基于ACO-BP模型的岩石爆破破碎块度预测  

Prediction of Blast-Induced Rock Fragmentation Based on ACO-BP Model

作　　者：陈莎莎 何理[1,2,3] 李腾飞 张鑫玥 彭胜 姚颖康[3] 刘昌邦 陈江伟[6] CHEN Shasha;HE Li;LI Tengfei;ZHANG Xinyue;PENG Sheng;YAO Yinkang;LIU Changbang;CHEN Jiangwei(Hubei Province Key Laboratory of Systems Science in Metallurgical Process,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Hubei Provincial Intelligent Blasting Engineering Technology Center,Wuhan 430065,Hubei,China;Hubei Provincial Key Laboratory of Blasting Engineering,Jianghan University,Wuhan 430056,Hubei,China;School of Urban Construction,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Wuhan Explosion&Blasting Co.,Ltd.,Wuhan 430056,Hubei,China;China Construction Seventh Engineering Division Co.,Ltd.,Zhengzhou 450004,Henan,China)

机构地区：[1]武汉科技大学冶金工业过程系统科学湖北省重点实验室,湖北武汉430065 [2]湖北省智能爆破工程技术研究中心,湖北武汉430065 [3]江汉大学爆破工程湖北重点实验室,湖北武汉430056 [4]武汉科技大学城市与建设学院,湖北武汉430065 [5]武汉爆破有限公司,湖北武汉430056 [6]中国建筑第七工程局有限公司,河南郑州450004

出　　处：《矿冶工程》2024年第5期12-16,21,共6页Mining and Metallurgical Engineering

基　　金：国家自然科学基金(52274136,51904210);爆破工程湖北省重点实验室基金项目(BL2021-11);湖北省重点研发计划(2020BCA084)。

摘　　要：为了对岩石爆破破碎块度进行有效预测,设计开展了混凝土试件钻孔爆破试验,得到不同试验条件下的破碎块度归一化值分布,最终选取试块尺寸40 mm以上进行研究。采用Spearman相关系数分析各试验条件参数之间的相关性,再采用蚁群算法(ACO)优化BP神经网络的初始权值和阈值,构建ACO-BP模型。结合现场试块爆破破碎块度数据对模型进行了训练和测试,并将预测模型与BP神经网络模型、随机森林(RF)模型、极限梯度提升(XGboost)模型进行了对比。结果表明,ACO-BP模型预测爆破块度均方根误差为0.13,平均绝对误差为0.11,决定系数为0.92,预测精度和适用性更高,能够更准确地预测岩石爆破破碎块度。In order to effectively predict blast-induced rock fragmentation,a distribution of normalized rock fragmentation under different conditions was obtained by performing a designed experiment on drilling and blasting of a concrete specimen,and then the rock fragmentation exceeding 40 mm was selected for study.The correlation among variables under different testing conditions was analyzed by using Spearman correlation statistics,and the initial weights and thresholds of the BP neural network were optimized by using the ant colony optimization(ACO)to construct an ACO-BP model.The model was then trained with rock fragmentation by on-site blasting,and tested.Based on the comparison of such prediction mode with BP neural network model,random forest(RF)model and extreme gradient boosting(XGboost)model,it is found that the ACO-BP model is highly reliable in predicting blast-induced rock fragmentation,presenting a root mean square error of 0.13,an average absolute error of 0.11,and a coefficient of determination of 0.92.It is concluded that this model,with higher accuracy in prediction and applicability,can accurately predict blast⁃induced rock fragmentation.

关 键 词：岩石爆破 破碎块度 模型试验 块度预测 ACO-BP模型 

分 类 号：TD235[矿业工程—矿井建设]