基于DBO-SVR算法的爆破振动预测比较研究

Research on Blasting Vibration Prediction Based on DBO-SVR Algorithm

为提高预测精度和适应性,基于梅山铁矿的爆破工程,提出了一种基于蜣螂算法优化的支持向量回归(Dung Beetle Optimize Support Vector Regression,DBO-SVR)模型用于爆破时质点峰值振动速度(Peak Particle Velocity,PPV)预测,使用皮尔逊热图分析各特征与PPV的相关性,并使用均方误差和决定系数作为模型评估指标,对比分析DBO-SVR,DBO-XGB,SVR,XGB四个算法的预测结果,四个算法的均方误差分别为0.028,0.152,1.084,0.226,决定系数分别为0.985,0.917,0.408,0.877。研究结果表明,DBO-SVR算法的预测效果优于其他几个模型;DBO-SVR算法模型综合考虑了多个爆破设计参数对PPV的影响,极大缩短样本数据的训练时间,并加快模型的收敛速度以满足实际爆破振动的速度预测要求,预测结果更精确,误差更小,可为类似爆破工程的峰值振动速度的预测提供参考。

In order to improve the prediction accuracy and adaptability,based on the blasting project of Meishan Iron Mine,this paper proposes a Dung Beetle Optimized Support Vector Regression(DBO-SVR)model for PPV prediction.The Pearson heat map is used to analyze the correlation between each feature and PPV(Peak Particle Velocity),and the mean square error and coefficient of determination are used as the model evaluation indexes.The prediction results of DBO-SVR,DBO-XGB,SVR and XGB are compared and analyzed.The mean square errors of the four algorithms were 0.028,0.152,1.084,0.226,and the determination coefficients were 0.985,0.917,0.408,0.877,respectively.The results show that the prediction effect of DBO-SVR algorithm is better than that of other models.The DBO-SVR algorithm model comprehensively considers the influence of multiple blasting design parameters on PPV,greatly shortens the training time of sample data,and accelerates the convergence speed of the model to meet the speed prediction requirements of actual blasting vibration.The prediction results are more accurate and the error is smaller,which can provide reference for the prediction of peak vibration velocity of similar blasting projects.