基于微震特征参数的秦岭隧洞岩爆实时预测模型

A real-time rockburst prediction model for Qinling Tunnel based on the characteristic parameters of microseismic monitoring

岩爆是深地工程施工中的主要灾害之一,微震监测是岩爆短期预测的主要方法。为了解决岩爆事故短期预测主要依靠经验的问题,基于引汉济渭工程秦岭隧洞,建立微震监测及岩爆事件的机器学习样本库。采用卷积神经网络模型,将一段时间内所有微震事件的能量、位置、震级等特征参数作为输入,并考虑掌子面位置对岩爆的影响,建立基于微震特征参数的岩爆实时预测模型。采用样本库对模型进行训练、验证和测试,针对模型结构、回溯时间段、预测时间段、训练世代对模型进行了优化。模型可以合理描述微震事件分布特征、施工进度等因素对岩爆可能性的影响。经测试,该模型能够实时预测未来48 h的岩爆发生概率,预测准确率超过80%,可为岩爆实时预测提供一种有效的技术途径。

Rockburst is one of the main disasters in the construction of deep earth engineering,and microseismic monitoring is the main method of short-term prediction for rockburst.In order to solve the problem that the short-term prediction of rockburst mainly depends on experience,a database was established consisting of the microseismic monitoring and the rockburst record events at Qinling Tunnel of the water diversion project from Hanjiang River to Weihe River.Based on all characteristic parameters during a period,e.g.,energy,position,and magnitude of the micro seismic event,a real-time prediction model for rockburst was established using the convolutional neural network.Besides,the influence of work surface position on rockburst was also considered.According to the training,validation and test results,the model structure,selection of lookback time,prediction time and training epoch were optimized.The model could reasonably describe the influence of the distribution feature of microseismic and the construction progress on the rockburst probability.Based on the test results,the model can predict the probability of rockburst in the next 48 hours continuously.The accuracy of the prediction is more than 80%,which provides an effective technical way for real-time prediction of rockburst.