EMD-GRU组合模型在煤矿瓦斯体积分数预测中的应用  

Application of the EMD GRU combined model for predicting gas volume fraction in coal mines

作　　者：盛武[1,2] 樊斌斌 SHENG Wu;FAN Binbin(College of Economic and Management,Anhui University of Science and Technology,Huainan 232001,Anhui,China;State Key Laboratory of Mining-Induced Response and Disaster Prevention and Control in Deep Coal Mines,Anhui University of Science and Technology,Huainan 232001,Anhui,China)

机构地区：[1]安徽理工大学经济与管理学院,安徽淮南232001 [2]安徽理工大学深部煤矿采动响应与灾害防控国家重点实验室,安徽淮南232001

出　　处：《安全与环境学报》2025年第4期1339-1348,共10页Journal of Safety and Environment

基　　金：国家自然科学基金项目(71971003);安徽省自然科学基金项目(1808085MG212)。

摘　　要：为更好厘清瓦斯体积分数数据中长期依赖关系,实现更加精确的瓦斯体积分数预测,提出一种基于经验模态分解算法—门控循环单元(Empirical Mode Decomposition-Gated Recurrent Unit,EMD-GRU)组合的瓦斯体积分数预测方法。该方法通过经验模态分解算法(Empirical Mode Decomposition,EMD)将复杂瓦斯体积分数数据分解为多个平稳序列,并对每个序列采用双层门控循环单元(Gated Recurrent Uni,GRU)进行时序预测;然后,将各分量预测结果叠加重构,得到最终的预测结果;最后,将EMD-GRU模型应用于某矿智能综采工作面瓦斯气体体积分数预测。结果表明:EMD-GRU模型较经验模态分解-长短期记忆(Empirical Mode Decomposition-Long Short-Term Memory,EMD-LSTM)神经网络模型、传统长短期记忆(Long Short-Term Memory,LSTM)神经网络模型和GRU模型,其均方误差(ERMS)分别减少11.57%、33.86%、48.78%,平均绝对百分比误差(EMAP)分别降低19.55%、28.23%和32.76%,同时,在测试集中有着更高的拟合效果(R2=0.9789),验证了该模型有较高的精准度和泛化性。A prediction model for gas volume fraction in coal mines,based on Empirical Mode Decomposition(EMD)and Gated Recurrent Unit(GRU),is proposed.This model is designed to effectively capture long-term dependencies in gas volume fraction data,resulting in more accurate predictions and enhancing the prevention and control of gas-related accidents.The model begins by decomposing the input gas volume fraction time series data using the EMD algorithm,resulting in multiple intrinsic mode functions and a residual term.This decomposition allows for the exploration of various periodic features within the data.Subsequently,each subsequence is forecasted using a two-layer Gated Recurrent Unit(GRU)network for time-series prediction.The predicted results of each component are then summed and reconstructed to obtain the final gas volume fraction prediction.Finally,the model s performance is validated using gas volume fraction monitoring data from an intelligent coal mining face.The results are compared with those of traditional neural network models,including the LSTM model,GRU model,and the combined Long Short-Term Memory and Empirical Mode Decomposition(EMD LSTM)model.The results indicate that the Root Mean Square Error(RMSE)of this model is 0.0084,which is 11.57%lower than that of the EMD LSTM model,33.86%lower than the LSTM model,and 48.78%lower than the GRU model.Additionally,the Mean Absolute Percentage Error(MAPE)is 9.71%,representing a reduction of 19.55%compared to the EMD LSTM model,28.23%compared to the LSTM model,and 32.76%compared to the GRU model.The R 2 value for this model is 0.9789,which is 0.0062 higher than that of the EMD LSTM model,0.0275 higher than the LSTM model,and 0.0598 higher than the GRU model.Additionally,the relative errors for this model in predicting abnormal data points range from 0.30%to 12.01%,surpassing the performance of the EMD LSTM,LSTM,and GRU models.The validation results demonstrate that the EMD GRU model effectively captures the various periodic trends in gas volume fraction data and exh

关 键 词：安全工程 经验模态分解 门控循环单元 井下监测数据 瓦斯体积分数预测 

分 类 号：X936[环境科学与工程—安全科学]