基于RNN算法的煤矿井下瓦斯浓度预测研究

Study on prediction of gas concentration in coal mine underground based on RNN algorithm

煤矿瓦斯事故往往是由瓦斯浓度过高引起的,为使瓦斯浓度保持在安全范围内,利用Lasso回归算法实现瓦斯浓度时间序列的特征选择,并以瓦斯浓度特征集合为对象,建立了基于递归神经网络(RNN)的瓦斯浓度预测模型。以平均绝对百分比误差(MAPE)为性能指标,对RNN算法与SVR和BP神经网络算法模型进行对比分析,结果表明:RNN算法不仅提高了预测精度,而且将相对误差限制在最小范围内,具有更高的稳定性,MAPE可降低到0.305%,预测某矿1206工作面9月28日9:30瓦斯浓度为0.8019%,建议工作人员实时关注该区域瓦斯浓度变化情况并采用必要的防治措施,能够为矿井瓦斯浓度预测提供理论指导。

Coal mine gas accidents are often caused by excessive gas concentration.In order to keep the gas concentration within a safe range,the Lasso regression algorithm was used to realize the feature selection of the gas concentration time series,and taking the gas concentration feature set as the object,a gas concentration prediction model based on recurrent neural network(RNN)was established.Using the average absolute percentage error(MAPE)as the performance index,the RNN algorithm was compared with the SVR and BP neural network algorithm models.The results show that the RNN algorithm not only improves the prediction accuracy,but also limits the relative error to a minimum range.The stability of the MAPE can be reduced to 0.3050%.It is predicted that the gas concentration of 1206 working face of a mine is 0.8019%at half past nine on September 28.It is recommended that the staff pay attention to the changes in the gas concentration in the area in real time and take necessary preventive measures.It can provide theoretical guidance for mine gas concentration prediction.