基于神经网络算法的爆破振动预测模型开发  

Developing a Blasting Vibration Prediction Model Based on Neural Network Algorithm

作　　者：徐国权 王鑫瑀[2] XU Guoquan;WANG Xinyu(School of Earth Science,East China University of Technology,Nanchang 330000,China;Hebei Iron&Steel Group Mining Co.,Ltd.,Tangshan 063000,China)

机构地区：[1]东华理工大学地球科学学院,南昌330000 [2]河北钢铁集团矿业有限公司,河北唐山063000

出　　处：《有色金属工程》2023年第5期94-102,共9页Nonferrous Metals Engineering

基　　金：国家自然科学基金资助项目(52008080)。

摘　　要：爆破地震效应是爆破过程中最为常见的负面效应之一。对地面振动进行准确预测是预防和控制爆破地震效应的前提。研究中,采集了76组爆破振动数据,基于神经网络算法,提出了8种不同的爆破振动预测模型,并考虑了神经元数量、传递函数和学习函数等超参数对神经网络模型预测精度的影响。选择了三种统计指标对模型性能进行评估,并对所开发模型的性能进行了比较。结果显示,建立的ANN-3模型具有最高的预测精度,其相关系数、决定系数和均方根误差分别为0.972、0.945和1.949。得到了神经网络的最佳参数配置:网络结构2-12-1,输入层-隐藏层-输出层之间的传递函数为logsig-pureling,学习函数为learngd。研究成果有助于提高矿山生产爆破的安全,特别是作为爆破设计过程中的辅助工具,便于爆破工程师更好的控制爆破振动,以及为围岩稳定性评估提供决策支持。Ground vibration is common negative effects during blasting.Accurate and effective estimation of blasting vibration is significant in preventing and controlling blasting adverse effects.In this study,eight different artificial neural network(ANN)models are proposed.Several model parameters have been investigated to optimize the ANN model′s performance,such as network structure,transfer function and learn function.A dataset including 76blast events was collected from Sijiaying iron mines in China,Hebei province.The maximum change per delay and the distance between blast points and monitor station were selected as the input variables,the output variable was peak particle velocity.By dividing dataset into training and testing,eight ANN models were constructed with the same train set.Some evaluation criteria were used to assess the prediction capability of the ANN models.The result revealed that all the ANN models can predict ground vibration with satisfaction accuracy.It was found that the third model namely ANN-3can provide the highest accuracy in predicting blast vibration in comparison with other ANN models.The values of 0.972,0.945and 1.949for correlation coefficient(r),coefficient of determination(R2)and root mean squared error(RMSE),respectively.The optimum parameter combination of neurons transfer function and learn function were determined.The parameter configuration of 2-12-1,logsig-purelin and learngd for network structure,transfer function and learn function.This main achievement and finding will increase the blasting safe in mining.Having such a simple and accurate model embedded in blasting design to decrease ground vibration would ease the blast engineer to control the blasting vibration effect and support their decisions to evaluate surrounding rock mass stability.

关 键 词：爆破振动 神经网络 预测 优化 机器学习 

分 类 号：TD235[矿业工程—矿井建设]