^(2))values of 0.2266,0.1875,and 0.8206,respectively,surpassing those achieved by GRU and LSTM models.These findings confirm the effectiveness of the proposed method in enhancing coal mine gas concentration predictions.

关 键 词：图神经网络 长短期记忆网络 无监督学习 煤矿瓦斯预测 

分 类 号：TD712[矿业工程—矿井通风与安全]