摘要
随着煤炭资源深部开采的进行,冲击地压灾害的发生风险不断上升,对矿井的生产效率和人员安全造成严重影响,精准可靠地预测冲击地压灾害尤为重要。应力和声发射信息通常被作为预测冲击地压的前兆信息,能够反应煤岩体破裂过程抗压强度和声发射信号变化趋势。随着近几年人工神经网络技术的飞速发展,利用人工神经网络对冲击地压前兆信息融合预测,为冲击地压预测研究提供了新的方法,可以在矿井生产中更好地保障人员的安全,提高矿井的生产效率。利用长短期记忆神经网络和一维卷积神经网络对冲击地压前兆信息进行数据融合,对比了LSTM、1DCNN与1DCNN-LSTM这3种模型的融合预测结果,结果表明,1DCNN-LSTM模型具有更高的融合准确度,融合抗压强度、峰值频率、振铃计数、能量和持续时间的R^(2)分别为0.9125、0.9392、0.9472、0.9151、0.9565,对冲击地压预测研究具有一定的参考意义。
With the advancement of deep coal resource extraction,the frequency and risk of rock burst disasters have been continuously increasing,posing serious threats to mine productivity and personnel safety.Accurate and reliable prediction of rock burst hazards is therefore crucial.Stress and acoustic emission data are commonly used as precursor information for rock burst prediction,as they reflect changes in the compressive strength and acoustic emission signal trends during the fracturing process of coal and rock masses.With the rapid development of artificial neural network technology in recent years,utilizing neural networks for the fusion and prediction of rock burst precursor information has introduced a new approach to rock burst prediction,enhancing personnel safety and improving mine productivity.Employs Long Short-Term Memory and One-Dimensional Convolutional Neural Network models to fuse rock burst precursor information.The fusion prediction performance of three models-LSTM,1DCNN,and 1DCNN-LSTM is compared.Results indicate that the 1DCNN-LSTM model achieves higher fusion accuracy,with R^(2)values of 0.9125,0.9392,0.9472,0.9151,and 0.9565 for compressive strength,peak frequency,ring count,energy,and duration,respectively,providing valuable insights for rock burst prediction research.
作者
梁燕华
张朝鹏
万陈瑞
吕航
LIANG Yanhua;ZHANG Zhaopeng;WAN Chenrui;LYU Hang(School of Electrical and Control Engineering,Heilongjiang University of Science and Technology,Harbin 150022,China)
出处
《煤炭技术》
2025年第8期141-145,共5页
Coal Technology
基金
国家自然科学基金专项项目(62441306)。
