Failure precursor of surrounding rock mass around cross tunnel in high-steep rock slope

The stability of the surrounding rock mass around cross tunnel in the right bank slope of Dagangshan hydropower station, in the southwestern China, was analyzed by microseismic monitoring as well as numerical simulations. The realistic failure process analysis code (abbreviated as RFPA3D ) was employed to reproduce the initiation, propagation, coalescence and interactions of micro-fractures, the evolution of associated stress fields and acoustic emission (AE) activities during the whole failure processes of the surrounding rock mass around cross tunnel. Combined with microseismic activities by microseismic monitoring on the right bank slope, the spatial-temporal evolution and the micro-fracture precursor characteristics during the complete process of progressive failure of the surrounding rock mass around cross tunnel were discussed and the energy release law of the surrounding rock mass around the cross tunnel was obtained. The result shows that the precursor characteristic of microfractures occurring in rock mass is an effective approach to early warn catastrophic damage of rock mass around cross tunnel. Moreover, the heterogeneity of rock mass is the source and internal cause of the failure precursor of rock mass.

基于秦岭输水隧洞微震参数的机器学习岩爆预测模型

随着地下工程不断向深部发展,岩爆的频次和强度日益增加。为了实现对岩爆烈度等级的预测,本文引入6种机器学习算法,建立了6个岩爆预测模型。以秦岭输水隧洞工程289天的微震监测数据与岩爆案例为基础,构建岩爆烈度等级预测数据集。在建模过程中,首先讨论了数据不均衡对模型性能的影响,得出Borderline-SMOTE1是消除数据不均衡最有效的方法。其次,对6个模型性能指标进行分析,发现Adaboost算法的岩爆预测模型性能最好,其精度、宏F1值、微F1值均最高,分别为0.938、0.937和0.938。最后将Borderline-SMOTE1-Adaboost岩爆预测模型应用于秦岭输水隧洞工程2020年6月1日至2020年6月10日的岩爆烈度等级预测,10次强烈岩爆均被准确预测,验证了通过微震参数来预测岩爆烈度等级的有效性,表明了Borderline-SMOTE1-Adaboost岩爆预测模型可为深埋隧道施工过程中岩爆灾害的预警提供参考。

基于改进LeNet5卷积神经网络的微震监测波形识别与过程解释

在微震大数据时代背景下,开发高精度、可解释、适应性强的波形自动分类算法变得越来越重要。针对现有网络波形识别和分类的不足,基于LeNet框架提出了一种适用于微震监测波形识别的改进模型。应用改进后的模型对引汉济渭工程8个月内出现的13种微震监测信号进行了研究。结果表明,改进模型中最佳框架的精度为0.98,比原模型提高了0.10。所有改进模型的平均精确度、召回率和F1值分别提高了0.11、0.12和0.12。同时,改进后的模型可以对整个波形的识别过程可视化。在某些信号类别中,改进的模型主要通过关注背景信息而不是波形来分类,为微震监测工程中信号的智能分类提供了参考。