基于堆叠式LSTM的采煤工作面瓦斯浓度预测方法

Coal mining face gas concentration prediction method based on stacked LSTM

随着淮南矿区煤炭开采工作向深部煤层开展,煤与瓦斯突出问题成为制约淮南矿区产能的重要因素,为提高瓦斯浓度预测方法的准确度,在长短期时间记忆网络(LSTM)的基础上,模仿深度神经网络通过多层堆叠提升特征能力的方式,提出堆叠式LSTM模型结构,通过多层LSTM结构深度挖掘瓦斯浓度时间序列数据中存在的关联性,从而提高瓦斯浓度预测的准确度。以淮南朱集东煤矿瓦斯监测数据为样本,通过滑动窗口方法制作数据集,并搭建相应实验平台,进行训练与验证。实验结果表明,采用堆叠式LSTM结构相较于传统LSTM结构能够降低网络结构的平均绝对百分误差,并更加准确地预测煤矿瓦斯浓度。

With the development of coal mining in the Huainan mining area towards deep coal seams,the issue of coal and gas outburst has become an important factor restricting the production capacity of the Huainan mining area.In order to improve the accuracy of gas concentration prediction methods,a stacked LSTM model structure is proposed based on the Long Short Term Time Memory Network(LSTM),imitating the deep neural network to enhance feature capabilities through multi-layer stacking.By deeply probing into the correlations in gas concentration time series data through multi-layer LSTM structure,the accuracy of gas concentration prediction can be improved.Using gas monitoring data from Zhujidong Coal Mine in Huainan as a sample,a dataset was created using the sliding window method,and a corresponding experimental platform was built for training and verification.The experimental results show that using stacked LSTM structure can reduce the average absolute percentage error of network structure compared to traditional LSTM,and more accurately predict coal mine gas concentration.