Adaboost集成学习优化的巷道围岩松动圈预测研究  

Prediction Study on Loosening Ring of Surrounding Rock Around Roadways Using the Optimized Ensemble Learning Algorithms Based on Adaboost

作　　者：方博扬 赵国彦[1] 马举[1] 陈立强 简筝 FANG Boyang;ZHAO Guoyan;MA Ju;CHEN Liqiang;JIAN Zheng(School of Resources and Safety Engineering,Central South University,Changsha 410083,Hunan,China)

机构地区：[1]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《黄金科学技术》2023年第3期497-506,共10页Gold Science and Technology

基　　金：‘十三五’国家重点研发计划课题“深部金属矿绿色开采关键技术研发与示范”(编号:2018YFC0604606)资助。

摘　　要：为提高巷道围岩松动圈预测准确率,给围岩支护和地压管理提供更科学的指导,提出了一种新的预测方法。采用改进的Adaboost回归算法对3种机器学习算法进行集成优化,即在Adaboost回归算法中寻找误差率阈值的最优值,实现Adaboost全局最优的集成效果。应用网格搜索对BP、SVM和RF的超参数进行优化,建立BP-Adaboost、SVM-Adaboost和RF-Adaboost回归预测模型。结果表明:BP-Adaboost模型的预测性能最好,误差率为7.65%。结合矿山松动圈测试实例进行验证分析,平均相对误差为4.15%。因此,所提出的模型能够为围岩松动圈预测提供参考,可以满足工程应用的需求。In order to improve the prediction accuracy of loose zone of excavation damaged zone around roadways and provide more scientific guidance for surrounding rock support and ground pressure management,a new prediction method was proposed.The improved Adaboost regression algorithm was used to integrate and optimize three machine learning algorithms,the optimal value of the error rate threshold was found to achieve the global optimal integration of Adaboost.The grid search was used to optimize the hyperparameters of BP,SVM and RF,and the regression prediction models of BP-Adaboost,SVM-Adaboost and RF-Adaboost were established.The results show that the prediction performance of BP-Adaboost is the best,it had the lowest error rate at 7.65 percent.The verification analysis was carried out based on the test example of excavation damaged zone around roadway,the results show that the mean relative error is 4.15%.Therefore,the model proposed in this paper can provide reference for the excavation damaged zone around roadway and meet the needs of engineering applications.

关 键 词：围岩松动圈 网格搜索 ADABOOST算法 BP神经网络 支持向量机 随机森林 

分 类 号：TD322[矿业工程—矿井建设]