Deep learning prediction and model generalization of ground pressure for deep longwall face with large mining height

被引：0
|
作者：
Zhao Y. [1 ,2 ]
Yang Z. [2 ]
Ma B. [2 ]
Song H. [2 ]
Yang D. [2 ]
机构：
[1] Beijing Key Laboratory for Precise Mining of Intergrown Energy and Resources, China University of Mining & Technology (Beijing), Beijing
[2] School of Energy & Mining Engineering, China University of Mining & Technology (Beijing), Beijing
来源：
Meitan Xuebao/Journal of the China Coal Society | 2020年 / 45卷 / 01期
关键词：
Deep learning; Large mining height; Longwall mining; Mine pressure analysis; Transfer learning;
D O I：
10.13225/j.cnki.jccs.YG19.0903
中图分类号：
学科分类号：
摘要：
The analysis and prediction of the mining pressure in the fully mechanized mining face is of great significance for managing the roof of the working face under complex geology and ensuring the safety of mine production.In this paper, a relational database is used to store the mine pressure data generated by the hydraulic support and the time series characteristics of mine pressure during the process of working face advance.The Long Short Time Memory (LSTM) deep learning method is used to study the mining pressure law of Hongqinghe 31101 fully mechanized mining face.The support working resistance, the unbalanced pressure of support, the state of safety valve and the initial pressure, and the periodic pressure are analyzed.Based on the established database, the mine pressure of the working face is predicted.The prediction results show that the LSTM method has high accuracy compared with Back Propagation (BP) model.In order to discuss the generalization ability of the LSTM network model used in this study, based on a small amount of data obtained, the transfer learning method is used to predict and verify the mine pressure of the Buertai 42103 working face and the 12401 working face of the Shangwan Mine.The verification results show that the method adopted in this study has a good generalization ability.The transfer learning can improve the generalization ability of the model.Finally, in order to improve the accuracy of the model, the large differences in the prediction performance of the model in the three working faces are discussed.It is found that the data itself have a great influence on the behavior of the model.Increasing the amount of data can make up for some of the original data itself.Finally, on the basis of the prediction model, a periodic weighting early warning model is designed, and the corresponding mine pressure analysis and early warning system is integrated.The results are verified by the in-situ engineering data. © 2020, Editorial Office of Journal of China Coal Society. All right reserved.
引用
收藏
页码：54 / 65
页数：11
相关论文
共 25 条
  • [21] Feng X., Wamg Y., Yao J., Adaptive neural network method for real-time prediction of mine roof pressure, Journal of China Coal Society, 20, 5, pp. 455-460, (1995)
  • [22] Hochreiter S., Schmidhuber J., Long short-term memory, Neural Computation, 9, 8, pp. 1735-1780, (1997)
  • [23] Karpathy A., Johnson J., Fei-Fei L., Visualizing and understanding recurrent networks, 17, pp. 1-12, (2015)
  • [24] Manaswi N.K., Understanding and working with keras, Deep Learning with Applications Using Python, 25, pp. 31-43, (2018)
  • [25] Pan S.J., Yang Q., A Survey on transfer learning, IEEE Transactions on Knowledge & Data Engineering, 22, 10, pp. 1345-1352, (2010)
123