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

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.