摘要
"透明工作面"是实现智能化无人开采的关键,但现阶段的煤层三维地质模型精度较低,无法满足构建高精度煤层地理信息系统的要求。为此,提出了以采煤机历史截割数据和煤层三维地质模型数据根据不同方式划分出2种数据组合,利用长短期记忆(Long-Short Term Memory,LSTM)神经网络挖掘煤层厚度的变化规律并预测煤层厚度分布。基于LSTM神经网络和编码——解码长短期记忆(Encoder-Decoder Long-Short Term Memory,Encoder-Decoder LSTM)神经网络分别建立了煤层厚度预测模型。结果表明:在超参数未优化时,2种模型的煤层厚度预测结果误差均较大;通过优化两种模型的超参数,并以均方根误差(Root Mean Square Error,RMSE)作为煤层厚度预测的评估标准。在第1种数据组合方式下,LSTM模型和Encoder-Decoder LSTM模型的煤厚预测RMSE分别为0.05、0.044 m;在第2种数据组合方式下,2种模型的煤厚预测RMSE分别为0.051、0.049 m。为进一步对比2种模型预测结果,引入绝对误差,求取预测范围内各点的煤厚预测值与真实值的差值。最后得出,2种数据组合方式下,Encoder-Decoder LSTM模型的预测误差在各较小误差范围内的占比始终优于LSTM模型,Encoder-Decoder LSTM预测模型在预测煤层厚度上表现较好,精度较高,其预测的煤层厚度能够修正煤层地质模型。
"Transparent working face"is the key to realize intelligent unmanned mining.However,the current coal seam 3 D geological model has low accuracy and cannot meet the requirements of building a high-precision coal seam geographic information system.To this end,this paper proposes to divide the two data combinations according to different methods based on the historical cutting data of the shearer and the three-dimensional geological model data of the coal seam,and use the Long-Short Term Memory(LSTM)neural network to mine the changes in the thickness of the coal seam.Regularly and predict the distribution of coal seam thickness.Based on the LSTM neural network and the Encoder-Decoder Long-Short Term Memory(Encoder-Decoder LSTM)neural network,a coal seam thickness prediction model was established.The results show that when the hyperparameters are not optimized,the prediction results of the coal seam thickness of the two models are relatively large;the hyperparameters of the two models are optimized,and the root mean square error(RMSE)is used as the evaluation standard of the coal seam thickness prediction.In the first data combination method,the RMSE of coal thickness prediction of the LSTM model and Encoder-Decoder LSTM model are 0.05 m and 0.044 m respectively;in the second data combination method,the coal thickness predictions of the two models are both the RMSE are 0.051 m and 0.049 m respectively.In order to further compare the prediction results of the two models,absolute error is introduced to obtain the difference between the predicted value of coal thickness and the true value of each point in the prediction range.Finally,under the two data combination methods,the Encoder-Decoder LSTM model’s prediction error is always better than the LSTM model in the smaller error range.The Encoder-Decoder LSTM prediction model performs better in predicting the thickness of the coal seam,and the predicted coal seam thickness can revise the coal seam geological model.
作者
梁耍
王世博
谢洋
葛世荣
LIANG Shua;WANG Shibo;XIE Yang;GE Shirong(School of Mechatronic Engineering,China University of Mining and Technology,Xuzhou 221116,China;School of Mechanical and Electrical&Information Engineering,China University of Mining and Technology-Beijing,Beijing 100083,China)
出处
《煤炭科学技术》
CAS
CSCD
北大核心
2021年第S01期150-157,共8页
Coal Science and Technology
基金
国家自然科学基金资助项目(51874279)
关键词
煤层地质模型动态修正
煤层厚度
煤矿智能化无人开采
深度学习
LSTM
dynamic correction of coal geological model
coal thickness
intelligent unmanned mining in coal mines
deep learning
LSTM
