基于机器学习方法预测煤矿瓦斯浓度的研究  

Research on Predicting Coal Mine Gas Concentration Based on Machine Learning Methods

作　　者：张晓雷 连瑾 郭继森 胡鹏飞 ZHANG Xiaolei;LIAN Jin;GUO Jisen;HU Pengfei(School of Geology and Mining Engineering,Xinjiang University,Xinjiang 830047;Jiaozuo College of Industry and Trade,Henan 454550;Manas County Tianxin Coal Industry Co.,Ltd,Xinjiang 832217)

机构地区：[1]新疆大学地质与矿业工程学院,新疆830047 [2]焦作工贸职业学院建筑工程系,河南454550 [3]玛纳斯县天欣煤业有限公司,新疆832217

出　　处：《中国煤层气》2024年第5期29-32,共4页China Coalbed Methane

基　　金：国家级创新训练计划一般项目(202310755004)。

摘　　要：采用人工神经元网络(ANN)和长短时记忆网络(LSTM)预测了准东煤田某煤矿工作面的瓦斯浓度,结果表明,在一个隐含层的ANN模型中,神经元数量为32时,预测瓦斯浓度均方误差(MSE)最低为0.5,当隐含层为两个,神经元数量分别为3和10时,预测瓦斯浓度MSE最低为0.03,即最优ANN模型为5-3-10-1。在一个隐含层的LSTM模型中,神经元数量为35时,预测瓦斯浓度MSE最低为0.02,当隐含层为两个,神经元数量分别为10和11时,预测瓦斯浓度MSE最低为0.008,即最优LSTM模型为5-10-11-1。Artificial Neural Network(ANN)and Long Short Term Memory Network(LSTM)are used to predict the gas concentration in a coal mine's working face in Zhundong Coalfield.The results show that in an ANN model with one hidden layer,when the number of neurons is 32,the minimum mean square error(MSE)for predicting gas concentration is 0.5.When there are two hidden layers with 3 and 10 neurons respectively,the minimum MSE for predicting gas concentration is O.03,indicating that the opti-mal ANN model is 5-3-10-1.In an LSTM model with a hidden layer,when the number of neurons is 35,the minimum MSE for predicting gas concentration is 0.02.When there are two hidden layers with 10 and 11 neurons respectively,the minimum MSE for predicting gas concentration is O.008,indicating that the optimal LSTM model is 5-10-11-1.

关 键 词：长短时记忆网络 人工神经元网络 瓦斯 浓度 预测 

分 类 号：TD7[矿业工程—矿井通风与安全]