基于HGS-ANN混合模型的爆破振动预测

Blasting Vibration Prediction Based on Novel HGS-ANN Model

将饥饿游戏搜索算法(HGS)与神经网络算法(ANN)相结合,开发了一种新的混合模型HGS-ANN,用来预测爆破振动。分别基于数据分组处理方法(GMDH)、支持向量机(SVM)、神经网络算法(ANN)以及萨道夫斯基经验公式建立了4种不同预测模型,并与HGS-ANN模型进行对比,评估模型性能。从某露天矿山收集了32组爆破数据,选择爆心距、最大单段药量、总药量、抵抗线、孔距、孔数、孔深等7个自变量作为输入参数,选择质点振动速度作为输出参数,以均方根误差(RMSE)和决定性系数(R^(2))作为模型性能评价指标,对所建立的模型性能进行对比。结果表明,HGS-ANN模型的RMSE和R^(2)分别为0.833和0.963,性能优于其他4种模型。HGS-ANN模型可以作为一个辅助工具来优化爆破设计,降低爆破地震效应。

Based on the combination of the hunger games search(HGS)algorithm and the artificial neural network(ANN),a new hybrid model of HGS-ANN was developed to predict blasting vibration.Four different prediction models were established based on group method of data handling(GMDH),support vector machines(SVM),ANN and Sadov′s empirical formula,and compared with HGS-ANN model in evaluating the performance of models.For this purpose,32 sets of blasting data of an open-pit mine were collected.7 independent variables,including detonation distance,maximum single-stage charge,total charge,burden,hole spacing,number of holes and hole depth were selected as inputs,while the particle vibration velocity was selected as the output.With the root-mean-square error(RMSE)and the decisive factor(R^(2))as the evaluating indicators,the established models was compared in terms of their performances.The results show that the HGS-ANN model,with the RMSE and R^(2) of 0.833 and 0.963,respectively,has performance better than the other four models.It is proposed that the HGS-ANN model can be used as an auxiliary tool to optimize the blasting design for reducing the blasting-induced seismic effect.