基于LSTM多变种模型的岩爆微震参数预测研究  

Research on the Prediction of Rockburst Microseismic Parameters Based on LSTM Multivariate Model

作　　者：马佳骥 马春驰[1,2] 曾俊 邓叶林 易聪 Ma Jiaji;Ma Chunchi;Zeng Jun;Deng Yelin;Yi Cong(School of Environment and Civil Engineering,Chengdu University of Technology,Chengdu 610059,P.R.China;State Key Laboratory of Geohazard Prevention and Geoenvironmental Protection,Chengdu University of Technology,Chengdu 610059,P.R.China)

机构地区：[1]成都理工大学环境与土木工程学院,成都610059 [2]成都理工大学地质灾害防治与地质环境保护国家重点实验室,成都610059

出　　处：《地下空间与工程学报》2022年第5期1481-1494,共14页Chinese Journal of Underground Space and Engineering

基　　金：军事科学院国防工程研究院协作项目(2019-JKCF-C01-1033)。

摘　　要：岩爆是深地工程和深部资源开采中必须要解决的核心问题之一。基于改进的LSTM神经网络,提出了用于时间序列预测的LSTM微震多参数预测模型,包括单变量时序预测模型和多元平行序列预测模型。并以峨汉高速大峡谷隧道微震监测数据对模型进行验证,同时与多项式回归方法结果进行对比分析。结果表明:单变量预测模型中堆叠式LSTM(S-LSTM)的预测精度最高;多变量预测模型中卷积LSTM(CNN-LSTM)对累积视体积和能量指数具有最好的预测效果,且余下几种LSTM模型仍可准确实现各参数演化趋势的预测,其精度均优于多项式回归分析方法。研究可为正确识别岩爆当前活动及未来状态的危险性提供理论支撑,为及时掌握岩爆未来活动状态提供重要依据。Rockburst is one of the key problems that must be solved in large-scale underground rock and soil mining and deep resource mining projects.Based on the improved LSTM neural network,a multi-parameters LSTM microseismic prediction model for time series prediction is proposed,including univariate sequence prediction model and multivariate parallel sequence prediction model.The model was verified with the microseismic monitoring data of the Grand Canyon Tunnel of E-Han Expressway,and the results of the polynomial regression method were compared and analyzed.The results show that the stacked LSTM(S-LSTM)in the univariate prediction model has the highest prediction accuracy.The convolutional LSTM(CNN-LSTM)in the multivariate prediction model has the best predictive effect on the cumulative apparent volume and energy index.In addition,the remaining several LSTM models can still accurately predict the evolution trend of each parameter,and their accuracies are better than polynomial regression analysis methods.This can provide theoretical support for correctly identifying the current and future state of rockburst,and provide an important basis for grasping the future state of rockburst in time.

关 键 词：隧道工程 岩爆 微震参数 时间序列预测 LSTM 

分 类 号：U456.3[建筑科学—桥梁与隧道工程]