基于CNN-GRU的瓦斯浓度预测模型及应用*

为解决传统瓦斯浓度预测方法预测精度低和适用性不强等问题，提出运用卷积神经网络（CNN）提取瓦斯浓度时间序列的变化趋势及局部关联特征，应用门自适应矩估计（Adam）优化的控循环单元神经网络（GRU），在关联特征基础上进行时序性预测的组合方法，并以铜川玉华煤矿监测数据为样本，对比CNN-GRU组合模型、传统机器学习模型LSTM和GRU模型的预测效果。研究结果表明:CNN-GRU模型的预测精度和收敛速度均优于LSTM和GRU模型；CNN-GRU平均绝对误差和均方根误差分别可降低至0.042，0.006，运行效率分别提高59.15%，35.04%，研究结果可为矿井瓦斯灾害防治提供依据。

Prediction model of gas concentration based on CNN-GRU and its application

Aiming at the problems of low prediction accuracy and weak applicability of the traditional methods for gas concentration prediction,the convolution neural network (CNN) was applied to extract the change trend and local correlation characteristics of gas concentration time series,and a combination method of time series prediction based on the correlation characteristics was put forward by using the controlled loop unit neural network (GRU) optimized by the gate adaptive moment estimation (Adam).Taking the monitoring data of Yuhua coal mine in Tongchuan as the sample,the prediction effect of CNN-GRU combined model was compared with those of the traditional machine learning LSTM model and GRU model.The results showed that the CNN-GRU model was better than LSTM model and GRU model in the prediction accuracy and convergence speed.The average absolute error and root mean square error of CNN-GRU could be reduced to 0.042 and 0.006 respectively,and the operation efficiency increased by 59.15% and 35.04% respectively,with higher application value.The results can provide basis for the gas disaster prevention and control in mines.