深部过应力岩体破坏模式智能预测模型
Intelligent hybrid model to classify failure modes of overstressed rock masses in deep engineering
刘子达 1李地元1
作者信息
- 1. School of Resources and Safety Engineering,Central South University,Changsha 410083,China
- 折叠
摘要
挤压大变形、板裂和应变岩爆是深部岩体开挖工程中过应力岩体的典型破坏形式.本研究采用完整岩石性质参数来预测挤压大变形、板裂和应变岩爆.基于收集的数据库,提出了Jaya算法与支持向量机结合的智能模型(JA-SVM)评估过应力岩体破坏模式.JA-SVM模型的训练准确率为0.970,测试准确率为0.875;排序系统和泰勒图显示提出的JA-SVM模型优于SVM、人工神经网络等机器学习模型;ROC曲线显示JA-SVM与其他广泛应用的机器学习模型相比,具有更好的预测应变岩爆和板裂能力;敏感性分析表明脆性指数和杨氏模量是评估过应力岩体破坏模式的重要变量.本文提出的智能模型可用于深部地下工程初期过应力岩体破坏模式识别,有利于施工现场根据评估结果提前采取相应的支护措施.
Abstract
Squeezing, slabbing, and strainburst are typical failure modes of overstressed rock masses in deep rock excavation engineering. This study considered intact rock properties to evaluate squeezing, slabbing, and strainburst, owing to the effectiveness and availability of these parameters. Hybrid models combining the Jaya algorithm and support vector machine (JA-SVM) were proposed to predict the failure modes of overstressed rock masses based on the collected database. JA-SVM model achieved a training accuracy of 0.970 and a testing accuracy of 0.875. Ranking system and Taylor diagrams showed that the developed hybrid model was superior to other machine learning (ML) models, including SVM, artificial neural network, etc. Receiver operator characteristic curves suggested that JA-SVM had a more powerful ability to predict strainburst and slabbing compared to other widely applied ML techniques. Performed sensitive analysis revealed that the brittleness index and elastic modulus were vital factors in estimating failure modes. The developed model can be applied to identify failure modes of overstressed rock masses in the initial phases of a deep underground project, and appropriate support measures can be prepared beforehand based on estimation results.
关键词
挤压大变形/板裂/应变岩爆/支持向量机/Jaya算法Key words
squeezing/slabbing/strainburst/support vector machine/Jaya algorithm引用本文复制引用
基金项目
National Natural Science Foundation of China(52074349)
出版年
2023
NETL
NSTL
维普
万方数据