爆破振动速度峰值（Peak Particle Velocity，PPV）的准确预测是有效控制爆破工程振动危害的前提。为了提高爆破振动速度峰值的预测精度，提出将黏菌算法（Slime Mould Algorithm，SMA）对随机森林（Random Forest，RF）中的树的个数和最小叶子点数2个超参数进行优化。以某露天爆破工程实例中收集的具有4个输入参数（最小抵抗线r、高差H、最大段药量Q _max、水平距离W）和1个输出参数（PPV）的23个样本的数据集为依据，将4种参数组合（Q _max -H-W-r、Q _max -H-r、Q _max -W-r、Q _max -r）作为随机森林算法中的输入参数，确定最优的参数组合。随后对SMA-RF模型、未优化RF模型和国内外常用的6组经验公式的预测结果进行比较，结果表明SMA-RF模型取得了最优的预测效果，因此在工程实践中推荐使用SMA-RF模型预测爆破振动速度峰值。

The vibration caused by blasting is likely to cause instability and failure of facilities such as underground roadways，high and steep slopes in mining areas or ground buildings under dynamic action.Therefore，it is particularly important to predict the intensity of blasting vibration.The accurate prediction of peak particle velocity（PPV） is the premise of effectively controlling the vibration hazard of blasting engineering，but the current empirical formula for predicting the peak particle velocity is not accurate enough.Machine learning has obvious advantages in solving the problem of nonlinear relationship.In order to improve the prediction accuracy of the PPV prediction model，this study proposes to optimize the number of trees and the minimum number of leaf points in the random forest （RF）by slime mould algorithm （SMA） ，which overcomes the inability to obtain the optimal hyperparameters by using a single RF algorithm.Based on a dataset of 23 samples with four input parameters （minimum resistance line-r，height difference-H，maximum segment dose-Q _max，horizontal distance-W） and one output parameter（PPV） collected in an open-pit blasting engineering example，the combination of four parameters of these four parameters （Q _max -H-W-r、Q _max -H-r、Q _max -W-r、Q _max -r） was used as the input parameters in the RF algorithm，and then MAE，RMSE，MEDEA and R ² evaluate the prediction effect of the SMA-RF model for four different input parameters to determine the optimal combination of parameters.In this model，the fitness function in SMA is defined as the root mean square error of the predicted value to enhance the robustness of the RF model.Then，the performance of SMA-RF model and unoptimized RF model and six empirical formulas commonly used in China and abroad were compared.The results show that the SMA-RF model has better prediction accuracy than the RF model，and the SMA-RF model has significantly better prediction effect than the six empirical formulas.In addition，Q _max -H-W-r can train the optimal SMA-RF model in the combination of four parameters，so it is recommended to be used to predict PPV in engineering practice.

http://www.goldsci.ac.cn/article/2023/1005-2518/1005-2518-2023-31-4-624.shtml