摘要
为实现爆破振动峰值速度的精准预测并降低爆破振动的危害,基于某工程实测数据,选取爆心距、单段药量、孔距、堵塞长度、孔深等5个关键参数作为影响因子,提出了一种基于粒子群优化算法(PSO)优化的Stacking集成学习模型(PSO-Stacking)。通过PSO优化Stacking基模型中随机森林(RF)的决策树数量与决策树深度、支持向量回归(SVR)的惩罚系数以及梯度提升决策树(GBDT)的树数量等超参数,显著提升了模型的预测性能。结果表明:PSO-Stacking模型的准确率(AR)为85.54%,均方误差(MSE)、均方根误差(RMSE)和决定系数(R^(2))分别为2.39、1.54和0.7361,较PSO-RF、BPNN、AdaBoost等6种模型表现出更优的预测性能与泛化能力。
To accurately predict blasting vibration peak particle velocity and mitigate vibration hazards,this study develops a Stacking ensemble learning model(PSO-Stacking)optimized via a particle swarm optimization(PSO)algorithm,using field data from engineering applications.Five critical parameters were identified as key influencing factors:blast center distance,single-stage drug volume,borehole spacing,stemming length,and hole depth.The model′s predictive performance was substantially enhanced through PSO-based hyperparameter optimization,including the number of depth levels of random forest(RF)decision trees,the support vector regression(SVR)penalty coefficient,and the number of trees in the gradient boosting decision tree(GBDT)within the Stacking framework.Experimental results demonstrate that the PSO-Stacking model achieves 85.54% accuracy(AR),with the mean square error(MSE),root mean square error(RMSE),and coefficient of determination(R^(2))being 2.39,1.54,and 0.7361,respectively,which shows better prediction performance and generalization ability than the six models,such as PSO-RF,BPNN,and AdaBoost.
作者
徐琛
李其乐
邱浪
王超
任高峰
赵亮
刘驰
穆鹏宇
XU Chen;LI Qi-le;QIU Lang;WANG Chao;REN Gao-feng;ZHAO Liang;LIU Chi;MU Peng-yu(Chinese Institute of Coal Science,Beijing 100013,China;School of Resources and Environmental Engineering,Wuhan University of Technology,Wuhan 430070,China)
出处
《爆破》
北大核心
2026年第1期213-224,共12页
Blasting
基金
国家自然科学基金项目(52409143、52304089)
天地科技股份有限公司科技创新创业资金专项资助项目(2023-TD-ZD001-004、2023-TD-QN003)。
关键词
爆破振动峰值速度
粒子群优化
集成学习
blasting vibration peak velocity
particle swarm optimization
ensemble learning
