基于GD-DNN模型的岩爆烈度等级预测方法与应用

岩爆是深埋隧道施工过程中开挖时形成临空面,引起能量突然释放的现象,轻则岩片剥落,重则造成人员伤亡和财产损失,其危害程度取决于岩爆烈度等级,因此岩爆烈度等级预测是急需解决的难题之一。本文针对单一指标预测法预测效果不理想的问题,首先设计并实现了综合指标法和针对多分类问题的分类器,其次提出并建立了基于梯度下降(Gradient Descent)算法优化深度神经网络(Deep Neural Network)的GD-DNN岩爆烈度等级预测模型。实验结果表明：GD-DNN模型预测的准确率达到95.8%,相比机器学习算法K最近邻(K-nearest neighbor,KNN)、支持向量机（support vector machine,SVM)和深度学习算法DNN分别提高了45.8%、38.7%和8.3%,同时在精确率、召回率和??1值三项指标上均优于其他模型。最后在秦岭隧道、大相岭隧道、通渝隧道和马路坪矿井4个实际工程中检验模型的预测效果,检验结果证明GD-DNN模型能够精准预测岩爆烈度等级,研究成果可应用于深埋隧道工程中。

Prediction method and application of rockburst intensity grade based on GD-DNN model

Rockburst is a phenomenon in which the air surface is formed during the excavation of a deeply buried tunnel, causing a sudden release of energy, ranging from flaking of rock flakes to causing casualties and property losses in severe cases. Therefore, the prediction of rockburst intensity grade is one problem that needs to be solved urgently. This paper aims at the unsatisfactory forecasting effect of the single index forecasting method. Firstly, a comprehensive index method and a classifier for multi-classification problems are designed and implemented. Secondly, a GD-DNN rockburst intensity grade prediction model based on a gradient descent algorithm to optimize deep neural network is proposed and established. The experimental results show that the prediction accuracy of the GD-DNN model reaches 95.8%, which is 45.8%, 38.7%, and 8.3% higher than the machine learning algorithm K-nearest neighbor (KNN), support vector machine (SVM), and deep learning algorithm DNN, respectively. At the same time, it outperforms other models in the three indicators of precision, recall, and ??1 value. Finally, the prediction effect of the model was tested in four actual projects Qinling Tunnel, Daxiangling Tunnel, Tongyu Tunnel, and Maluping Mine. The test results prove that the GD-DNN model can accurately predict the rockburst intensity grade, and the research results can be applied to deep tunnel engineering.