摘要
为了解决岩爆数据库中存在数据不均衡的问题,导致岩爆预测准确率较低等问题,基于SMOTE(synthetic minority oversampling technique)过采样算法提出了SMOTE-随机森林、SMOTE-梯度提升决策树、SMOTE-支持向量机、SMOTE-BP神经网络、SMOTE-卷积神经网络5种模型。选取6个指标,并将岩爆烈度等级划分为4个等级,以此建立岩爆指标体系。然后,针对岩爆数据库存在数据不均衡问题,使用SMOTE过采样算法扩增数据库。最后引入5种常用的机器学习模型预测岩爆烈度等级,并将这5种模型分别对原始的岩爆数据库和经过SMOTE算法后的岩爆数据库进行预测,验证预处理过程的有效性。结果表明:1)相比于传统模型,引入SMOTE算法后,模型预测准确率提高了10.000%~35.000%;2)SMOTE-随机森林模型相比于其他4种模型预测准确率最高。
In order to solve the problem of data imbalance in the rockburst database,resulting in low prediction accuracy of rockburst,five models were proposed based on the synthetic minority oversampling technique(SMOTE),including SMOTE-random forest,SMOTE-gradient boosting decision tree,SMOTE-support vector machine,SMOTE-BP neural network and SMOTE-convolutional neural network.In this paper,six indicators were selected and the rockburst intensity grade was divided into four grades,so as to establish a rockburst index system.Then,in view of the problem of data imbalance in the rockburst database,the SMOTE oversampling algorithm was used to expand the database.Finally,five commonly used machine learning models were introduced to predict the rockburst intensity level,and these five models were used to predict the original rockburst database and the rockburst database after SMOTE algorithm respectively,to verify the effectiveness of the pretreatment process.The results show that:1)Compared with the traditional model,the prediction accuracy of the model is improved by 10.000%~35.000%after the introduction of SMOTE algorithm;2)Compared with the other four models,the SMOTE-random forest model had the highest prediction accuracy.
作者
李璐佳
周爱红
袁颖
戎密仁
LI Lujia;ZHOU Aihong;YUAN Ying;RONG Miren(Hebei GEO University,Shijiazhuang 050031,China;Hebei Technology Innovation Center for Intelligent Development and Control of Underground Built Environment,Shijiazhuang 050031,China)
出处
《河北地质大学学报》
2025年第3期30-37,共8页
Journal of Hebei Geo University
基金
中央引导地方科技发展资金项目(246Z5405G)。
