基于深度学习的大采高工作面矿压预测分析及模型泛化  

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

作　　者：赵毅鑫 杨志良[2] 马斌杰 宋红华 杨东辉[2] ZHAO Yixin;YANG Zhiliang;MA Binjie;SONG Honghua;YANG Donghui(Beijing Key Laboratory for Precise Mining of Intergrown Energy and Resources,China University of Mining&Technology(Beijing),Beijing 100083,China;School of Energy&Mining Engineering,China University of Mining&Technology(Beijing),Beijing 100083,China)

机构地区：[1]中国矿业大学(北京)共伴生能源精准开采北京市重点实验室,北京100083 [2]中国矿业大学(北京)能源与矿业学院,北京100083

出　　处：《煤炭学报》2020年第1期54-65,共12页Journal of China Coal Society

基　　金：国家自然科学基金资助项目(51874312,U1910206);国家重点研发计划资助项目(2016YFC0801401)

摘　　要：综采工作面矿压显现的分析与预测对于复杂地质条件下工作面顶板管理,保证矿井生产安全具有重要意义。采用关系型数据库储存液压支架工作阻力数据以及利用工作面推进过程中矿压显现的时序特性,采用SQL语言,运用长短时记忆网络(Long Short Time Memory,LSTM)深度学习方法,以红庆河矿31101大采高综采工作面矿压规律为研究对象,对工作面支架工作阻力、支架不平衡力、支架安全阀开启情况及初次来压与周期来压等矿压显现规律进行分析;基于建立的数据库,预测了红庆河大采高工作面矿山压力,预测结果表明LSTM方法较BP神经网络预测更具准确性。为进一步讨论本研究采用的LSTM网络模型的泛化能力,在采用布尔台42103大采高工作面、上湾矿12401大采高工作面少量矿压数据的前提下,使用迁移学习方法,对矿压数据进行预测检验,结果表明:LSTM模型具有很好的泛化能力,相比于不使用迁移学习方法,迁移学习可提高模型的泛化能力。最后,探讨了模型在3个大采高工作面矿压预测表现的差别,发现数据量本身对模型预测行为影响较大,增大数据量可弥补原始数据缺失等问题。在预测模型基础上设计了周期来压预警模型,集成形成相应矿压分析与预警系统,经工程验证判定预警系统分析效果良好。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.

关 键 词：大采高 长壁开采 深度学习 矿压分析 迁移学习 

分 类 号：TD323[矿业工程—矿井建设]